# Changelog

All notable changes to the punkshell project version are documented here.
The latest `## [X.Y.Z]` header must match the `version` field in `punkproject.toml`.
Entries are newest-first; one bullet per notable change. See the root `AGENTS.md`
"Project Versioning" section for the bump policy.

## [0.50.0] - 2026-08-04

- punk executable -e one-liner support (G-077, achieved): `<punkexe> -e
  <script> ?args...?` and `<punkexe> script -e <script> ?args...?` run Tcl
  code as a one-liner in the punk script environment - the perl/python/node
  reflex now works on punk kits instead of erroring. `::argv0` is `-e`,
  trailing args land in `::argv`, a non-empty final result is echoed (as
  for the stdin form), errors print errorInfo to stderr with exit 1, and a
  `-e` with no following script argument is a usage error - never an
  interactive fall-through and never stock tclsh's -e misparse (argv-swallow
  + stdin read). The top-level form is a boot-core 0.4.0 reclassification
  ahead of the G-032 unknown-first-arg refusal (script-default kits only;
  tool-style kits keep `-e` routed to their processor, and `<punkexe> tclsh
  -e ...` keeps deliberate stock parity per G-118). app-punkscript 1.1;
  moduledoc punkexe 0.5.0 documents both forms; the unknown-first refusal
  hint and the degraded plain help name them. Characterization:
  scriptexec.test (-e section, 10 pins) + launcherhelp.test - 49/49 green
  on the rebaked punk902z/punksys/punk86.

## [0.49.8] - 2026-08-04

- commandstack 0.7.0 (G-160 follow-on, landed at user direction): the
  remove_rename convenience forms parked as code todos since 2021 -
  pop_rename (pop a renamer's topmost stack entry, per command or
  searched across live stacks), remove_renamer (remove ALL of a
  renamer's entries across every live stack - the unload-my-package
  form), and restore_original (unwind a command's whole stack to the
  original implementation regardless of renamer; deliberately not gated
  by known_renamers so it survives state loss such as a module
  re-source). All three route through remove_rename's re-linking/token
  map machinery; Rename_stack-parked stacks are invisible to the
  renamer-wide forms. commandstack suite 40/40 on tclsh90 (9.0.3) and
  punk86 (8.6); packagepreference consumer suites 6/6 on both. Minted
  as modules/commandstack-0.7.0.tm; bootsupport/vfscommon promotion
  left to the next promotion cycle (as with 0.6.0).

## [0.49.7] - 2026-08-04

- punk::tcltestrun 0.4.1 (G-161 follow-on, landed at user direction):
  errorInfo/errorCode values in failure reports no longer carry the
  historic leading space (the 16-character "---- errorInfo: " /
  "---- errorCode: " prefixes were captured from index 15 - the prefix's
  trailing space - instead of 16). Report-shape normalization only; the
  parser suite now pins the normalized shape exactly.

## [0.49.6] - 2026-08-04

- punk::tcltestrun 0.4.0 (G-161): `parse_testrun` now parses failing-test
  opening banners that span physical lines (test descriptions containing
  embedded newlines) with full fidelity - failure entry with name,
  description, actual/expected, and the summary line plus every later
  event intact - by anchoring banner recognition on tcltest's
  `---- <name> start` events with bounded continuation buffering.
  Previously such a failure wedged the per-file parser: every later event
  including the summary line was swallowed and the file degraded to
  warn/missing-cleanupTests with no failure detail. Also fixes
  result_expected capture: the "Result should have been" opener was
  consumed by the active result_was capture's reset branch, so FAILED
  reports never carried the expected value (and it leaked into the
  captured test body). runtests.tcl floors its require at punk::tcltestrun
  0.4.0-; the test-description single-line rule is relaxed to style
  guidance (src/tests/AGENTS.md); parser characterization suites live at
  src/tests/runner/testsuites/parser/.

## [0.49.5] - 2026-08-03

- punk::console 0.8.1: fixed `get_size` returning `columns {} rows <n>` on
  unix (WSL and native) on every call after the first. `get_size_using_tput`
  invoked tput once with two capnames, but multi-capname support varies by
  implementation (ncurses >= 6.1 and BSD emit one value per capname; older
  ncurses ignores the second and emits one value) - and the malformed result
  won the mechanism-timing cache without erroring. tput now autodetects
  multi-capname support by result shape (cached per tput path); new
  `get_size_using_sttysize` mechanism (`stty size` - single fork, fixed
  output shape) preferred ahead of tput; `size_via_query_mechanisms` now
  validates mechanism results (`size_result_valid`) so malformed dicts -
  including ANSI-mechanism timeout results - are treated as failures instead
  of winning silently.

## [0.49.4] - 2026-08-03

- shellfilter 0.2.5: fixed unix repl result-echo corruption - large multibyte-dense
  results (e.g `i <cmd>` help tables) displayed with contiguous missing ranges,
  fused mid-word seam lines or truncated bottom rows on unix ttys (WSL and native).
  Root cause: `shellfilter::chan::tee_to_pipe` declared the transchan `clear` op -
  the Tcl core delivers `clear` before every write when declared (the G-145 defect
  class) - and its `clear` discarded the held partial multi-byte character, leaving
  the next chunk starting with orphan continuation bytes; whole chunks then
  accumulated undecodable and successive clears discarded them. `clear` is no longer
  declared (method kept as state-preserving no-op). Also repaired `flush` in
  tee_to_pipe/tee_to_log/logonly (referenced an undeclared variable and threw on
  every invocation). Windows rendering paths unchanged.

## [0.49.3] - 2026-08-03

- commandstack 0.6.0 (G-160 hygiene pass): tokenids are unique and monotonic
  per (renamer, command) - same-renamer re-renames now dispatch through the
  full override chain and are individually removable by exact token (the
  counter was stuck at 1 on an apply-local variable, silently bypassing
  intermediate overrides and erroring on a third rename); get_next_command
  dispatch is a single dict lookup (token->implementation map, no stack scan);
  get_IMPLEMENTOR classifies builtins where tcl::info::cmdtype exists;
  informational warnings are debug-gated (a stray write of the literal word
  "stderr" to stdout is gone); rename records carry a trailing did_rename 0|1;
  misplaced -renamer flags error instead of silently misparsing; new
  commandstack::next delegation helper; Rename_stack-parked stacks are
  retrievable and keep dispatching; Delete_stack refuses to strand live
  overrides; known_renamers defaults match the vendored packagetrace/
  packagesuppress registrations.
- punk::packagepreference 0.2.1: a second install returns 0 silently as
  documented instead of throwing (the already-installed guard read a
  nonexistent record key; latent in kits, which install once per interp).

## [0.49.2] - 2026-08-03

- commandstack 0.5.0: `commandstack::help` now returns a real API overview
  (was an empty string), and every API proc carries a PUNKARGS documentation
  block registered lazily with punk::args (no punk::args dependency added -
  the module remains dependency-free). First test suite added:
  `src/tests/modules/commandstack/` characterises the rename/remove/stack
  behaviour the in-tree consumers (punk::packagepreference, punk::nav::fs,
  punk auto_execok, packagetrace, packagesuppress) rely on, and pins known
  defects as _GAP_ tests (tokenid stuck at 1 causing duplicate-token dispatch
  bypass, unreachable get_IMPLEMENTOR 'builtin' branch, Delete_stack
  live-rename recursion).

## [0.49.1] - 2026-08-03

- punk::libunknown 0.2.4: deep tm discovery (`register_all_tm`, used by
  `dev lib.search`) now skips the G-155 workdirs `_mint` and `_bake`
  alongside legacy `_build` in both the directory walk and the tm-file
  filter - loose staged `.tm` files in `<srcdir>/_mint` were being
  registered as junk `_mint::*` modules.
- punk::tcltestrun 0.3.1 + runtestmodules.tcl: the unbuilt-modpod scan's
  subfolder excludes likewise missed the `_build` -> `_mint` staging
  rename - staged pods were registered as bogus `*::_mint::*` modules in
  test interps; now `-exclude-paths {**/_build/*** **/_mint/***}` (test
  tooling - listed for the record, not itself a bump trigger).

## [0.49.0] - 2026-08-03

- punk::path 0.5.0: `subfolders1` REMOVED (superseded by `subfolders`;
  no call sites existed repo-wide). `subfolders` ***-tail exclude
  patterns now prune the walk - `X/***` is handled internally as the
  equivalent pair `{X/** X}` - giving identical results while no longer
  descending the excluded subtree (measured ~250x faster on a bulky
  excluded tree); a bare `***` exclude now short-circuits to an empty
  result with a warning, like bare `**`. The missing punk::winpath
  dependency is now declared (normjoin errored in a minimal interp that
  required only punk::path). Help consistency pass: the *|**|***|? path
  glob syntax is documented once (shared argdoc text fragment) across
  globmatchpath, subfolders, treefilenames and the new pathglob_as_re
  argdoc; treefilenames no longer claims unsorted results and its list
  options are typed. `strip_prefixdepth` FIXED: it called a nonexistent
  `norm` (broken since its 2024 copy from
  punk::repo::path_strip_prefixdepth) and now uses the textual
  punk::path::normjoin; normjoin and strip_prefixdepth gain argdocs.
  Module test suite extended 40->67 tests, now covering every
  punk::path function in use in the codebase (relative, normjoin,
  pathtype, scriptlib_resolve and treefilenames_zipfs had no direct
  tests) plus strip_prefixdepth characterisation.

## [0.48.0] - 2026-08-03

- Launcher help + parsed subcommands (G-032, boot core 0.3.0): the kit
  launch surface self-documents through punk::args - `<punkexe> help
  ?subcommand?`, `<punkexe> -help` and `<punkexe> <subcommand> -help`
  render tabled usage (package modes + all subcommands including
  project-registered ones with their summaries; unambiguous subject
  prefixes resolve). New built-in subcommand `help`; an unknown first
  argument naming no existing file (and no lib:* reference) is now
  refused with usage on stderr (exit 1) instead of silently attempted
  as a script; buildinfo argument rejection carries the tabled usage
  beside its historic one-liner (exit 2 unchanged). Registration model
  extension `::punkboot::project_subcommand_info` (summary/argsid/
  package/parse - documented by punk::args::moduledoc::punkboot 0.2.0):
  project subcommands appear in help by registration alone and may opt
  their arguments into punk::args validation. Carve-outs: tool-style
  kits (unknownfirst -> project subcommand) keep help flags routed to
  their processor; tclsh intercepts a help flag only as its sole
  argument (stock dash parity); argsid-less registrations keep -help
  passthrough. Degradation per the G-030 doctrine: normal boots never
  load punk::args, and without the stack (or with PUNKBOOT_PLAIN=1)
  help is a plain subcommand list, parsing reverts to the historic
  switch semantics, boot never fails. moduledoc::punkexe 0.4.0
  documents the surface ((script)::punkexe::help added); pinned by
  src/tests/shell/testsuites/punkexe/launcherhelp.test on zipfs and
  metakit kits.

## [0.47.2] - 2026-08-03

- modpod 0.1.6 (G-111 tidy): the broken tar-era connection API is removed
  (connect/disconnect/get and their system:: helpers - dead since the tarjar
  era, no callers), is_valid_tm_version fixed, make_mountable_zip debug
  output removed (-offsettype file no longer errors under the module's
  declared dependencies), struct::set dependency dropped, and the load stub
  gains the unwrapped #modpod-<pkg>-<ver> redirect: an extracted folder
  beside the .tm is sourced directly - loads with neither zipfs nor vfs::zip
  available. First modpod testsuite added (src/tests/modules/modpod/)
  covering wrap/load round-trips including a probed binary payload; green on
  the tcl9 kit and native tclsh90 (7/7) and punk86 8.6 (6/7 - the
  zipfs-container test skips on 8.6, recorded per G-034).

## [0.47.1] - 2026-08-02

- Boot core 0.2.1: the built-in subcommands that require app packages
  (shellspy/punk/script/shell) fail gracefully by name when a kit does not
  carry the package - "script: this kit does not carry the app-punkscript
  package (...)" + exit 1 - instead of a raw package-require trace. Found
  adopting the shared core in the tomlish kits (which carry app-tomlish
  only); punkshell's own kits carry all app packages and are unaffected.
  tomlish itself now boots the componentized boot: its 547-line forked main
  is replaced by a thin declarative main (the G-031 derived-project proof).

## [0.47.0] - 2026-08-02

- Punk executables self-report build provenance (G-025): new 'buildinfo'
  launch subcommand prints a machine-parseable key=value report on stdout and
  exits 0 - the build STAMP baked into the kit (new punkkit-stamp.toml at the
  payload root: project version, source vfs folder, input runtime + its
  G-117 build_id when recorded, kit/target names, boot-core version -
  deliberately no time-of-bake field, so identical inputs stamp
  identically), LIVE facts (real executable path, patchlevel, payload base, package
  modes), the governing punkproject.toml version when findable
  (live_vs_stamp=same|differs), and the inherited runtime-layer
  punkbin-artifact.toml record. Identity comes from embedded data: a renamed
  or copied kit keeps reporting its stamp. Single implementation is the new
  punk::buildinfo module (0.1.0, dependency-free - works in the repl code
  interp and tclsh-hosted shells); the bare 'buildinfo' shell command is an
  aliascore alias (0.2.0) preloaded by punk::repl (0.5.4); unstamped contexts
  report stamp_present=0 rather than fabricating. Boot core 0.2.0 adds the
  buildinfo builtin and publishes ::punkboot::boot_state (core_version,
  main_script, package_modes, proj_scope, proj_root, kit_payload_base,
  src_project_root) - the single boot running-state surface G-089
  anticipated. punk::args::moduledoc::punkexe 0.3.0 documents the
  subcommand; new punk::args::moduledoc::punkboot 0.1.0 documents the G-031
  boot-core API (thin-main contract, registration model, persistent
  ::punkboot procs) runtime-queryably.

## [0.46.0] - 2026-08-02

- Generated projects get the componentized boot + its guidance (G-031
  increment 3): project_main.tcl reborn as the heavily commented thin-main
  SAMPLE (1132-line 2025-vintage monolith retired) - declare subcommands and
  tool-style launch defaults at ==CUSTOMIZE== points, opt-out documented; the
  project-0.1 layout _config now carries the thin punk_main.tcl,
  the sample and the punkboot_core.tcl master (hand-swept - layout _config is
  outside the sync channels), with punkboot/core.tcl seeded in the layout
  _vfscommon.vfs. src/vfs README (repo + layout) gains the boot-entry
  section: fauxlink-into-_config recommended path, plain root main.tcl
  alternative, one-startup-script rule (collision refusal), _config
  inertness; src/vfs/AGENTS.md updated (ownership split, G-129 consolidation
  note). E2E-proven on a freshly generated project: boots through its own
  trio via the sibling master, and a 'demo' subcommand added by editing ONLY
  the thin main dispatches in all three launch forms (explicit,
  noargs, unknownfirst).

## [0.45.0] - 2026-08-02

- Componentized kit boot (G-031 increment 2): every punkshell kit now boots
  through a THIN project-owned main.tcl that declares its launch surface and
  hands over to a shared layout-owned boot core - <vfsroot>/punkboot/core.tcl,
  delivered by the _vfscommon merge (master src/vfs/_config/punkboot_core.tcl,
  regenerated by 'make.tcl vfscommonupdate'). The core carries the boot
  machinery (G-058 static capture; zipfs/metakit/cookfs detection incl the
  G-129 backport mount derivations; package modes incl proj:; libunknown;
  src-mode modpod registration) plus the default dispatch, and is versioned
  (::punkboot::core_version, 0.1.0). Thin mains may declare project
  subcommands (::punkboot::project_subcommands - built-ins cannot be
  shadowed) and launch defaults (noargs/unknownfirst): the G-031 registration
  model. A main.tcl executed from the source tree finds its sibling master
  punkboot_core.tcl, so filesystem-driven launches keep working. Behaviour
  parity verified: full punkexe suite (110 pass / 0 fail), kitmountpoint
  across the 8.6-backport / 32-bit-modern / 64-bit runtime families, 11-kit
  default bake with no new warnings. Layout _config copies and the
  project_main.tcl thin-sample rebirth follow in the next increment.

## [0.44.0] - 2026-08-02

- `make.tcl bake` refuses a kit whose source .vfs carries two root-level
  suppliers of the startup-script name main.tcl - an actual main.tcl plus a
  fauxlink resolving to that name, or several such fauxlinks (G-031 collision
  gate). Previously merge_over materialised whichever supplier traversal met
  last (its fauxlink branch overwrites with no warning), so the built kit's
  boot script depended silently on directory order. Refusal is per-kit under
  FAILED KITS, before any build product is written: kits from other .vfs
  folders proceed and a previously deployed bin/<kit> is never replaced over
  the ambiguity. Folders with NO startup script stay legal (warning only, as
  before); nested app/main.tcl links (the undroidwish convention) are not
  root suppliers. Verdict predicate:
  punkboot::utils::vfs_startup_script_report (0.7.0), fixture-characterized
  in src/tests/modules/punkboot/utils/testsuites/utils/startupscript.test;
  same guarded-require degradation as the G-125/G-133/G-134 checks when the
  bootsupport snapshot is stale.

## [0.43.0] - 2026-08-02

- Launch package_mode `dev` renamed `minted` (clean break, NO alias): the mode
  that loads the mint output trees (<projectroot>/modules, lib etc) now names
  what it consumes, completing the stage-true set - src (unbuilt working tree)
  / minted (mint outputs) / internal (baked-in snapshots) / os (system
  packages) - and ending the overload with the `dev` shell command, the
  `-dev` kit scheme role and "dev modules" (magic-version sources), which all
  keep their names. `<kit> dev` on current kits now falls through to
  subcommand handling; old kits accept only `dev` until rebaked. Both boot
  mains, the punk::args punkexe moduledoc (0.2.0) and docs updated; layout
  _config copies hand-swept (they are outside the thin-layout sync channel).
  Deliberate versioning exception (alpha): recorded as minor despite the
  changed launch invocation.

## [0.42.0] - 2026-08-02

- Stage-true vocabulary part 2 (G-155): make.tcl speaks strict stage verbs -
  build = compile (buildsuites/tool only), mint = stamp/pack modules from src
  into the projectroot trees (the modules/libs/packages subcommands), promote
  = the bootsupport/vfscommonupdate gates, bake = kit assembly. `make.tcl
  workflow` now opens with a TERMINOLOGY key and the charter reads "one mint
  stage, two promotion gates, one bake"; help/summaries/guidance swept.
- Recapped warning tag renamed BUILD-WARNING -> BAKE-WARNING (every emitter is
  bake-domain; PROVENANCE-WARNING unchanged). Grep contracts and test pins
  updated; captured-output consumers should grep for BAKE-WARNING.
- Stage workdirs renamed to teach the vocabulary: kit assembly now uses
  src/_bake (was src/_build) and module staging <srcdir>/_mint;
  src/buildsuites/_build keeps its name (true compile area). Ignore rules
  updated in git + fossil; stale _build dirs are inert leftovers.
- bakelist deployed-state token nobuild -> nobake; detail label "bake
  product:". punk::mix::base 0.2.0 renames get_build_workdir ->
  get_bake_workdir (legacy alias retained) and retires the dead vfs-cksums
  quartet; punk::mix::cli 0.6.0 stages modpods under _mint;
  punkboot::utils 0.6.2 wording. build_modules_from_source_to_base keeps its
  name (punkcheck installer identity - G-156 owns any rename).

## [0.41.4] - 2026-08-02

- Generated projects complete their trailing fossil initial commit (G-146):
  every vendor/punk layout seeds .fossil-settings/binary-glob (*) plus a
  crlf-glob covering src/embedded/*, src/bootsupport/* and bin/*, so the
  binary payloads generation injects (zip-based .tm modules, bin/sdx.kit)
  and CRLF-carrying injected bootsupport files no longer prompt/abort the
  commit.

## [0.41.3] - 2026-08-02

- punk::args 0.14.0 (G-152): multiform parse failures now report a sound
  per-form suffix-viability verdict - a form is 'incomplete' (viable) only
  when the supplied words are a valid prefix of it and parsing stopped from
  exhaustion at end-of-input; words contradicting a form (e.g a wrong action
  word) now report it 'invalid' instead of the former false 'incomplete'.
  The verdict rides the noformmatch errorcode classes/-formerrors and the
  parse_status formstatus records (documented consumer contract in the module
  argdoc), and rendered noformmatch reports mark each candidate form line
  '(viable - needs more arguments)' or '(not viable)'.

## [0.41.2] - 2026-08-02

- make.tcl top-level overview interim layout (user decision): -choicecolumns
  back to 1 on both the overview and the lean dispatch definition - the long
  one-line summaries make the two-column layout too wide without table cell
  wrapping. Revisit -choicecolumns 2 when cell wrapping is implemented; the
  '## FORM' header-line dropping in the overview cells is accepted as-is.

## [0.41.1] - 2026-08-02

- make.tcl top-level help is a rich overview (user-directed, the repl 'i info'
  style): bare 'make.tcl help'/'make.tcl'/'-help' now render per-subcommand
  cells carrying the one-line summary plus that subcommand's auto-generated
  bracket-notation synopsis line(s) - per-form lines for the multi-form
  tool/buildsuite - grouped by the existing subcommand groups in two columns.
  Built with stock punk::args machinery (rich -choicelabels generated via
  punk::args::synopsis per subcommand id + -choicecolumns 2), hung on a
  separate (script)::punkboot.overview id resolved only on the help path so
  dispatch pays nothing; the lean top-level definition (now also
  -choicecolumns 2 for its unknown-subcommand error render) remains the
  dispatch surface, and the plain PUNKBOOT_PLAIN fallback help is unchanged.

## [0.41.0] - 2026-08-02

- make.tcl bake/bakelist kitnames are declaration-authoritative (user-directed,
  completing the G-143 arc): the configured kit names + @group selectors are now
  RESTRICTED punk::args choices, so an unknown kit is rejected at dispatch with
  the choice error (before any build - the no-build guarantee moves to the
  gate), unambiguous prefixes resolve to canonical names, and 'make.tcl help
  bake <line>' dry-run mirrors the verdict (the out-of-order 'bake punk91
  -confirm 0' is now a pointed choice error naming '-confirm' instead of a
  silent value-swallow). The handlers' own validation remains the backstop
  under PUNKBOOT_PLAIN degraded mode and when the definition-time mapping read
  fails (the choices clause is omitted entirely in that case - no gate).
- make.tcl kit-mapping reads consolidated to a single memoized parse per
  invocation (punkboot::lib::mapvfs_model): the definition-time kitname
  choices and every handler (selective bake, main bake, bakelist, vfslibs
  %platform% derivation, check's smoke listing) consume the same model
  (mapvfs_parse is pure, so cached and fresh models are interchangeable;
  success is cached, a throwing parse diagnoses fresh per caller; a
  PUNK_MAPVFS_CONFIG override keys its own cache entry).

## [0.40.3] - 2026-08-02

- make.tcl help dry-run purified (G-143 follow-up, user-directed): the 0.40.2
  bare-action carve-out is removed - 'make.tcl help buildsuite build' now falls
  through to the punk::args diagnosis like any other incomplete line (exit 1;
  the error table carries the form's argument rows and names the missing
  suitename, so the error IS the form documentation). Accepted lines gain a
  one-line received-args report after the usage table ('dry-run: line accepted
  (form build) - action = build | -test = 0 | toolname = punkzip'), which makes
  punk::args' value-swallow visible ('kitname = punk91 -confirm 0') - an
  interim make.tcl-side clue until punk::args grows an annotated success
  render.

## [0.40.2] - 2026-08-01

- make.tcl help is now a dry-run of the supplied command line (G-143 follow-up,
  user-directed): 'make.tcl help <subcommand> ?arg ...?' (and '<subcommand>
  ?arg ...? -help') parses the words through the subcommand's punk::args
  declaration exactly as dispatch would - an accepted line renders the matched
  action's single-form usage ('help tool build -test 0 punkzip', prefix words
  like 'help tool bu' included), while a rejected line (unknown action,
  option-first line, unknown flag in option position) emits the same punk::args
  noformmatch diagnosis dispatch gives - all-forms synopsis plus per-form
  reasons - on stderr with exit 1. One carve-out: a bare action word (or
  unambiguous prefix) renders that form's usage directly even where the real
  command requires further values ('help buildsuite build'). This replaces
  0.40.1's leading-word-only selection and its option-first whole-usage
  fallback: the parse error table is itself usage documentation, and help
  verdicts otherwise mirror dispatch verdicts exactly.

## [0.40.1] - 2026-08-01

- make.tcl help refinement (G-143 follow-up, user-directed): 'make.tcl help
  <subcommand> ?arg ...?' now accepts the subcommand's own command line and
  renders the usage most specific to it - 'make.tcl help tool build -test 0
  punkzip' (equivalently the same command line with -help appended) renders the
  tool build form via its leading action word; words after the action are
  tolerated and ignored, an option-first line falls back to the whole
  subcommand's usage, and single-form subjects (e.g 'help bake punk91 -confirm
  0') ignore the words entirely. The 0.40.0 numeric form-index acceptance
  ('help tool 2') is withdrawn as unintuitive - an unknown or numeric action
  word remains a pointed exit-1 usage error. This exceeds the repl 'i' alias,
  which tolerates trailing argument words but does not form-narrow on flat
  multi-form commands.

## [0.40.0] - 2026-08-01

- make.tcl help depth (G-143): the tool and buildsuite subcommands are multi-form
  punk::args definitions (one @form per action with a literal action leader), so
  'make.tcl help tool'/'help buildsuite' render one synopsis line per action, and
  'make.tcl help <subcommand> <action>' - equivalently '<subcommand> <action> -help',
  with a 0-based form index accepted in place of the action word - renders that
  action's single-form usage; unknown actions are pointed punk::args usage errors
  (exit 1). tool dispatch now parses through the definition: unknown actions and
  flags are usage errors (exit 1), and the declared positional model puts options
  before tool names ('make.tcl tool build -test 0 <name> ...' - the historic
  flag-anywhere order is no longer accepted; a misplaced flag earns a stderr hint,
  and the PUNKBOOT_PLAIN degraded scan keeps the historic parse). buildsuite
  driver-arg passthrough is unchanged. Per-subcommand explicit @form -synopsis
  overrides retired in favour of the G-144 automatic bracket-notation synopses
  (top-level make.tcl line kept). Pinned by
  src/tests/shell/testsuites/punkexe/maketclhelp.test.

## [0.39.6] - 2026-08-01

- punk::args 0.13.0 (G-144): auto-generated synopses for `(script)`-prefixed
  constructed definition ids now lead with the declared `@cmd -name` (the
  invocation name - e.g `make.tcl modules` for `(script)::punkboot::modules`)
  instead of the raw id, across the punk::args::synopsis full/summary renders
  and the usage/arg_error synopsis sections that funnel through them.
  `(autodef)` and other constructed id families are deliberately unchanged
  (their id-minus-tag is already the invocation, or no invocation-style
  -name exists); explicit `@form -synopsis` overrides still win, so make.tcl
  and punkexe help render as before - retiring those overrides is a follow-on.

## [0.39.5] - 2026-08-01

- Project-layout template payloads no longer self-censor in git (G-012): layout
  `.gitignore` payloads are stored inert as `gitignore.in` and materialized to
  `.gitignore` at project generation (punk::mix::commandset::layout 0.3.0 -
  rename map in `layout_materialize_renames`, staging-time rename after overlay
  composition, in-place fast path withheld from inert-payload folders).
  make.tcl's thin-layout sync now refreshes every vendor/punk layout's
  `gitignore.in` from the canonical root `.gitignore` (punkcheck-tracked), so
  generated projects receive current ignore rules and template files can no
  longer silently drop out of git tracking.

## [0.39.4] - 2026-08-01

- make.tcl piped output no longer leaks orphan ANSI fragments ('0;1m', '[0m', a lone
  'm' beside table borders) into usage/help tables (G-145). Root cause: the
  ansistrip channel transform's `clear` handler discarded the split-sequence carry
  at every write-chunk boundary - the Tcl core delivers `clear` to the transform
  stack before every write op (intentional but under-documented core behaviour,
  documented as read-side/seek-scoped). The transform no longer declares the
  `clear` op, so the carry survives chunk boundaries and split sequences are
  stripped whole; `finalize` still drops an end-of-stream partial sequence.
  Interactive and forced-colour output unchanged; pinned by a zero-remnant test
  across every help subject in maketclcolour.test.

## [0.39.3] - 2026-08-01

- make.tcl help restyle: every punk::args definition in `::punkboot::argdoc` is
  now a braced file-style block with `-&` record continuations and tstr
  placeholders (the punk module doc-block idiom; the SUBOPTS/OPT_SYNOPSES/
  VALUES_SYNOPSES/SUBVALUES define loop is unrolled into per-subcommand blocks).
  Help bodies are display fields expanded deferred, so each subcommand's
  Description now renders block-form with its HELPTEXTS indentation - the
  centred look of `make.tcl help tool` - instead of left-aligned (`@normalize`
  re-basing dropped). No parsing or interface changes; narrow usage tables
  widen slightly to fit the indented Description.

## [0.39.2] - 2026-08-01

- make.tcl tool build: installed exe mtime is now stamped with the install time.
  Previously `file copy` preserved the zig-out artifact's mtime and a zig
  cache-hit rebuild left that at the original link time, so once any
  non-build-input file in the tool tree (e.g. PROVENANCE.md) was newer, the
  tool list/info bin state reported 'stale' permanently - no `tool build` run
  could clear it.

## [0.39.1] - 2026-08-01

- make.tcl deprecated aliases removed (G-112 transition window closed): `project` and
  `vfs` no longer map to `bakehouse`/`bake` - they now produce an unknown-subcommand
  usage error. All alias plumbing is gone (dispatch mapping, punk::args definitions,
  the "deprecated aliases" help group, plain-help and `workflow` mentions).

## [0.39.0] - 2026-07-31

- Target-keyed kit output locations (G-127): a kit whose mapvfs entry declares
  a target other than the host's default now builds under
  `src/_build/kits/<platform>/` and deploys under `bin/kits/<platform>/`
  (per-tier punkcheck ledger; sidecar, arch scan, offset probe, smoke probe
  and sweep guards follow), while default-target kits keep their flat
  locations byte-identically. Same-named kits for different targets coexist -
  the pre-G-127 cross-run overwrite and runtime-name disambiguation for
  cross-target namesakes are gone; a kit name spanning targets selects all of
  them in `bakelist` and selective `bake`. `bakelist` marks such rows with
  `out=kits/<platform>/` and prints tiered paths in the detail block.
  RELOCATIONS: linux `punkshell902` -> `bin/kits/linux-x86_64/punkshell902`;
  the `verify-ix86` kits land at `bin/kits/win32-ix86/` on their next bake
  (pre-move flat copies are orphans to remove by hand).
- Per-platform payload declarations (G-127 on the G-115 surface): a
  `[payload.*]` source containing `%platform%` materializes once per CONSUMING
  kit target (platform set derived from the kit mapping) into VCS-ignored
  `_targets/<platform>/` staging inside the `.vfs` folder, and each kit's
  merged image receives only its own target's subtree. punk9linux.vfs's
  tcllibc entry migrated to the axis and its flat committed
  `lib_tcl9/tcllibc` tree was removed from VCS (the packages tier and the
  declaration govern it).

## [0.38.0] - 2026-07-31

- Kit mapping converted to toml (G-024, achieved): `src/runtime/mapvfs.toml`
  (tomlish-parsed `[kit.<name>]` / `[group.<name>]` / `[scheme.<name>]` tables)
  replaces the mapvfs.config line format, which remains readable as a
  deprecated fallback for un-migrated trees (toml wins when both exist).
  Strict entry-named validation (unknown keys/types/schemes, conflicting
  targets and parse failures are fatal; a full bake refuses a default entry
  whose vfs folder is missing and warn-skips entries whose runtime file is
  absent from its store tier). Named groups select in `bake`/`bakelist` as
  `@<group>`; `bake_default=false` entries are excluded from full bakes but
  bakeable by name/group. The G-023 `versioned` scheme expands one declaration
  into `punk<gen>-<version>` + `punk<gen>-dev` + the release-gated plain
  `punk<gen>` (created when absent; a normal bake never overwrites it) from
  the punkproject.toml version - no per-version config edits. New
  `PUNK_MAPVFS_CONFIG` env seam points bakelist/bake at an alternate mapping
  for characterization. Verified: legacy-vs-toml identical kit rows, and a
  forced punkluck86 rebuild content-identical across formats (identical
  runtime prefix, identical 2993-member payload).
- Declarative .vfs payload composition (G-115, achieved): a sibling
  `src/vfs/<name>.vfs.toml` declares package folders materialized INTO the
  `.vfs` folder (`[payload.<name>]`: source, source_root src|packages, target,
  supersedes, replace) by `make.tcl vfslibs` and the bake phase -
  punkcheck-tracked with drop-in-wins precedence (`-overwrite synced-targets`:
  undeclared or hand-modified files are preserved and reported, never
  overwritten; supersedes/replace are the explicit exceptions). A `.vfs`
  with no declaration is untouched. The per-package
  `src/runtime/vendorlib_vfs.toml` surface is migrated into per-.vfs files
  and retired (a leftover copy is ignored with a warning). Format +
  precedence spec: `src/vfs/README.md`; the project-0.1 layout seeds the
  convention (mapvfs.toml + vfs README) for derived projects. New
  demonstration kit `punkdeclare` (group `declare-demo`, bake_default=false):
  VCS carries only its boot fauxlink and the whole binary payload (tcludp
  vendor tree + tcllibc packages tier) rematerializes from the declaration -
  proven by clean-tree rebuilds with passing in-kit smoke requires and by
  drop-in survival runs.

- Buildsuite artifact identity fields (`project`, `project_url`, `packager`)
  now default from the enclosing project's `punkproject.toml` `[project]`
  table (G-117/G-123 lineage): both suite recipes locate the nearest
  manifest by a bounded walk-up from the build root, so a copied suite
  reads its own project's file. Chains - `project`: `-Dproject` > manifest
  `name` > "punkshell" (previously hardcoded); `project_url`:
  `-Dprojecturl` > manifest `url` > "unrecorded"; `packager`: `-Dpackager`
  > `PUNKBIN_PACKAGER` env > manifest `packager` > git identity > `fossil
  user default` (new fossil-only fallback) > "unrecorded". Verified
  end-to-end in both suites' emissions. The root `punkproject.toml` gained
  an active `url`, and `dev project.new`'s seeded manifest now carries
  commented `url`/`packager` placeholders documenting where those fields
  are consumed (punk::mix::commandset::project 0.4.1).

## [0.36.0] - 2026-07-31

- `punk-runtime` fetch reports the artifact's build-origin class after
  retrieving its metadata sidecar (`provenance: class=<c>` - table-aware
  schema-v2 read), completing the G-123 consent surface: a trusted fetch from
  any origin now makes the artifact's provenance class visible in the fetch
  report. Both payloads; `bin/punk-runtime.cmd` re-wrapped; pinned in the
  provenance characterization suite.

## [0.35.1] - 2026-07-31

- `bin/dtplite.cmd` repaired (G-139 fallout, present since the vendored-tree
  removal on 2026-07-30): its no-tcllib fallback still pointed at the retired
  `src/vendorlib_tcl9/<arch>/tcllib*` trees, so every execution usecase failed
  with "can't find package dtplite". The fallback now resolves the artifact-fed
  DEPLOYED tree `lib_tcl9/allplatforms/tcllib<ver>` first (vendorlib patterns
  kept for derived projects that still vendor); dtplite.test back to 8/8.

## [0.35.0] - 2026-07-31

- `punk-runtime` fetch now retrieves the beside-toml metadata sidecar for ANY
  artifact name, not only `-r<N>` family names - bare pre-family artifacts
  carry G-123 retroactive schema-v2 sidecars on the server, so a fetched
  third-party runtime arrives with its provenance record (absence stays
  tolerated: toml-less entries keep degrading as no-basis rows). Both
  payloads; `bin/punk-runtime.cmd` re-wrapped.
- New characterization suite
  `src/tests/shell/testsuites/binscripts/runtimecmd_provenance.test` (G-123):
  fixture-server coverage of the server-trust gate (refusal + flag + env
  acknowledgement forms), bare-name sidecar fetch, `class=` listing tags with
  suite-built quiet and no-basis degradation, and `info` schema-v2 field/
  `provenance_class` parity between the payloads.

## [0.34.0] - 2026-07-31

- `punk-runtime` fetch now carries the G-123 server-trust consent gate (one
  vocabulary with `make.tcl libfetch`'s `-trust-server`): fetching runtime
  binaries from a non-canonical origin (`PUNKBIN_URL` override) requires the
  explicit `-trust-server` flag or `PUNKBIN_TRUST_SERVER=1` (the unattended
  form); the canonical punkbin origin gates nothing, and the metadata-only
  listings (`list -remote`, `platforms -remote`) never gate - never an
  interactive prompt. Both payloads; `bin/punk-runtime.cmd` re-wrapped, with
  a spacer adjustment clearing the 512-byte cmd label-scan boundary the
  payload growth pushed a template label onto (caught by scriptwrap
  checkfile exactly as designed).

## [0.33.0] - 2026-07-31

- Runtime artifact records move to schema v2 (G-123): `[provenance]` gains the
  build-origin `class` line (`suite-built | third-party | local`), emitted by
  the suite_tcl90 kit-family steps (both record copies; family_check asserts
  it) and documented in the punkbin repo alongside 12 retroactive sidecars
  authored for every pre-family artifact on the server (evidence-based
  classes; additive - artifact bytes and names unchanged).
- `punk-runtime` (both payloads, `bin/punk-runtime.cmd` re-wrapped): local
  listings and `list -remote`'s sparse identity column now carry a compact
  `class=<build-origin>` tag for third-party/local runtimes (suite-built rows
  stay quiet; tags derive from local records only - no per-row remote
  fetches), and `info` shows the schema-v2 provenance fields (`builder`,
  `source_url`, `upstream_ref`, `retrieved`) plus a table-aware
  `provenance_class` row (the flat field scan deliberately keeps reporting
  `[artifact] class` per the documented v2 ordering caveat).
- `punk-runtime` bash payload fix: `info -platform <p>` was advertised but
  silently ignored ('info' was missing from the option-scan action list) -
  the per-name identity report now honours the tier argument like the ps1
  payload.

## [0.32.3] - 2026-07-31

- `punk-runtime` bash payload: `fetch` and `list`'s local-file comparison now
  strip CR characters before matching sha1sums.txt entries, fixing the
  "Unable to locate hash" abort when the artifact server publishes
  sha1sums.txt with CRLF line endings (GNU/BSD grep anchor `$` strictly, so
  the end-of-line filename match failed even with the entry present; the ps1
  payload was unaffected via Get-Content line splitting). All sha1sums lookups
  in the payload now share the CR-stripping form `list -remote` already used.

## [0.32.2] - 2026-07-30

- Linux lib-tier emission (G-140, achieved): suite_tcl90 gains a `tcllibc-linux`
  step cross-building the tcllib critcl accelerators for linux-x86_64 (critcl
  driving `zig cc -target x86_64-linux-gnu.2.17` - dynamic-glibc floor, stubs
  linkage, no linux build host), structurally ELF-gated on the build host
  (`tools/elf_gate.tcl`) and emitted as `lib/linux-x86_64/tcllibc-tcl9-r<N>.zip`
  by the `library-artifacts` step. The punkshell902 (punk9linux) kit payload now
  ships this artifact-fed tcllibc with its embedded punkbin record - the
  hand-dropped tcllibc + stray tcllibc2.0 folders are retired
  (`libpackages.toml` + `vendorlib_vfs.toml` declarations; accelerator engage
  verified under WSL for both the materialized tier and the baked kit).
- library_artifacts build_id seeds now include the tier path so same-named
  artifacts in different target tiers (win32/linux tcllibc) get distinct
  identity digests.

## [0.32.1] - 2026-07-30

- `make.tcl libfetch` gains env-only isolation seams: `PUNK_LIBFETCH_CONFIG`
  (alternate declarations file) and `PUNK_LIBFETCH_PACKAGES` (alternate tier
  root), so fixture/mirror runs never read the real declarations or write the
  real `bin/packages` tier; neither seam bypasses the `-trust-server` consent
  gate. New characterization suite
  `src/tests/shell/testsuites/punkexe/maketcllibfetch.test` pins the G-139
  behaviours previously verified only by manual session evidence - the
  server-trust gate (refusal with exit 3 before any network access),
  `-serverurl`-over-`PUNKBIN_URL` precedence, fetch/verify/materialize,
  idempotency vs `-force`, declared-revision-change re-materialization,
  sha1-mismatch rejection and the no-config no-op - against
  httpfixture-served and `file://` fixture origins (the canonical origin is
  never contacted).

## [0.32.0] - 2026-07-30

- Kit payloads switched to punkbin lib-tier artifacts (G-139): the 8.6-family
  kits upgrade tcllib 1.21 (+its 2023 tcllibc 0.4 and md5c) to suite-built
  tcllib 2.0 + current tcllibc; the tcl9 kits' hand-dropped tcllib2.0 copies
  (which nested md5c/tcllibc) are replaced by the published artifacts with
  sibling tcllibc. Every installed package tree in deployed lib_tcl<N> and
  kit payloads now carries an embedded punkbin-artifact.toml provenance
  record. The vendored src/vendorlib tcllib trees are retired (the first
  major G-004 payoff). Validated by kit-hosted runtests on punk86 against a
  same-day baseline (identical results). r1 artifacts published to the
  punkbin origin; `make.tcl libfetch` verified live against it.

## [0.31.1] - 2026-07-30

- `make.tcl libs`/`packages`/`bakehouse` gain the PACKAGES_tcl<N> phase
  (G-139): materialized punkbin lib-tier packages under `bin/packages/`
  deploy into `lib_tcl<N>/allplatforms` + `lib_tcl<N>/<platform>` with
  punkcheck provenance (silent no-op when the tier is absent). The vfslibs
  declaration schema (`src/runtime/vendorlib_vfs.toml`) gains `source_root =
  "packages"` (sources resolved against the bin/packages tier, with an
  actionable libfetch hint when missing) and `replace = true` (clean-slate
  same-name refresh - supersedes deliberately cannot name the folder being
  installed). tcllib/tcllibc enter the kit vfs declarations: tcl9 kits
  switch their hand-dropped tcllib2.0 (which nested md5c/tcllibc) to the
  suite-built artifact pair; tcl8 kits upgrade tcllib1.21 (+nested tcllibc
  0.4/md5c) to tcllib 2.0 + current tcllibc via supersedes; punk9linux
  (linux target) gets only the platform-neutral tcllib2.0.

## [0.31.0] - 2026-07-30

- New `make.tcl libfetch` subcommand (G-139 phase 1): fetches the punkbin
  lib-tier library artifacts declared in `src/runtime/libpackages.toml`
  (suite-built tcllib/tcllibc zips, both Tcl generations - G-138) into the
  untracked `bin/packages/<target>/` input tier with sha1 verification
  against the server's per-target sha1sums.txt, and materializes the
  installed-shape package trees under `bin/packages/<target>/tcl<N>/`
  (self-describing via their embedded punkbin-artifact.toml records).
  Consent is keyed to server trust: the canonical punkbin origin needs no
  flag; any other origin (`-serverurl` / env `PUNKBIN_URL`) requires the
  explicit `-trust-server` flag. `file://` origins serve local mirrors and
  fixtures. Idempotent: present artifacts that verify are skipped and
  current trees are left untouched (`-force` overrides).

## [0.30.7] - 2026-07-29

- Baked kits pinned archive-relative (G-134): every assembled kit image is
  probed post-assembly with `punk::zip::archive_info`
  (`punkboot::utils::kit_offsetstyle_report`, punkboot::utils 0.6.0) and a
  FILE-relative attached zip payload earns a recapped BUILD-WARNING -
  advisory, the kit still builds and deploys. No-zip results (the metakit
  kit shape) and bare zips stay silent; today's bakes warn on nothing
  (measured baseline). `make.tcl check` reports the pin
  ACTIVE/UNAVAILABLE with the same guarded-require degradation as the
  G-125/G-133 checks. `punk::zip::mkzip -offsettype file` remains available
  for deliberate manual file-relative artifacts - the pin asserts only what
  the pipeline emits.

## [0.30.6] - 2026-07-29

- punkres is publishable as a punkbin artifact (G-128 distribution route):
  `win32-x86_64/tools/punkres-x86_64-windows-<ver>.exe`, sha1-verified via
  the tools tier's `sha1sums.txt`, placed at `bin/punkres.exe` - the
  no-toolchain route that does not depend on the committed twapi. 0.3.1
  staged in the punkbin repo (push maintainer-gated); route recorded in
  `src/tools/punkres/PROVENANCE.md` and `bin/AGENTS.md`, and the kit-icon
  seam's skip notice now names both remedies (build from vendored source,
  or fetch the prebuilt artifact).

## [0.30.5] - 2026-07-29

- `punk-getzig.cmd` fixed for spaced checkout paths and Windows PowerShell 5:
  the launcher moves off the `pwsh -c` form (which tokenized a spaced script
  path - same `cmd.exe /c powershell ... -File` fix punk-runtime received in
  G-119) and the payload drops its pwsh-isms: `Resolve-Path
  -RelativeBasePath` (7.4+) becomes the Push-Location `-Relative` form,
  `Invoke-WebRequest` gains `-UseBasicParsing` (PS5 IE-engine
  NullReferenceException) with progress suppressed for the large download,
  and two `Invoke-Expression "$zigexe version"` sites (which split spaced
  paths under ANY powershell) become call-operator invocations. Verified end
  to end from a "program files" checkout: mirror-latency selection, fetch
  from the local punkbin mirror, minisign verification, extraction and both
  the fresh-install and already-installed branches.

## [0.30.4] - 2026-07-29

- `punk-runtime fetch` now prints a one-line materialization hint when the
  fetched `-rN` artifact's WORKING name (what mapvfs and bakes reference) is
  absent: "note: bakes and projects reference <working> - materialize it
  with: punk-runtime use <artifact>". Both payloads (ps1 + bash, parity
  wording); silent once the working copy exists. Closes the surprise where
  fetch reports an active runtime but `make.tcl bakelist` still reads
  runtime=missing until `use` runs.

## [0.30.3] - 2026-07-29

- Fleet renamed: every win32 kit now carries the embedded `PUNKSHELL` group
  icon via punkres (punksys, punk86, punk91, punkbi, punkluck86, punk902z,
  punk905, punk91ix86, punk9_beta, punk9bi_beta; punkmagic excluded - its
  runtime is absent; punkshell902 is a linux target, not applicable).
- punkres 0.3.0: new `set-icon -drop-opaque-overlay` consent DISCARDS non-zip
  trailing stub bytes (the COFF symbol/debug tables mingw-built runtimes
  carry, which ride into extraction heads and previously made punkres refuse
  those stubs), zeroing the orphaned symbol-table header pointer; distinct
  from `-allow-opaque-overlay` which RELOCATES a real self-locating payload.
  The bake's punkres arm passes the drop flag - matching the twapi arm's
  measured long-standing EndUpdateResource stripping - so mingw-lineage stubs
  (tclsh902z, tclkit-win64-dyn family) stamp identically to MSVC/zig ones.

## [0.30.2] - 2026-07-29

- Baked kits' embedded RT_GROUP_ICON is now NAMED from the icon file's
  uppercased rootname - `PUNKSHELL` for `punkshell.ico` (default and per-kit
  overrides alike) - instead of retaining the stub lineage's name ("SFE",
  "TCLSH"). Applied identically by both icon mechanisms (punkres 0.2.0
  `set-icon -group`; the twapi arm gained the matching override), language
  still adopted from the replaced group. The `<kit>.resources.toml` sidecar
  records the name as `icon_group`. Already-deployed kits keep their old
  group name until their next rebuild.

## [0.30.1] - 2026-07-29

- Portable PE resource stamping (G-128, first increment): new vendored zig tool
  `src/tools/punkres` (upstream c:/repo/jn/zig/punkres, BSD-2-Clause) builds to
  `bin/punkres(.exe)` via the existing `make.tcl tool` step with zero make.tcl
  edits. punkres stamps RT_ICON/RT_GROUP_ICON into a FINISHED PE from any build
  host and reads its own work back (info/list/extract-ico + built-in set-icon
  verification), preserving an appended zip payload: archive-relative moved
  verbatim, file-relative refused by default and shifted (offsets rewritten,
  convention preserved) with `-allow-file-relative-shift`, zip64/multi-disk
  always refused, non-zip overlays moved only with `-allow-opaque-overlay`.
- The G-057 kit icon seam now selects punkres when built (any host), with the
  twapi arm as the unchanged fallback; sidecar `embedding` records which
  mechanism ran (`punkres` | `twapi` | `none`). Parity verified: both
  mechanisms produce byte-identical icon resources from the same input.

## [0.30.0] - 2026-07-29

- Kit builds carry a configurable icon (G-057): every baked kit ships a
  `bin/<kitname>.resources.toml` sidecar recording the build-time icon choice
  (default `src/runtime/punkshell.ico`; per-kit override = root `punkshell.ico`
  in the kit's own custom `.vfs` folder), with icon identity and provenance
  hashes reaching back through the G-135 assetorigin records to the SVG master.
  On win32 targets baked by a twapi-capable host the icon is embedded as PE
  RT_ICON/RT_GROUP_ICON resources - the payload-free runtime stub copy is
  stamped before the vfs payload is attached (tcl-sfe mechanism, Ashok P.
  Nadkarni), so built kits now wear the punkshell mark in Explorer and the
  taskbar instead of their runtime's icon. Non-PE targets skip as not
  applicable; incapable hosts (no twapi / non-windows build host) skip with a
  distinct notice; the sidecar is written in every case and runtime store
  originals are never modified.

## [0.29.0] - 2026-07-28

- `proj:` package-mode scope prefix (G-033): an installed or standalone punk shell
  binary can visit the punk project containing the current directory -
  `<punkexe> proj:internal-src shell` (canonical, kit copies win same-version ties)
  or `proj:src` (project copies win). The boot walks up from the cwd to the nearest
  git/fossil repo root with a punkshell-style src tree and feeds it to the existing
  src/dev path machinery (src/modules + bootsupport + vendormodules + src/lib,
  `package prefer latest`, #modpod registration), reporting the detected root and
  effective precedence on stderr - warn-and-proceed (dev/src unbound) when no
  project is found, never an exe-relative fallback rebind. Exe-relative plain
  `src`/`dev` launches are unchanged. The packagemode contract (blocks, ordering,
  proj: examples) is now live punk::args documentation in
  punk::args::moduledoc::punkexe 0.1.1 (rendering wiring remains G-032).

## [0.28.4] - 2026-07-27

- punk::zip 0.3.0 (G-126): optional punkzip extraction accelerator. New
  `punk::zip::accelerator` (auto|none|<path>; auto probes env(PUNKZIP_EXE) then a
  punkzip beside the executable, resolution cached); `unzip` hands whole-archive member
  writing to the binary when the call is one it serves identically (whole archive,
  default -overwrite/-mtime/-verify, ascii names) and re-stamps mtimes to punk::zip's
  local-time convention afterwards, falling back to pure Tcl silently otherwise
  (punk::zip::last_unzip_engine / last_accelerator_note record the decision). Preflight
  refusals, member selection and returned names always come from punk::zip's own
  reader. Parity suite src/tests/modules/punk/zip/testsuites/zip/zipaccel.test proves
  identical names, bytes and mtimes through both engines over the G-124 shapes and the
  real file-relative kit runtime. Measured: material from ~50 members; x2.3-2.6 at kit
  scale under a native tclsh; x3.8-4.9 kit-hosted (e.g. punk91.exe 21.9s -> 5.7s).

## [0.28.3] - 2026-07-27

- New `make.tcl tool` subcommand (G-126): surface for the vendored first-party zig
  build tools under src/tools (punkzip first). `tool list|info` report each tool's
  version, declared zig floor, installed-binary state and vendored upstream commit
  (from its PROVENANCE.md); `tool build ?name ...? ?-test 0|1?` resolves a pinned
  toolchain (PUNK_ZIG override, else the lowest satisfying release under
  bin/tools/zig*), runs the tree's own 'zig build test' as an exit-code gate, builds
  ReleaseSafe and installs to bin/. zig stays optional: packages/bake never require
  the step, and without a suitable toolchain the state is reported (build/test exit
  nonzero with bin/punk-getzig.cmd guidance). Piped characterization:
  src/tests/shell/testsuites/punkexe/maketcltool.test.

## [0.28.2] - 2026-07-27

- Bakes verify payload/target consistency (G-133, both checks advisory - recapped
  BUILD-WARNINGs, the kit still builds and deploys): a structural binary-arch scan of
  each merged kit tree (PE/ELF/Mach-O headers vs the kit's target platform, platform
  subdirs like win32-ix86/ and vendor spellings like win-x64/ exempt, cross-target kits
  included) plus a post-build smoke-require probe (packages declared as a mapvfs.config
  5th entry element are plain-'package require'd inside the freshly built artifact via
  its tclsh subcommand - the only check that sees wrong-arch version shadowing).
  `make.tcl check` reports both. First real sweep flagged zint.dll (32-bit) carried by
  the x64 tcl8 kits punk86/punkbi/punksys - a genuine latent load failure.

## [0.28.1] - 2026-07-27

- The win32-ix86 verification kits (punkluck86 / punk91ix86) wrap new MINIMAL payloads
  `punk8min.vfs` / `punk9min.vfs`: boot fauxlinks plus multi-arch binary packages only
  (iocp-2.0.2 both majors, twapi-5.0b1 for tcl8, via directory fauxlinks - no binary
  duplication), pure-tcl set via the dll-free `_vfscommon.vfs` overlay. The previous
  full 64-bit payloads broke basic shell functionality on the 32-bit kits: a
  higher-versioned x64 Thread shadowed the LUCK runtime's working 32-bit copy, and
  x64-only dlls could never load. Both kits now run a working piped shell session and
  resolve Thread/iocp/twapi (punkluck86 additionally udp) at correct architecture.

## [0.28.0] - 2026-07-27

- Kit boot finds its payload wherever the runtime mounted the attached archive (G-129):
  zipfs presence now keys on `tcl::zipfs::mount` and the payload mount base derives from
  the runtime's own mount table (the entry whose archive file is the executable) instead
  of assuming `tcl::zipfs::root` + `//zipfs:/app`. The androwish/undroidwish Tcl 8.6
  zipfs backport - no `root` command, archive mounted at the executable's own path - now
  carries a working punk kit. A mounted archive the boot cannot attribute to the
  executable is reported on stderr instead of silently configuring no internal paths.
- New 32-bit windows verification kits in the bake matrix (target win32-ix86):
  `punkluck86` (LUCK tclsh 8.6.10 zipfs-backport runtime) and `punk91ix86`
  (tcl-sfe 9.1b0). Kit-wrappable runtime requirements documented in `bin/AGENTS.md`.
- `punk8win.vfs` payloads carry `app/main.tcl` beside root `main.tcl` (the undroidwish
  family's in-archive startup hook; inert elsewhere). make.tcl fauxlink targets now
  resolve link-location-relative (symlink semantics; identical for root-level links).
- make.tcl's runtime capability probe keys `has_zipfs` on `tcl::zipfs::mount` (was
  `tcl::zipfs::root`, which blocked zip assembly for the backport family).

## [0.27.4] - 2026-07-27

- `make.tcl bootsupport` no longer fails to replace a mounted modpod. A zip-based `.tm`
  cannot be overwritten in place while any process has it zipfs-mounted, because the mount
  memory-maps the archive and windows refuses to overwrite a file with a user-mapped
  section open (`ERROR_USER_MAPPED_FILE`). The update now falls back to delete-then-place,
  which windows does permit - unlinking a mapped file leaves the holder reading its own
  mapping while new content lands at the name. Replacement content is staged to a sibling
  temp first, so a mid-sequence failure can never leave the target missing.
- Failures of that kind are now explained instead of being reported as
  `error copying ...: invalid argument`. That message is Tcl's catch-all:
  `Tcl_WinConvertError` maps only Win32 codes 0..267 and sends everything above them to
  `EINVAL`, so every high-numbered windows error arrives with the same uninformative text
  (identical in 8.6 and 9.x). `make.tcl` now prints what it actually means and which
  process class to look for.
- The mechanism, and the warning not to read `invalid argument` at face value on windows,
  is documented in `src/bootsupport/AGENTS.md`.

## [0.27.3] - 2026-07-27

- `bin/punk-runtime.cmd list` and `use` no longer offer support files as selectable
  runtimes. The payloads already filtered `.txt/.toml/.tm/.tmp/.log`; the recognised set
  now also covers docs/text (`.md .html .htm .pdf .url`), config/metadata
  (`.json .xml .yml .yaml .ini .cfg`), dev leftovers (`.bak .old`), checksums and
  signatures (`.sha1 .sha256 .asc .sig`) and archives (`.zip .tar .gz .bz2 .xz .7z`).
  Prompted by a runtime tier where a build-request `.cfg` and a `.url` shortcut were
  listed as runtimes.
- The set is now defined ONCE per payload (`$PunkRuntimeSupportExt` /
  `Test-PunkRuntimeSupportName` in the powershell payload, `punk_support_ext_re` /
  `is_punk_support_name` in the bash payload) instead of repeated at each of the nine
  filter sites, so the two payloads cannot drift - their listings are required to agree
  byte-for-byte.
- `bin/runtime/README.md` now states what a platform folder is for and what is tolerated
  in it: anything not recognised as a support file is offered as a runtime, so unrelated
  executables do not belong there.

## [0.27.2] - 2026-07-27

- The bake's boot-precondition gate (G-125) recognises a third tcl-library convention:
  `tcl<major>.<minor>/` at the vfs root, used by runtimes whose attached archive mounts at
  the executable's own path rather than at `//zipfs:/app` - so `[info library]` is
  `<exe>/tcl8.6`. The androwish/undroidwish zipfs backport for Tcl 8.6 mounts this way, and
  a kit built on such a runtime was being refused as unbootable even though its library was
  present and complete. `punkboot::utils` 0.3.0 -> 0.4.0; the report's `checked` list gains
  the new location, and `make.tcl check` and `make.tcl workflow` name all three.
- The root `tcl8/`/`tcl9/` module trees this project's own kits carry are matched by the
  same glob but carry no `init.tcl`, so they are examined and correctly not mistaken for a
  tcl library. All ten assembled kit trees still pass with unchanged detected locations.

## [0.27.1] - 2026-07-27

- Third-party `pkgIndex.tcl` files that define commands at index-source time for their own
  `package ifneeded` scripts to call now work again (punk::libunknown 0.2.2 -> 0.2.3).
  0.2.0 moved index sourcing into a proc of the `punk::libunknown` namespace to stop index
  helper variables leaking into the global namespace; that also changed the *current
  namespace*, so an index's relative `namespace eval foo` created
  `::punk::libunknown::foo` instead of `::foo` - and stock `tclPkgUnknown`, being a proc in
  `::`, has always given indexes global resolution. Such packages failed at require time
  with `invalid command name`. The index body now runs via `apply` with `::` as its
  namespace, keeping the frame isolation and restoring the resolution context.
  Found on twapi 4.7.2 as shipped in a third-party 8.6 kit; the pattern is common.

## [0.27.0] - 2026-07-27

- A kit that cannot boot is no longer deployed (G-125). `make.tcl bake` now checks each
  kit's merged `.vfs` for a tcl library before building it; a kit with none is listed under
  `FAILED KITS` with a reason naming the cause and NOTHING is written - no
  `src/_build/<kit>`, no deploy, and the previously deployed `bin/<kit>` is left
  byte-identical. Previously the build warned and carried on, deleting the working kit and
  replacing it with an artifact that died at startup with
  "application-specific initialization failed: Cannot find a usable init.tcl".
- The gate tests the boot precondition, not whether extraction ran: a `.vfs` that supplies
  its own tcl library builds and deploys exactly as before. The check is structural and
  executes nothing, so it covers cross-target kits this host could never run, and it is
  cheap enough to run for every kit on every bake.
- `tclsh src/make.tcl check` reports the gate as ACTIVE or UNAVAILABLE, and
  `tclsh src/make.tcl workflow` shows it in the kit data flow (key note K10). A stale
  bootsupport snapshot degrades the gate to a NOTE rather than failing every kit - the same
  guarded-require treatment the dirty-src provenance check already had.
- New `punkboot::utils::vfs_boot_library_report` (0.2.0 -> 0.3.0) is the predicate behind
  the gate, with its own test suite over directory fixtures.

## [0.26.0] - 2026-07-26

- punk::zip can READ a zip archive using only stock Tcl - no zipfs, no vfs::zip, no tcllib
  (G-124). Three new shell-reachable commands: `punk::zip::members` lists what is inside an
  archive without extracting it (per-entry name, directory-vs-file, sizes, compression
  method, mtime, crc and the stored attributes), `punk::zip::unzip` extracts all or part of
  it to a directory with every member's crc32 verified, and `punk::zip::archive_info`
  reports where the archive begins inside its host file and which offset convention it
  uses. All three work on a plain .zip and on a zip attached to an executable or script
  prefix - a punk kit, a tclsh carrying a zipfs image, a zip-based .tm modpod - so
  "what is in this thing" is now answerable on a runtime without zipfs.
- Reading an executable-prefixed archive no longer depends on splitting it first. The base
  offset is derived from the central directory and the members are read from the original
  file, which makes the archive-relative and file-relative offset conventions
  indistinguishable to callers. A file-relative artifact (`tclsh90b4_piperepl.exe` in the
  runtime store) previously split into a .zip that tcllib rejected with
  "Bad zip file. Bad closure."; it now reads and extracts.
- make.tcl's zipfs-less kit extraction consumes that reader instead of tcllib's
  `zipfile::decode`, which was a system package punkshell does not vendor. A tclsh with
  neither zipfs nor tcllib (msys2's `/usr/bin/tclsh8.6`) can now carry a runtime's
  `tcl_library` into the kit it bakes, where before the bake warned and produced an
  artifact that could not initialise.
- Unsupported archives fail by name rather than emitting partial output: encrypted members,
  zip64 archives, unknown compression methods, crc mismatches and member paths that would
  escape the target directory are each refused before anything is written.
- First test coverage for punk::zip (`src/tests/modules/punk/zip/`), including the
  mkzip -> read round trip that verifies what punkshell writes is readable.

## [0.25.0] - 2026-07-26

- make.tcl host/target platform split (G-122): everything a bake EMITS is now keyed by the
  artifact's TARGET platform rather than the driving tclsh's personality - the
  bin/runtime/<tier> store a runtime is read from, .exe suffixing of runtime files and kit
  outputs, presence checks, and whether the pre-deploy process sweep uses tasklist/taskkill
  or ps/kill. Host semantics (copy commands, path handling, filesystem case rules, prompts)
  stay keyed to the host. The default target is the host's platform canon EXCEPT for a
  cygwin-family host (msys2/cygwin-runtime tclsh: reports tcl_platform(platform) unix on
  windows), which targets win32-x86_64 - such a host now drives the identical kit set,
  names and store addressing as a native tclsh, where before it reported every kit
  runtime=missing against a nonexistent mingw64-x86_64 store.
- mapvfs.config entries take an optional 4th element naming the kit's target platform, so a
  runtime kept in another platform's store tier is a first-class mapping entry: it is read
  from that tier, named with that platform's executable convention, and skipped by the
  process sweep. The existing linux tclkit entry now declares linux-x86_64 and bakes to a
  suffixless bin/punkshell902.
- make.tcl check reports the derivation on one line: `platform (G-122): host=... target=...
  store=... exe-suffix=... process-tooling=...`, with the tcl_platform values it came from.
- Zip-type kits assemble without zipfs in the driving tcl: when tcl::zipfs::mkimg is absent
  (any 8.6) the image is built by raw-runtime split + punk::zip::mkzip + concatenation - the
  same helper the zipcat kit type uses. Both mount identically (archive-start-relative
  offsets). Note the runtime's own zip is still EXTRACTED via tcllib's zipfile::decode,
  which is not vendored; without it the bake emits a recapped BUILD-WARNING that the kit
  will carry no tcl_library.
- Fixes exposed by the above: the runtime capability probe looked for a suffixless filename
  and so reported find-fail for every runtime on windows (its results now feed the
  extraction-type choice, which additionally now tests each capability's VALUE rather than
  its mere presence); BUILDCOPY names follow the runtime's own filename; and native windows
  command lines run through a guard that suppresses msys2's posix-path argument rewriting -
  without it `taskkill /PID <n>` reached taskkill as `taskkill C:/<msysroot>/PID <n>` and
  the pre-deploy sweep could find a running kit but never replace it.
- make.tcl diagnoses being handed a windows-style script path by a cygwin-family (POSIX)
  tclsh, which silently resolves it relative to the cwd, and names the path spelling to use.
- punk::platform: the msys2 family (mingw64/mingw32/ucrt64/clang*) folds onto one stable
  `msys` tag and CYGWIN_NT onto `cygwin` - msys2's uname renders that token from MSYSTEM,
  so one binary otherwise canonizes differently per launching shell. New
  punk::platform::normalize_os; cygwin-x86_64 added to the platform table; 'help platforms'
  documents the fold.

## [0.24.1] - 2026-07-26

- make.tcl bakelist/bake: a nonexistent per-platform runtime store folder is now flagged
  loudly instead of silently yielding all-missing rows - bakelist prints a stderr WARNING
  showing the platform-canon derivation (tcl_platform os/platform -> canon) and marks the
  header line (FOLDER MISSING); bake's no-runtimes exit carries the same self-diagnosis.
  Classic trigger: an msys/cygwin-runtime tclsh (reports platform unix, canon e.g
  mingw64-x86_64) deriving a store name that is not the windows store the kits use.
- make.tcl shell: 'help platforms' now renders (punk::platform added to the bootsupport
  module set - it was absent from the snapshot, so the bootsupport-hosted shell degraded
  to a package-not-available notice on any interpreter, not just msys hosts).

## [0.24.0] - 2026-07-26

- make.tcl bakelist + selective bake (G-121): new informational `bakelist ?kitname ...?`
  subcommand reports the kit outputs configured in src/runtime/mapvfs.config - kit name,
  kit type, runtime (with presence in the runtime store), vfs folder, and the deployed
  state of bin/<kit> vs the src/_build build product (current|stale|absent|nobuild), with
  anomalies (runtime=missing, vfs=missing, rtrev=r<cur><r<max> runtime-materialization
  staleness) in a trailing notes column; name filtering doubles as per-kit detail.
  `bake ?kitname ...?` now bakes and deploys only the named kits - other kits' _build/bin
  artifacts and punkcheck records are untouched, the vfslibs phase narrows to the named
  kits' vfs folders, and an unknown name errors before any build, listing the configured
  names. Bare `bake` is unchanged (all configured kits). Both surfaces (and the argdoc
  kit-name choices) consume one parsed mapping model (::punkboot::lib::mapvfs_*), so the
  config-format work (G-024 toml direction) can swap the reader underneath.

## [0.23.3] - 2026-07-25

- make.tcl colour policy: unix-class Tcl 8.6 terminals auto-detected via the tty channel
  signature - real ttys (isatty-gated channel type) expose the serial/tty options
  (`-mode` etc) that pipes/files lack. Covers linux/WSL and msys2/cygwin-runtime tclsh
  builds (which report platform unix even on windows - e.g the scoop-installed msys2
  `usr/bin/tclsh8.6`, previously colourless-with-transform on its pty). Verified: WSL
  8.6.14 pty mode=tty / piped tcl86-plain; msys2 8.6.12 piped tcl86-plain (interactive
  carries `-mode` per field report). With the 0.23.2 windows console-encoding signature
  this completes dependency-free 8.6 tty detection on all supported hosts; the force
  env vars remain for exotic environments. Also confirmed: 8.6 and 9.x piped help
  output is byte-identical - no 8.6-specific rendering defect existed; all observed
  mangling was the (fixed) utf-16 transform corruption.

## [0.23.2] - 2026-07-25

- make.tcl colour policy defect fix + 8.6 windows console auto-detection: the G-113
  ansistrip transform was being pushed onto Tcl 8.6 windows CONSOLE channels, which
  encode utf-16 ("unicode") - at byte level ESC arrives as 1B 00, so sequences passed
  unstripped and the hold-back logic deferred/dropped real content, seriously mangling
  interactive `punk86`/`punksys src/make.tcl help` layout. The 8.6 windows console is
  now detected by that very encoding signature (only console channels report
  `-encoding unicode`; pipes/files get the system encoding): mode=tty, colour ON, no
  transform - interactive 8.6 windows colour works with no env var and no twapi.
  The transform push is now PER CHANNEL and never wraps a utf-16-class channel (e.g
  `make.tcl ... > file` from an 8.6 console wraps stdout only; the console stderr
  keeps its ANSI). Policy/check lines report the wrapped channels
  (`ansistrip=stdout+stderr|stdout|0`). Chunk-boundary handling verified sound over
  byte-transparent encodings (invariant across chunk sizes 1..4096); 8.6 unix/WSL
  remains plain-by-default pending the probe-ladder goal.

## [0.23.1] - 2026-07-25

- punk-runtime `use` accepts runtime/artifact names with or without the `.exe` suffix
  (both payloads + rewrapped `bin/punk-runtime.cmd`; exact names - including extensionless
  unix artifacts in foreign `-platform` staging folders - still match first).
- make.tcl runtime-materialization staleness guard: `bake`/`bakehouse` emit a
  BUILD-WARNING (recapped at end of run) when a working runtime they wrap is materialized
  from an older revision than an `-r<N>` artifact present beside it (beside-toml revision
  comparison, local-only, no network; pre-family runtimes without metadata stay silent).
  Closes the forgot-to-switch-back hazard behind the 2026-07-25 stale-r1 punk9_beta
  incident - a deliberate `punk-runtime use <old-rN>` test excursion no longer bakes
  stale kits silently.

## [0.23.0] - 2026-07-25

- make.tcl stage-true subcommand rename (G-112): `vfs` becomes `bake` (kit assembly, now
  including the vfslibs phase so a bake cannot ship stale binary libs), `project` is retired
  in favour of `bakehouse` - a consumer umbrella chaining packages + bake that REFUSES
  uncommitted src by default (-dirty-abort defaults ON; the refusal names the granular
  developer flow: packages -> test -> bootsupport/vfscommonupdate -> commit -> bake).
  `project` and `vfs` remain as deprecated aliases printing a one-line notice (kept until
  the 8.6 arc completes). `vfslibs` survives as a standalone narrowing for declaration
  iteration. Help/usage tables regrouped (build & bake / promotion gates / source
  maintenance / deprecated aliases); `workflow` output teaches the two-persona model.
- make.tcl buildsuite surface (G-104): `buildsuite list` discovers the suites under
  src/buildsuites (directories carrying a suite.tcl driver; _build excluded) with
  one-line descriptions; `buildsuite info <name>` renders the suite's configured detail
  (sources.config records, pinned zig + presence, products, doc pointers); `buildsuite
  build <name> ?driver-args?` runs the suite driver forwarding args untouched, streaming
  output and exiting with the driver's status. Suites are self-describing via new
  sources.config records (description/product/doc/zigpin - same manifest the driver
  parses; suite.tcl's pinned-zig default now reads the zigpin record), so copied
  retargeted trees appear automatically. Contract documented in src/buildsuites/README.md
  (new) + AGENTS.md (new). Works under plain tclsh and punk kit executables.

## [0.22.2] - 2026-07-25

- make.tcl colour output is terminal-aware (G-113): piped/redirected runs produce fully
  ESC-free output automatically (ansistrip channel transform on stdout+stderr - covers
  module-side emitters too), interactive terminal runs keep colour (stdout `-winsize` tty
  probe on Tcl 8.7+/9; verified on a real Windows console and a WSL pty). NO_COLOR always
  suppresses colour; PUNK_FORCE_COLOR/FORCE_COLOR re-enable ANSI on piped output; Tcl 8.6
  (no probe) fails OFF. `make.tcl check` reports the active policy; `make.tcl shell` pops
  the transform for the repl. New characterization suite
  `src/tests/shell/testsuites/punkexe/maketclcolour.test`. make.tcl converted to LF endings.

## [0.22.1] - 2026-07-25

- punk::libunknown 0.2.2 (G-120): the package-unknown static-builtin shortcircuit no longer
  leaks its version-probe child interp. On kits whose runtime provides Thread (or any
  required package) as a true static (e.g the tclsfe 9.1b0 runtime behind punk91), an
  unversioned `package require` made the probe's `package vsatisfies` throw (it requires at
  least one requirement argument), skipping the interp delete; the leaked child's std-channel
  registrations then disabled Tcl's std-slot recycling process-wide - after
  `chan close stderr` the slot stayed occupied and the next opened channel got a fileN name,
  so `puts stderr` failed. The probe interp is now auto-named and deleted in a try/finally,
  zero-requirement requires accept any provided static version, and probe errors fall
  through to the normal tm search instead of aborting the require.

## [0.22.0] - 2026-07-25

- punk-runtime `list -remote` identity columns (G-119 follow-on; both payloads + rewrapped
  `bin/punk-runtime.cmd`): rows the server listing does not name carry an untitled middle
  column - `from=<artifact>` for materialized working copies (content verified against the
  source artifact's server sha1, so the Remote cell shows a real status: Same version /
  CONTENT DIFFERS / artifact-not-listed) and `sha1=<name>` for byte-copies of server
  artifacts. The starred active row now changes with every `use` switch instead of sitting
  on the dead-end "(not listed on server)" (which now means genuinely unknown), and the
  `(server default)` row annotation follows artifact identity. Table header/rules widened
  to the three-column layout.

## [0.21.0] - 2026-07-25

- punk-runtime freshness verdict (G-119; both payloads + rewrapped `bin/punk-runtime.cmd`):
  `list -remote` now states the relationship between the ACTIVE runtime and the SERVER
  DEFAULT whenever both are known - explicit `freshness:` AHEAD/BEHIND verdicts naming both
  revisions (BEHIND adds fetch+use update guidance), current-verdict/annotation for matches,
  a different-series incomparable note, and a stated no-revision-basis note for pre-family
  actives with sha1 VALUE-match identification when the active is a byte-copy of a server
  artifact. A no-name `fetch` prints the same verdict as a one-line note when the active
  diverges from the resolved default. Verdict lines are identical across the powershell and
  bash payloads; pinned by a new fixture-punkbin-server suite
  (`src/tests/shell/testsuites/binscripts/runtimecmd_freshness.test` +
  `src/tests/testsupport/httpfixture.tcl`).

## [0.20.1] - 2026-07-25

- Launch subcommand family documentation (G-118 item 10): new module
  `punk::args::moduledoc::punkexe` carries the punk executable launch contract as punk::args
  definitions - ids `(script)::punkexe` (dispatch: package modes, subcommand enumeration,
  default script/shell reclassification) and `(script)::punkexe::tclsh/script/shell/punk/shellspy`.
  The tclsh definition is the full contract: stock argument-form parity (incl. the deliberate
  no `-e` one-liner boundary), `lib:` refusal, piperepl `::tclsh` launch-state array
  (istty/dorepl/evalinput/inputbuffer/reopened) and env gates, plus the intended divergences
  (early `tcl_interactive` link; whole-buffer piped eval with honest exit codes on
  machinery-less runtimes). View via `punk::args::usage {(script)::punkexe}` in any punk shell
  (repl code-interp setup loads the module; punk::repl 0.5.3). Authored G-032-compatible: the
  future launcher help/parse wiring renders from these definitions.
- punk::args 0.12.7: tag-prefixed id namespaces (`(script)::...`, `(autodef)::...`) resolve
  first-class in `update_definitions` - `punk::args::usage` on a registered-but-unloaded
  script-level id no longer errors `no such id`, and the spurious
  `warning: update_definitions received unqualified ns: (script)` stderr noise is gone.
- ARCHITECTURE.md: `tclsh` subcommand bullet added to the boot/launch chain section.

## [0.20.0] - 2026-07-24

- `tclsh` subcommand review fixes (G-118 items 3/4/5/9): no-script forms set `::argv0` to the
  executable path (stock parity - previously leaked the kit boot script path); a missing script
  file gets a clean `script file not found` message with exit 1 instead of a raw errorInfo trace;
  on runtimes without the piperepl patch (or with `TCLSH_PIPEREPL=0`) a console-interactive
  `tclsh` invocation now fails fast with usage guidance instead of blocking in a raw stdin read
  (piped degraded-mode evaluation unchanged; the patch-missing notice is now `note:` and a
  deliberate `TCLSH_PIPEREPL=0` opt-out is quiet); `project_main.tcl` (generated-project kits)
  now carries the full tclsh dispatch in parity with `punk_main.tcl` (script-arg/leading-dash/
  `-encoding`/`lib:`-refusal forms included) and the project-layout `_config` copies are
  refreshed to byte-parity.

## [0.19.0] - 2026-07-23

- `bin/punk-runtime.cmd`: sha1 hashing (fetch verification, `list -remote` comparison) no longer depends on `Get-FileHash` - a script-defined function in Windows PowerShell 5.1 that vanishes on machines with a damaged PSModulePath while compiled cmdlets keep working (observed in the field: "Get-FileHash ... not recognized" under PS 5, making `list -remote` report spurious UPDATE AVAILABLE and fetch verification fail closed). Hashing now uses .NET directly (`Get-PunkFileSha1`), working on any PowerShell edition/state, with a one-time informational note when the damaged condition is detected. The bash payload's multi-tool sha1 probing was already robust and is unchanged.
- `tclsh` subcommand: stock tclsh argument forms - a leading `-` argument no longer errors as a
  script name: all arguments stay in `::argv` and the subcommand proceeds to the repl/stdin
  evaluation, matching stock `tclsh - args...`; `-encoding name fileName` is recognised and the
  script is sourced with the named encoding (incomplete `-encoding` forms fall through to argv,
  as stock). `lib:` scriptlib names are refused with a pointer to the `script` subcommand
  (exit 1) - the tclsh subcommand keeps plain tclsh semantics and loads no punk modules.
  Pinned by `src/tests/shell/testsuites/punkexe/tclshcmd.test` (new suite, also covering the
  piperepl launch-state contract on patched kits).

## [0.18.9] - 2026-07-22

- punk::mod 0.1.2: punk::apps `latest` app-version selection fixed (version-selection audit) - a plain lexical sort picked the LOWEST version as latest (and misorders 1.10 vs 1.9); now `package vcompare`-sorted highest, empty unversioned-main entries excluded. Audit outcome: all other punk-owned `[package versions]` consumers already sort with vcompare; naive sorts remain only in vendored third-party code (tcllib doctools, pattern-1.2.8.tm) - flagged, not edited, per the vendor no-edit policy.

## [0.18.8] - 2026-07-22

- `bin/punk-runtime.cmd` `list -remote`: remote-only rows skip support files (artifact metadata tomls etc now ride in the server's sha1sums beside runtimes - same extension set the local candidate filter excludes); the `(= server default)` active marker also recognizes a materialized working copy via its beside-toml artifact identity (the default names an immutable `-r<N>` artifact while the active is typically its working name - exact-name matching alone would never fire post-family). Verified live against the first published family artifacts.

## [0.18.7] - 2026-07-22

- `bin/punk-runtime.cmd`: deterministic listing order everywhere. Investigation of an ordering discrepancy between launch paths found two edition divergences: the `.ps1` twin runs under the INVOKING powershell (pwsh 7) while the `.cmd` routes via wrap-pinned Windows PowerShell 5, and both Hashtable enumeration order AND culture-sensitive `Sort-Object` collation (NLS vs ICU) differ between those editions. All name orderings now use ordinal comparison (ps1) / `LC_ALL=C` (bash), which agree byte-for-byte across pwsh, powershell and bash. Also corrected `bin/AGENTS.md`: the `.ps1` twin is SELF-MATERIALIZED by the polyglot's batch layer (created when missing, `fc`-compared and re-copied when stale) - the previous "refresh both on re-wrap" manual-copy guidance was stale; twin verified regenerating byte-identical after deletion.

## [0.18.6] - 2026-07-22

- `bin/punk-runtime.cmd` `list -remote`: surfaces the selection state - summary lines show the platform's server default (per `defaults.txt`, best-effort fetch) and the locally active runtime (annotated `(= server default)` on match); the active runtime's row carries the local listing's `*` marker and the default's row (local or remote-only) is annotated `(server default)`. Also fixed a latent bash-payload bug the work exposed: CRLF server sha1sums made locally-present runtimes additionally appear as remote-only rows (`\r`-suffixed name failed the existence test; msys grep strips `\r`, bash `read` does not) - punkbin's maintenance script now writes LF sha1sums too.

## [0.18.5] - 2026-07-22

- `bin/punk-runtime.cmd`: the default runtime a no-name `fetch` retrieves now comes from the artifact server's curated root-level `defaults.txt` (`<platform> <runtimename>` per line - a punkbin release decision, updated there in the same change-set that publishes the artifact it points at; validated by punkbin's maintenance script) instead of values baked into the payloads. Works for any `-platform` (the recommendation is per-platform server data); platforms without a recorded default, and servers without the file, get an actionable message; cached-copy fallback on network failure. The baked per-platform defaults were removed from both payloads - one of them (linux-arm) had already drifted from the server's actual artifact name, which is precisely the failure mode the server-side file eliminates.

## [0.18.4] - 2026-07-22

- Platform matrix: `win32-ix86` (32-bit x86 windows) added to `punk::platform` / `help platforms` - status supported, runtime+lib tiers, buildsuite `candidate` (zig can target it; a buildsuite is undetermined - hosting for available third-party runtimes/libs is supported regardless). `vendorlib_tcl8`/`_tcl9` gained the platform dirs; `punk-runtime` detects genuine 32-bit windows hosts (32-bit shells on 64-bit OS keep the x86_64 default); `punk::platform::normalize` folds hand-typed `i386`/`i486`/`i586`/`i686` to `ix86`.

## [0.18.3] - 2026-07-22

- `bin/punk-runtime.cmd` `list -remote` (powershell payload): the local column now uses the shared runtime-candidate filter, so `active.toml`, artifact metadata tomls, stray `.log` files, `.tmp`/build copies and directories no longer appear as local runtimes in the server comparison (bash payload already filtered).

## [0.18.2] - 2026-07-22

- `bin/punk-runtime.cmd`: new `platforms ?-remote?` action - lists local `bin/runtime/*` platform folders (local platform marked), or the platforms the artifact server serves via the server's new root-level `platforms.txt` discovery manifest (raw-file servers have no directory listing; the manifest is now part of the punkbin layout contract, generated by punkbin's maintenance script, and third-party mirrors using the layout carry the same file). Local presence marked, cached-copy fallback, actionable message for pre-convention servers.

## [0.18.1] - 2026-07-22

- `punk::platform` records + `help platforms` gain a `buildsuite` axis (supported/planned/candidate/none): whether the punkshell zig buildsuites produce runtimes for a platform - deliberately separate from the artifact-tree tiers, since punkbin-structured repos can host runtimes built by any mechanism (a platform can be runtime-tier hosted while buildsuite=none, e.g. netbsd/dragonflybsd). openbsd/netbsd/dragonflybsd records now carry the runtime tier (hosted third-party runtimes possible).

## [0.18.0] - 2026-07-22

- Canonical punkshell platform names (platform-folder synchronization survey; G-105 groundwork). New module `punk::platform` 0.1.0: `platforms` (canonical records with status supported/dormant/recognized and artifact-tree tiers), `normalize` (folds Tcl platform-package version-dependent aliases: amd64->x86_64, aarch64->arm64, macos->macosx, macosx arm->arm64), `local ?-tier lib|runtime?` (runtime tier collapses macOS per-arch to the universal `macosx` runtime-store name). New `help platforms` shell topic (punk 0.2.7) renders the canon with the running interpreter marked and the raw platform-package identifiers for comparison. Boot machinery (punkboot platform_generic snips in punk_main.tcl/project_main.tcl/make.tcl) now normalizes via an inline copy of the same mapping (platform_punk). `src/vendorlib_tcl8`/`_tcl9` platform subfolders synchronized to the canon: freebsd-amd64 renamed freebsd-x86_64, new freebsd-arm64/linux-arm64/macosx-arm64 (+ tracked linux-arm), msys-x86_64 marked dormant pending a utility decision. punk-runtime bash platform prongs emit canonical names (aarch64->linux-arm64 with no fetch default yet, arch-aware freebsd/openbsd/netbsd/dragonflybsd).

## [0.17.9] - 2026-07-22

- `bin/punk-runtime.cmd` (G-103/G-105 groundwork): cross-platform staging surface - `fetch`/`list`/`use` accept `-platform <p>` (punkbin platform-dir names; default = local platform, overridable via `PUNK_RUNTIME_PLATFORM` env) for listing/fetching/selecting runtimes of OTHER platforms (cross-build staging and guest provisioning); foreign fetches require an explicit runtime name; `list` flags runtimes whose metadata `target` disagrees with their folder (`!TARGET-MISMATCH`); `run` stays local-only (rejects `-platform`, ignores the env override). New `help` action (full operator reference: actions, options, env vars, examples) and a proper usage block on no-args invocation. Bash payload's FreeBSD platform dir aligned to punkbin's actual `freebsd-x86_64` (was `freebsd-amd64`). Layout-shipped copy refreshed.

## [0.17.8] - 2026-07-22

- punk::console 0.8.0 (G-106): the powershell console-mode fallback (raw mode on twapi-less windows runtimes) is now dependable and quiet. The persistent pwsh/powershell server starts lazily on first raw enable/disable (loading punk::console no longer spawns a process or prints "twapi not present..." to stderr), is shared process-wide, watches the owning process's pid and exits with it (no orphan pwsh processes), and no longer dies mid-session (the old 20s keepalive with no ping sender, plus a listener defect that killed it after the first message, are both fixed). disableRaw now works via the server on runtimes without `-inputmode` (tcl 8.6 path); enableRaw confirms the flip via the live tcl9 `-inputmode` read before returning. The server script resolves from env `PUNK_PS_CONSOLEMODE_SCRIPT`, the project tree, or an embedded copy in the module - kits and unusual cwds no longer depend on argv0-relative luck. `PUNK_PS_CONSOLEMODE_DEBUG=1` surfaces diagnostics. `scriptlib/utils/pwsh` reconciled to the one canonical server script (experiment variants retired, README added).

## [0.17.7] - 2026-07-22

- `bin/punk-runtime.cmd` (G-103): `list` surfaces per-runtime artifact metadata from `<rootname>.toml` records beside the runtimes (variant, tcl patchlevel, revision, piperepl policy, source artifact of a materialized copy); `use <artifact-r<N>-name>` materializes an immutable `-r<N>` artifact into its working name (metadata toml copied alongside) and selects it; `fetch` of an `-r<N>` artifact also retrieves its metadata toml from punkbin (absence tolerated). Runtime-candidate listing now excludes directories and `.log` files in both payloads. Layout-shipped copy refreshed.

## [0.17.6] - 2026-07-21

- punk::mix::cli 0.5.2: the modpod zip build survives a failing `zipfs mkzip` by falling back to `punk::zip::mkzip` (the established path for zipfs-less interpreters), clearing the partial zip mkzip leaves behind and noting the interpreter + upstream ticket on stderr. Motivation: Tcl 8.7 builds predating core fix `c971e6c7c4` (tkt 7d5f1c13089d463e7796, "zipfs mkzip broken on Windows dotfiles") die with "non-unique path name" on dot-prefixed entries — hit by the templates modpod's layout `.fossil-custom` payloads when building under 8.7a6. Fallback-built modpods verified mounting identically under 9.0.3 and 8.7a6.

## [0.17.5] - 2026-07-21

- punk86 kit restored to the build: mapped in `src/runtime/mapvfs.config` as tclkit-win64-dyn-8617 (Tcl 8.6.17) + punk8win.vfs, replacing the unmapped tclkit8613-era build whose fossilized kit-stamped `punk::args` 0.1.0 broke runtests under the kit ("invalid command name punk::args::define" - a stale registration satisfies an unversioned require, so `package unknown` never scans the modern paths). All kits rebuilt after a `vfscommonupdate`, so shipped kits now carry punkcheck 0.6.0 (G-095 lock protocol) and punk::lib 0.5.1.
- runtests hardening (test tooling, rides this release): the runner's own requires are version-floored (`punk::args 0.12-`, `punk::lib 0.5.1-`, punk::path already floored) - the floors alone resurrect even the old fossilized kit; punkcheck's concurrency suite floors its child-side require so a stale kit host fails loud instead of mysteriously.

## [0.17.4] - 2026-07-21

- punk::lib 0.5.1: `punk::lib::compat::lpop` now matches the 8.7/9 builtin exactly on error paths too - out-of-range errors use the builtin's message and `{TCL VALUE INDEX OUTOFRANGE}` errorcode (previously a nonstandard `tcl_lpop ...` message), and variable-access failures (unset variable, whole-array name, missing array element) reproduce the builtin's message and errorcode naming the caller's variable. Validated by a differential parity matrix against the real builtin, pinned in the punk::lib compat testsuite; no known divergence remains (8.6's index parser already accepts TIP-176 `M+N` arithmetic).
- runtests (test tooling, rides this release): the harness now runs under a native Tcl 8.6 tclsh - guarded `punk::lib` compat loads (lpop family) at both boot levels, no-ops on Tcl 9 (fast `-discover-only` boot unchanged). Full-suite mode parity re-verified on Tcl 9.0.3 (`PARITY: ok`, baseline exec-14.3 only).

## [0.17.3] - 2026-07-21

- punkcheck 0.6.0 (G-095, goal achieved): concurrent writers to one `.punkcheck` file can no longer corrupt it or silently erase each other's records, and readers never see torn content. Saves are write-temp-then-atomic-rename (with reader-side retry for the Windows replace-window open denial); whole installer events hold an advisory sibling `.punkcheck.lock` (new `punkcheck::lock` namespace - holder pid/host/installer/timestamp in the lockfile, backoff-retry acquisition to a timeout that errors naming the holder, stale-break on provably-dead holder pid or age, in-process reference-counted re-acquisition; env overrides `PUNKCHECK_LOCK_TIMEOUT`/`PUNKCHECK_LOCK_STALEAGE`); deferred flushes merge own records into freshly-loaded file state; an mtime/size tripwire warns on protocol-bypassing writers; the duplicate-INSTALLER recovery raises a clean actionable error instead of prompting when stdin is non-interactive. Verified by a new concurrency suite (real child installer processes) on Windows Tcl 9.0.3 + 8.6.17 and linux (WSL tclsh 8.6.14).

## [0.17.2] - 2026-07-21

- punkcheck 0.5.0 (G-094, goal achieved): the installtrack/installevent object layer is the sole implementation of the .punkcheck record lifecycle, with a per-event persistence policy (`start_event ?-persistence eager|deferred?`; deferred updates flush via `$installer flush`, QUERY is never persisted) and a single save chokepoint (installtrack `save_working_recordset` - the G-095 attachment point). `punkcheck::install` is now a tree-walking OO consumer in deferred mode and its records carry the full field set including `-targets_cksums`. The zero-caller legacy proc pipeline (`start_installer_event`, `installfile_begin/started_install/finished_install/skipped_install`) is retired to error shims pointing at `punkcheck::installfile_help` (rewritten for the OO lifecycle). Sequential events on one installtrack no longer clobber prior targetset writes (ending an event now clears the active slot), and several latent bugs are fixed (installer-record `-name` trailing space on flush, event-history timestamp rewrites on reconstruction, `get_recordlist`, `targetset_dict`). Build skip/copy decisions are unchanged against pre-refactor `.punkcheck` state.

## [0.17.1] - 2026-07-20

- punk::path 0.4.0 (G-093, goal achieved): pathglob syntax gains `***` - as a whole segment it matches ZERO or more path segments, so `X/***` addresses X and everything below it (`***/f.txt` matches at any depth including the root). The directory-pattern lattice (`X` in-dir, `X/*` one below, `X/**` strictly below, `X/***` base-inclusive) is now separable in both directions: an exact or single-segment-glob include match no longer drags the whole subtree in, and non-spanning exclude matches drop only the matched directory's files instead of pruning (only `**`/`***` tails prune - the documented `**/_aside` vs `**/_aside/**` distinction is now real). The legacy bare-`*` match-everything include alias is removed, and the default-nocase drive-letter exception survives glued-group patterns like `c:/***`.
- runtests.tcl targeting (test tooling, rides this release): `-include-paths`/`-exclude-paths` pass through to treefilenames - one pathglob semantics at every layer; subtree runs are `-include-paths <path>/***`. New `-exclude-paths` (exclusion wins) and subsecond `-discover-only` (two-stage boot); `-serial-paths` reconciled to the same directory-of-file reading; a received include pattern matching nothing earns a stderr advisory; the json report no longer breaks when a failed test's result value carries surrounding whitespace.

## [0.17.0] - 2026-07-19

- G-087 stages 4+5 (goal achieved) - overlay layout composition: `dev project.new` now materializes a layout's overlay chain before copying - custom-tier files override same-name vendor-tier files (folders merge recursively), zero-byte `<item>.anti` marker files delete lower-layer items, and a `<newname>@<base>` folder presents as a derived layout starting from its base layout's chain (new `punk::mix::commandset::layout::lib::layout_materialize` staging primitive; single-folder layouts such as punk.project are used in place, unchanged). The sample fixtures demonstrate all three (`punk.othersample` is a working derived layout with a `.anti` deletion). `layout.files`/`layout.templatefiles` list the effective merged view.
- Generated projects are version-seeded at creation (each only-if-missing): `punkproject.toml` (`[project]` name + version 0.1.0), a matching `CHANGELOG.md` `## [0.1.0]` seed, and `punkorigin.toml` - a generation provenance baseline (layout name + resolved store target, punk::mix::templates version, punkshell version when derivable, origin path) as the interim carrier for the planned G-027 pull updates.
- project.new fixes/robustness: module names (including the projectname-derived default) are validated before any folder or fossil-repo creation (a dashed project name previously failed late with the tree and repo already created; use -modules {} for dashed names wanting no initial module); generation now also works programmatically/script-mode (explicit shellrun dependency, fully-qualified punk::mix::util calls). The retired unimplemented `src/PROJECT_LAYOUTS_*` marker folder and its install/sync exclusions are gone; the `_project` pseudo-vendor store level is retired for layouts (custom/<vendor> is the one override level).

## [0.16.0] - 2026-07-19

- G-087 stage 3 - thin project layouts with generation-time injection: `dev project.new` now injects bootsupport into the generated project from the generating shell's current trees (modules per the layout's `include_modules.config` manifests plus adjacent support files; bootsupport lib trees from a shell project; standalone kits resolve modules from their own module paths and emit a notice for the manifest-less lib set). The ~40MB of stored bootsupport/vfs module snapshots left `src/project_layouts` (fat `custom/_project` fragments and their `_overrides` refs retired; vendor layouts thinned to structure/scripts/docs/manifests).
- A bare punk kit outside any project can now list and generate the `punk.project` layout: `punk::mix::templates` 0.2.0 ships the thin payload inside the modpod (build-synced from the store), `punk::cap::handlers::templates` 0.3.0 resolves module-pathtype layout refs against the module's own store, and the `shellproject_multivendor` declaration was fixed to `src/decktemplates` (shell-project layouts/templates were invisible when cwd was outside any project). `project.new -layout` is now prefix-matched against the live layout set instead of a load-time snapshot.
- make.tcl: former custom/_project layout bootsupport-sync retired; new thin-layout sync keeps layout `make.tcl`/`build.tcl`/manifest copies and the modpod payload current (workflow DIAGRAM 1b); bootsupport copy onto a self-mounted modpod now converges when content is already identical.
- punkcheck 0.4.0: `punkcheck::install` event headers no longer record literal `$eventid`/`$rel_source`/`$rel_target`/`$config` strings (latent since inception; collided when mixing install + installtrack writers in one db), and sources sharing no common root with the punkcheck folder (e.g `//zipfs:/` module-mounted layouts) are recorded as independent absolute paths instead of erroring. punk::mix::base 0.1.2: `cksum_path` no longer hangs forever on vfs-mounted files (tcllib `-file` digest modes rely on fileevents, which zipfs channels never deliver - data-mode digests are used for non-native filesystems).
- bootsupport `include_modules.config` completed with previously-recordless-but-required entries: `cmdline`, `struct::set`, `textutil`, `textutil::expander`, `textutil::patch` (generated projects' make.tcl now boots under a plain tclsh).

## [0.15.1] - 2026-07-19

- punk::ns 0.9.1: `get_ns_dicts` per-namespace `packagetails`/`packageprefixes` fix - the two lists were initialized once before the matched-namespaces loop, so with multiple matched namespaces (path globs, trailing `**`) every dict after the first carried the union of all earlier namespaces' entries, mis-marking non-package children in `nslist`/`n//` underline styling and in the returned data; a child with several deeper packages below it was also recorded once per matching package (now once). Single-namespace queries (the nslist.test tier A pins) were unaffected; the multi-namespace reset is now pinned in nslist.test. Documentation-only PUNKARGS added for `get_ns_dicts` (documents the returned per-namespace dict keys and `-allbelow`/`-nspathcommands`) and `nslist_dict`.
- New developer/agent namespace-discovery script `scriptlib/developer/nslist.tcl` backing the new cross-framework `tcl-nslist` agent skill (canonical `.agents/skills/`, synced copy in `.claude/skills/`): the discovery counterpart of `whatis.tcl` - lists what a namespace or package actually exposes at runtime via `punk::ns::get_ns_dicts` (child namespaces with loadable-package markers, commands classified by kind, punk::args-documented commands, `-synopsis` whole-namespace usage lines, `-pathcommands` namespace-path resolution), auto-`package require`s unloaded packages and follows ensemble commands to their implementation namespace. The `tcl-whatis` skill/script now cross-reference it (whatis redirects with a hint when the queried name is a namespace rather than a command).

## [0.15.0] - 2026-07-18

- shellrun 0.2.0: `runx` gains `-teelog` - a live tee of the captured stdout/stderr to the shellfilter::log mechanism (shellthread worker threads; `-tag`/`-syslog host:port`/`-file`/`-raw` settings; log sources `<tag>stdout`/`<tag>stderr`) while the captured/returned data stays unchanged. Implemented as write traces on the capture variables rather than additional channel transforms: probing showed the unaudited float-placement logic can starve a tee_to_log float combined with the float-locked capture junction, and shellthread's parked-worker adoption does not re-apply settings (logs silently misrouted to a previous target) - both findings recorded for a future shellfilter/shellthread audit. Tag workers are left open for reuse across calls; a conflicting reconfiguration errors explicitly.
- runtests.tcl UDP watch mode (test tooling, rides this release): env `PUNK_TEST_UDPTEE=<port>|<host>:<port>` streams runner lifecycle events (tag `runtests`, tcl-list `RUNTESTS-EVENT` messages) and test output (tags `teststdout`/`teststderr`) to a local UDP listener - live per output chunk in single-process mode via `runx -teelog`, per completed file in multi-process mode. In watch mode the singleproc loop also terminates the shellfilter::log worker threads spawned inside each per-file testinterp (unwatched runs keep the pre-existing small teehandle-worker leak rather than pay ~0.6s/file terminate handshakes - noted for the shellfilter/shellthread audit). Full-suite result parity with the tee on and off verified via `scriptlib/developer/runtests_parity.tcl`.
- punk::config 0.2.0: `PUNK_TEST_UDPTEE` registered in the known punk env vars table; module converted from manually-versioned `config-0.1.tm` to standard buildversion tracking (magic-version source filename + `config-buildversion.txt` - punk::config was never in the documented manual-versioning exception list).

## [0.14.0] - 2026-07-18

- `lib:` scriptlib scripts may now be extensionless: an extensionless (or unknown-extension) `lib:<name>` call also matches a file named exactly `<name>` when its first lines identify it as tcl - a leading `# tcl` comment, a shebang naming tclsh/tclkit/wish, or the portable sh-trampoline (`#!/bin/sh` with a commented `exec tclsh` continuation). When both `<name>` and `<name>.tcl` exist, the spelling matching the call wins, checked per scriptlib location so kit-internal precedence is unaffected. Implemented in punk::path 0.3.0 (`scriptlib_resolve` + new internal `scriptfile_is_tcl`; result dict gains `scripttype`/`notes`, not-found errors list skip notes); `.kit` is now a known extension so `lib:<name>.kit` resolves as itself (previously it wrongly searched `<name>.kit.tcl`, making the shell subcommand's kit dispatch unreachable - now restored). The `shell` subcommand's `lib:` handling (app-punkshell, app_shellrun) is refactored onto the shared resolver, fixing two silent-failure bugs in its former inline copy (no message when a scriptlib dir existed but lacked the script; wrong kit-internal path in the error listing). Test fixtures + coverage added under `scriptlib/_punktest/` and `src/tests/shell/testsuites/punkexe/scriptexec.test`.

## [0.13.0] - 2026-07-18

- punk::ns 0.9.0: cmdhelp (`i`) gains `-return text` - plain-text argument documentation for machine/LLM consumption: no ANSI codes, no table layout, and the argument section of every command form rendered in a single output under a common header (`i -return text after` shows all six forms), with G-041 form selection still honoured when argument words are supplied. No advisory-parse error preamble (use `-return dict` for parse status).
- punk::args 0.12.6: usage/error display refinements (user direction; the string-form layout is explicitly not frozen - no external consumers yet). The error message and its per-form match detail now render exactly once, BELOW the usage display (previously above and below). The string renderer's argument rows are single-line `name  TYPE:type(constraints) ?DEFAULT:'val'? ?MULTI:yes?` - type constraints parenthesize onto the base type instead of line-breaking mid-row, and empty DEFAULT/MULTI labels and whitespace-only help blocks are omitted. Choices render as markdown-subset bullets (`` - `choice` - label``) with label continuations at a fixed deeper indent, so choice names are machine-distinguishable from label text without ANSI (structural contract: column-0 lines bearing TYPE: are argument rows, indented lines are documentation, backticked bullets are choices).
- New developer/agent introspection script `scriptlib/developer/whatis.tcl` (backing the checked-in cross-framework `tcl-whatis` agent skill in `.agents/skills/` with a synced copy in `.claude/skills/`): reports a command's resolved name, subcommand-chain resolution (e.g `whatis string is`), kind, providing package + loaded version, origin file, proc definition file:line, punk::args synopsis, and optionally the runtime body with file-relative line numbers (`-body`) or the full plain-text argdoc (`-doc`).

## [0.12.42] - 2026-07-17

- make.tcl: new informational subcommand `workflow` - prints an embedded plain-ASCII overview of the build/release data flow (release-ready command sequence, source->built-packages->bootsupport/kit-vfs/kit folder diagrams, keyed notes on version stamping, punkcheck provenance, the staleness gate, -confirm and locked-exe deploys). punk::args-documented like the other subcommands (tabled help + PUNKBOOT_PLAIN fallback covered); exempt from the bootsupport staleness gate alongside check/projectversion. The text lives in ::punkboot::workflow_text so it travels with make.tcl into generated-project layouts; update contract for keeping it current documented in src/AGENTS.md Work Guidance and in the output's MAINTENANCE key.

## [0.12.41] - 2026-07-17

- punk::args 0.12.4: resolve's undefined-id miss path is now collision-aware - an inline definition (`parse ... withdef`, direct resolve) whose @id is already registered with different content quietly purges the id's existing cache entries before registering (matching define's re-creation behaviour), instead of repointing the id while stale cache entries lingered (previously a subsequent define with the original text was a silent no-op, leaving withid lookups stuck on the inline definition). The disabled blanket 'called with undefined id' stderr warning removed (it flagged supported usage). Known residual: the miss path still records no -defspace for @dynamic definitions first seen via resolve. New define.test pin (define_resolve_id_shadow, characterized pre-fix).

## [0.12.40] - 2026-07-17

- G-087 stage 2: layout refs migrated to fauxlink files. fauxlink 0.2.0 (vendored from upstream) implements fauxlink::link_as - the encoder counterpart of resolve with a round-trip guarantee - and the migration script used it to generate every new ref name. punk::cap::handlers::templates 0.2.0 resolves layout refs via the fauxlink module (<alias>#<encodedtarget>.fauxlink|.fxlnk, target relative to src/project_layouts); the bespoke <alias>@<target>.ref grammar is retired (.ref files no longer recognised). Layout refs must carry a non-empty alias - alias-less (leading-#) refs are warned about and skipped (punkcheck excludes leading-# filenames). All 7 live refs renamed (alias-less ones gained their target-tail aliases); visible layout names unchanged. punk::mix::templates 0.1.5 carries the migrated module-shipped ref. Verified end-to-end: dev project.new resolves the default punk.project layout through fauxlink refs and materializes a full project.

## [0.12.39] - 2026-07-17

- fauxlink 0.1.2 vendored from upstream (c:/repo/jn/tclmodules/fauxlink, now under fossil version control with a 62-test suite): punk::args documentation for the whole developer-facing API including a (package)fauxlink id (inert until G-075 surfaces package ids); fixes for decode_unicode_escapes dropping the first character when a hexless %U was present, single-comment (3-section) links skipping validation/decoding, and the link_as stub silently no-oping (now raises 'not yet implemented'). Consumers (fcat redirect-follow, make.tcl merge_over/startup-script check, kit boot fallback) unchanged.
- tomlish 1.1.12 + test module vendorupdate catch-up; bootsupport snapshot refreshed to current module versions (incl. the 0.12.38 hygiene bumps) with superseded copies pruned.

## [0.12.38] - 2026-07-17

- G-087 hygiene stage (layout system): `dev project.new`'s default -layout now resolves the exact name `punk.project` (layout ref renamed to carry the alias; previously tcl::prefix ambiguity between punk.project-0.1 and punk.project-0.1_overrides). Junk/test layout refs no longer selectable (test2, a duplicate sample-0.1, and the a.txt-only punk.shell-0.1 whose ref name lied about its content). No-handler warning paths in punk::mix::commandset::{layout 0.1.1, project 0.2.2, module 0.1.1, scriptwrap 0.1.1} and punk::mix::base 0.1.1 fixed ('put' -> 'puts', and return empty dict instead of erroring on an unset variable); punk::cap::handlers::templates 0.1.1 fixes the non-braced expr that errored on any absolute-pathtype template provider registration, and the duplicate-registration guard that errored ('key not known in dictionary') for any second provider package registering the same capability. src/project_layouts/README.md rewritten to describe the implemented punk.templates+layout_refs mechanism including the ref filename grammar; dead modpod templates/layouts payload (punk::mix::templates 0.1.4) and make.tcl old_layout_update_list residue removed.

## [0.12.37] - 2026-07-15

- tablelist vendored 6.22 -> 7.11, now as a pkgIndex-based vendorlib package (src/vendorlib/tablelist7.11) instead of .tm vendormodules, propagated to _vfscommon.vfs; _vfscommon built punk modules refreshed (ansi 0.1.3, args 0.12.3, lib 0.5.0, ns 0.8.0, repl 0.5.2).
- punk9win.vfs split into per-Tk-version kits: punk9wintk90b2.vfs (Tk 9.0b2-era dll, parked) and punk9wintk903.vfs (newer Tk 9.0.3 dll, adds apave 4.4.10, drops bundled Thread modules in favour of the tclsfe runtime's static Thread; now the punk91/tclsfe-x64 build source). mapvfs.config and vendorlib_vfs.toml retargeted.
- make.tcl kit builds now emit a noticeable BUILD-WARNING (recapped at end of run) when a .vfs folder being built has no root main.tcl and no fauxlink resolving to main.tcl with an existing .tcl target; the build still proceeds.

## [0.12.36] - 2026-07-14

- punk::ns 0.8.0: removed the leftover global command aliases `nscommands1` and `nscommands2` (user direction; flagged as removal candidates in the 0.7.0 hygiene pass). Both were undocumented pipeline-based development iterations superseded by the `nscommands` proc; `nscommands1` was broken (referenced nonexistent commands) and errored on first use. Patch-level project bump despite command removal: neither was part of the documented user-facing shell contract.

## [0.12.35] - 2026-07-14

- punk::ansi 0.1.3 (documentation-only): grepstr's -return summarydict help corrected - the summary dict's linemap key is always present (the old text claimed it appeared only with -n; the always-present behaviour is pinned in ansi/grepstr.test). This was the doc/behaviour reconciliation deferred from the grepstr characterization wave to the punk::ns hygiene pass.

## [0.12.34] - 2026-07-14

- punk::ns 0.7.0/0.7.1 comment/documentation hygiene pass + pinned-wart fixes: ~1200 lines of dead code removed (the fully commented-out arginfo predecessor of cmdhelp, the never-exported name-primitive twins nsparts1/nsprefix1/nsprefix_orig/nstail1/nstail_orig whose divergence was pinned before removal, obsolete nsglob_as_re1, empty stubs, misc commented remnants); stale docs corrected (nsprefix :::a comment, nsglob_as_re 'should be fixed' note, nstree_list empty PUNKARGS help, cmdhelp/corp help typos). Shell-visible fixes (0.7.1, pins flipped): bare `nslist` (no -types) now displays all member types instead of erroring with a malformed literal-$known_types message; nsjoinall's absolute-path error is now prefixed 'nsjoinall:' (was 'nsjoin:'); a development trace puts removed from nscommands' weird-namespace path. Flagged unchanged: get_nslist's apparently-unreachable alias display branches, superseded nscommands1/nscommands2 aliases, cmdtrace output noise (deferred to proposed G-085).

## [0.12.33] - 2026-07-14

- punk::repl 0.5.2: fixed the codethread seeding template creating a stray namespace named after the ::punk::nav::ns::ns_current variable in every code interp (a phantom 'ns_current' child visible when browsing the nav namespace with n/). Subshell namespace carry-over semantics unchanged. Plus n/ n// n/// display-and-navigation characterization groundwork for the planned punk-tables/width-responsive rework: new punk/ns nslist.test (classification buckets as machine contract, per-element marking pins, REWORK-flagged layout pins, bare -types quirk), new punk/nav/ns navns.test (navigation state machine + ns_current contract), new punk/repl nscurrent.test (real-codethread inscope evaluation, retention, navigation, auto-create, seeding-fix guard).

## [0.12.32] - 2026-07-14

- punk::lib 0.5.0: new punk::lib::check::has_tclbug_nestedswitch_tracelines (+ pure signature classifier) - live behavioural probe of the upstream nested-switch trace-line mis-attribution (tcl tktview 5d5b1052280c976ea3d4), surfaced in the 'help tcl' warning report (level medium) alongside the existing has_tclbug_* checks. No version gate: the probe matches the characterized wrong/right arm signature, so it flips automatically on a fixed Tcl runtime; a fixed-canary test in ns/cmdtrace.test then fails first, triggering the documented followup (flip GAP pins, update cmdtrace argdoc caveat, record fixed-in version).

## [0.12.31] - 2026-07-14

- punk::ns 0.6.0: cmdtrace gains -pause 0 for non-interactive use (bypasses the enter-key pause so the marked-up traced-body report can be captured by scripts/tests). The nested-switch line-mismark was characterized with a punk-free minimal repro (identical on Tcl 8.6.17/8.7a6/9.0.3) confirming the bug upstream (core.tcl-lang.org tktview 5d5b1052280c976ea3d4): arm bodies whose split-list index lands on a literal word of the switch command report container-relative lines; new ns/cmdtrace.test pins correct flat/2-word-form marking and GAP-pins the upstream mismarks; cmdtrace argdoc caveat updated with the ticket and pattern.

## [0.12.30] - 2026-07-14

- punk::args 0.12.3: fixed the stringstartswith tail-clause reservation bug flagged in the 0.12.2 comment review - the multi-member tail-clause walk in get_dict_can_assign_value compared the prefix against the type string instead of the candidate value, so an optional trailing clause like {literal(with) stringstartswith(v)} lost its matching words to a preceding -multiple argument, and certain prefix spellings caused spurious "Received more values than can be assigned" errors. Characterized pre-fix, pin flipped (allocation.test), guards added for the literal-only and no-match cases.

## [0.12.29] - 2026-07-14

- punk::args 0.12.2 comment/documentation hygiene pass (no behaviour change): ~1100 lines of superseded commented-out code removed (old check_clausecolumn, abandoned _check_clausecolumn2 experiment with its error-collection idea preserved as a note for G-072-era rework, pre-parsekey ordering loop, disabled switch arm, debug remnants); MAINTENANCE loop-parity comments upgraded to a dated refactor assessment (shared-helper extraction deferred to ride G-084 under G-046 hot-path constraints); user-visible doc fixes (get_dict return-shape and directive lists corrected, false solo-flag/last-value-only claims removed, two broken doc examples fixed); goal cross-refs (G-053, G-084) annotated at the relevant code sites; a suspected latent bug in the multi-member clause stringstartswith( arm flagged in a comment (matches the type string instead of the value) without changing behaviour.

## [0.12.28] - 2026-07-14

- G-082 (achieved): punk::args 0.12.1 single-form parse error selection - a word rejected by an optional argument's allocation screen (restricted choices, or the basic int/double/bool/number/dict type screen) that then overflows the argument list now reports the specific rejection (choiceviolation naming the word and allowed choices, or typemismatch naming the type, with -badarg/-badval) instead of the generic "Received more values than can be assigned" overflow - restoring the 0.5.0-era pointedness for optional choice-restricted dispatch shapes (shell-visible in usage errors from punk::args-parsed commands). Genuinely-surplus words still report toomanyarguments; multiform selection unchanged. New errorselection.test (pre-fix behaviour characterized first, pins flipped deliberately); the make.tcl `-optional 0` authoring workaround is no longer needed on fixed versions (src/modules/AGENTS.md guideline re-documented; make.tcl keeps it while running against the bootsupport 0.12.0 snapshot).
- Also: punk::args goals G-083 (argument relations: strict mutual exclusivity, parsekey-group integrity) and G-084 (-parsekey completeness) proposed, with parsekey characterization tests (mash/prefix/shared-default/collision pins + dev known-bug pins) landed alongside.

## [0.12.27] - 2026-07-14

- make.tcl shell under a built punk executable: fixed missing prompts and the raw-mode 'invalid command name struct::set' error. Both stemmed from the kit's script-mode boot pre-loading the punk stack: (1) app-punkscript correctly forces ::tcl_interactive 0 for script runs, and since punk::repl was already loaded its load-time interactivity probe never re-ran - the shell branch now recomputes ::tcl_interactive via punk::repl::is_interactive against the real input channel before repl::start; (2) the accelerator-reload block (package forget + command destroy of sha1/md5/struct::*) relied on later package-require chains to restore the packages, which never re-fire when consumers (punk::lib, punk::du, flagfilter etc) were pre-loaded at kit boot - the block now records what was actually loaded and re-requires it immediately (verified: repl-thread interp state now identical under tclsh and punk91). src/AGENTS.md gains a "pre-loaded interp, not a virgin one" note covering the class: interp surgery and load-time-state assumptions in make.tcl must restore deliberately and be tested under both tclsh and a punk exe (includes the kit-vs-bootsupport module provenance-mixing caveat).

## [0.12.26] - 2026-07-14

- make.tcl under a built punk executable: kit builds gain a self-build guard - the kit whose deployed executable is running the build is skipped with a clear warning (previously the pre-deploy process sweep would taskkill the build's own process mid-run with -k, since it matched processes by executable name only; without -k the polite kill's failure was the only thing saving the run). The sweep now also excludes the build's own pid unconditionally (covers a same-named copy driving the build from a non-deploy path). Informational/update subcommands were already safe (they exit before the kit machinery); building *other* kits from a punk exe remains supported. Also quietened the 'punk::libunknown::init already done - unnecessary call?' stderr diagnostic when a punk exe runs make.tcl: make.tcl now leaves an already-active libunknown in place instead of re-sourcing the bootsupport copy over it and tripping init's rename guard. src/AGENTS.md build guidance updated.

## [0.12.25] - 2026-07-14

- G-030 follow-up (user review): make.tcl's constructed punk::args definitions adopt the G-045 authoring mechanisms - @cmd -help bodies converted to block-form values with a bare @normalize per definition, fixing the constructed-definition indent leak (first Description line rendered less indented than continuations; the braced option fragments never leaked because literal block-shaped parts get file-style normalization). Added a define-time capability probe (get_spec on representative ids) so a stale punk::args that accepts the text but cannot resolve a mechanism degrades to the plain fallback immediately. AGENTS infrastructure for future documentation work: new "punk::args definition authoring ergonomics" subsection in src/modules/AGENTS.md (constructed-def @normalize + block form, -unindentedfields, -& record continuation, container quoting rules, explicit @form -synopsis for script-level ids, the optional-choice error-quality workaround), cross-referenced from src/AGENTS.md with ::punkboot::argdoc named as the in-tree exemplar. Layout make.tcl copies followed via the punkcheck channel.

## [0.12.24] - 2026-07-14

- G-030 (achieved): make.tcl dogfoods punk::args. All 15 subcommands are punk::args-declared ((script)::punkboot definitions with single-source summaries, shared option fragments and explicit @form -synopsis lines): bare `make.tcl`/`-help` render tabled usage with grouped subcommand summaries, `make.tcl help ?subcommand?` / `<subcommand> -help` show per-subcommand definitions, invalid flags and unknown subcommands raise punk::args usage errors (exit 1) with unambiguous-prefix resolution. New `-confirm 0|1` prompt policy on every confirmation point (bootsupport minor-staleness gate, vfscommonupdate REPLACE, kit-type mismatch): default prompts only on an interactive terminal and aborts fast with guidance on non-interactive stdin (the piped-`y` agent workflow is retired - use `-confirm 0`; the old vfscommonupdate decline half-abort now exits 1). punk::args joined the bootstrap-tracked staleness set (parsing contract only - the punk::ansi/textblock rendering stack degrades instead: minimal errorstyle/plain help when unhealthy, and full scan-based fallback dispatch when punk::args itself is unavailable; PUNKBOOT_PLAIN=1 forces the degraded mode). Bootsupport snapshot refreshed (args 0.5.0->0.12.0 plus punk/lib/ns/repl/tclcore moduledoc) and layout make.tcl/bootsupport copies followed via the punkcheck channel. Docs: src/AGENTS.md, src/bootsupport/AGENTS.md, src/modules/AGENTS.md, src/tests/shell/AGENTS.md. Dogfooding finding recorded in the goal file: punk::args 0.12.0 reports a generic overflow error where 0.5.0 reported a pointed choiceviolation for optional-choice mismatches (worked around with required leaders; candidate punk::args error-selection improvement).
- Also since 0.12.23 (landed before this version's bump, entry added retrospectively 2026-07-14): punk::args -> doctools and punk-native HTML/Markdown converter prototypes under src/scriptapps/tools (punkargs_to_doctools.tcl, punkargs_punknative.tcl); vendored punkdoc-mono web font (OFL subset of Cascadia Mono 2404.23) with reproducible fontprep builds and -assets embed|link|none in the punknative generator; punknative generator refinements (nav index, per-form arg tables, terminal-black backdrop, integer row pitch, settled ansi-art html rendering recipe). Goals G-078..G-081 drafted (punk::ansi->html api, default-colour semantics, cell-grid rendering mode, argdoc build pipeline).

## [0.12.23] - 2026-07-13

- Restored `dev doc.validate`: tcllib dtplite wrapped as the generated polyglot bin/dtplite.cmd (multishell scriptset: src/scriptapps/dtplite.tcl + dtplite_wrap.toml, LF-only payload; falls back to the project-vendored tcllib under src/vendorlib_tcl9/<arch> when the invoking tclsh lacks the dtplite package). Previously the repl unknown-handler's auto_execok found no dtplite executable ("invalid command name dtplite" - only the legacy misnamed bin/dtplite_run.bat existed). New src/tests/shell/testsuites/binscripts/dtplite.test pins the artifact (LF-only, checkfile-clean, byte-identical round-trip re-wrap) and exercises usage-error/validate/html usecases. Verified: 'dev doc.validate' validates all 64 src/doc .man files clean. The doctools pipeline remains interim - punk::args is intended to become the documentation source of truth. (Entry added retrospectively 2026-07-14; the version bump landed in the original commit without one.)

## [0.12.22] - 2026-07-13

- G-074 (achieved): punk::args 0.12.0 adds punk::args::formcheck - on-demand multiform ambiguity analysis. It reports the form pairs of a definition that some argument list could cleanly match simultaneously, confirming every candidate with a real parse of a synthetic witness arglist against both forms - so fully discriminated form pairs can never false-alarm, and each reported witness is a genuine multipleformmatches input. Findings classify as type-weakness (a literal/choice discriminator aligned with a permissive non-validating type such as expr/any/string - the 'lseq 1 count 5' class) vs structural (the forms genuinely share an argument shape - the 'after cancel id|script' class). The new @form -overlapallowed <formname-list> key sanctions a known/documented overlap for formcheck reporting only (parse behaviour untouched; unknown form names rejected at definition resolve); tclcore moduledoc 0.3.4 adopts it for the after cancel pair, leaving ::after with zero unsanctioned findings while ::lseq's expr-typed-end findings stay visible as actionable. New testsuite args/formcheck.test.

## [0.12.21] - 2026-07-13

- G-076 (new goal, active): 'help tcl' now warns about the tcl9 dead-console defect (upstream ticket f10d91c2d3, root-caused in G-039: a dead console is never delivered to the script as a fileevent while the core's console reader thread busy-loops) via punk::lib 0.4.3's has_tclbug_console_deadspin — version-based detection through the pure classifier tclbug_console_deadspin_applies, gated by check::tclbug_console_deadspin_fixed_in (empty until a released Tcl contains the upstream fix verified by re-running the G-039 kill procedure). punk::repl 0.5.1 arms the dead-console watchdog only when that same check reports the runtime affected, so recording the fixed release once silences the warning and stops arming the watchdog together. Buginfo dicts gain a mitigated/mitigation axis orthogonal to level (punk 0.2.6 renders it): the deadspin warning keeps severity major but displays "(mitigated)" in subdued grey with the watchdog's scope described when punk::repl >= 0.5.0 is available to the runtime; non-repl console reads remain exposed and unmitigated warnings render unchanged.

## [0.12.20] - 2026-07-13

- tclcore moduledoc 0.3.3: removed the unnecessary @dynamic tags from the split, array and join definitions - no bad-@dynamic warnings remain anywhere in the module, and the three definitions now cache their display expansion like the rest instead of re-expanding per render.

## [0.12.19] - 2026-07-13

- punk::args 0.11.2: the "bad @dynamic tag" warning now emits once per definition instead of on every resolve ('i join' emitted it 4x, user-reported) - the underlying inefficiency fixed: parse-inert @dynamic definitions now cache their round-1 resolution like legitimately dynamic ones instead of redoing the full text substitution on every resolve. The warning wording now notes @dynamic still forces display-field re-expansion per render for such definitions. tclcore moduledoc 0.3.2: fixed the malformed ::tcl_startOfNextWord definition (unescaped quotes in its embedded man-page example prevented it resolving at all - 'i tcl_startOfNextWord' now works), found by sweeping all 462 registered ids; the sweep identified ::split, ::array and ::join as the parse-inert @dynamic carriers.

## [0.12.18] - 2026-07-13

- tclcore moduledoc 0.3.1 (authoring-style only): the after id-shape harvest is consumed via tstr placeholders with the variable in the argdoc namespace (the defspace), replacing the interim string-map token - the module now showcases the neater placeholder style documented in punk::args 0.11.1, with string map reserved for when build-time substitution is genuinely necessary. Behaviour unchanged.

## [0.12.17] - 2026-07-13

- punk::args 0.11.1 (documentation-only): the define help now documents the two definition-registration styles (direct define vs lazy PUNKARGS/register::NAMESPACES registration used by module templates and moduledocs) and the tstr interpolation rules - display-field deferral, parse-field expansion at first resolve, the defspace rule (argdoc child namespace wins when present), the silent-literal fallback for unresolvable placeholders, and the safe patterns for load-time-computed values.

## [0.12.16] - 2026-07-13

- tclcore moduledoc 0.3.0: 'after cancel <word>' and 'after info <id>' now discriminate by the after-id shape, harvested from the running interpreter at load time (user-directed). 'i after cancel someid' resolves cleanly to the cancel-script form exactly as real Tcl treats it (silent script-match no-op), and only genuinely id-shaped words ('after#N') report the cancelid/cancelscript ambiguity - the junction real Tcl itself resolves by id liveness at runtime. Parity pins added (tclcoreparity.test) including the accepted dead-id over-acceptance boundary.

## [0.12.15] - 2026-07-13

- G-041 doc surface: the help system presents the command form matching the supplied words - 'i after cancel <id>' presents the cancel form's argument table (candidates are named when no form or several forms match, e.g 'after cancel someid' reports the genuine cancelid/cancelscript ambiguity of the documented forms), and 's after cancel someid' underlines the matching synopsis line(s) (punk::ns 0.5.0). punk::args 0.11.0: the documented @form -synopsis override now renders in synopsis output, and candidate-form ranking recognises -choices discriminators (the tclcore models' subcommand words) via the shared choice resolver.

## [0.12.14] - 2026-07-13

- punk::args 0.10.0 (G-041 increment): multi-form argument definitions now parse by automated form selection - punk::args::parse without -form attempts every form and auto-selects the one that cleanly matches ('lseq'-style multiform commands no longer fail with the first form's error when the arguments fit a later form). No matching form raises an error naming each candidate form's failure (best candidate first); arguments matching several forms raise an error naming them, with -form (which now accepts the documented list of form names/indices) restricting to the intended form. Parse results and parse_status report the selected form and every candidate's status ('form'/'formstatus' keys).

## [0.12.13] - 2026-07-13

- G-051 achieved: punk::ns 0.4.0 - the help system's doc walk now accepts the same choice-word abbreviations parsing accepts: 'i string is tr' resolves to the 'string is true' documentation exactly as 'string is tr 1' executes (unique prefixes, aliases, nocase; ambiguous/denied/unknown words stay at the parent exactly when parse rejects them). cmdinfo also reports the truthful cmdtype 'doconly' (was 'notfound') for documentation-only levels such as 'string is <class>' and documented TclOO method docids. punk::lib 0.4.1 adapts its script analysis to the new value with identical behaviour.

## [0.12.12] - 2026-07-12

- punk::ns 0.3.0: 'i <cmd>' with no argument words again signals when the command cannot be called bare - e.g 'i if' shows "Bad number of trailing values for if. Got 0 values. Expected at least 2" in the error scheme above the usage table (regression vs older builds, reported by the user). The G-046-era suppression of that render existed because the old message carried internal-looking attribution ("for punk::args::parse ..."); the G-049 -caller attribution fixed the message wording, so the suppression was reversed rather than reworded - the original 'i string is' complaint case now renders an accurate "Bad number of leading values for string is ..." instead of being hidden.

## [0.12.11] - 2026-07-12

- G-071 achieved: punk::args 0.9.0 allocation choice screen - optional arguments with restricted choice sets no longer greedily consume non-choice words at allocation time, so noise-word grammars parse correctly with the noise word omitted. 'i lseq 0 10 2' (and the '0 10 by 2' / '1 5 by 0' shapes) now parse and render correctly against the tclcore moduledoc; genuinely invalid arglists report a plain excess-values error instead of blaming an unrelated optional argument. Correction recorded: parse_status already accepted -form before the withid/withdef tail per its documented synopsis - the earlier 'missing -form' finding was an argument-order mistake, now pinned by test.

## [0.12.10] - 2026-07-12

- punk::args 0.8.3 (G-071 increment): silenced four more unconditional debug lines that printed to stderr during normal argument parsing whenever an optional element was skipped (visible when using 'i'/help display against commands modelled with noise-word arguments such as if and lseq). New allocation characterization testsuite pins the optional-element mis-allocation ('lseq 0 10 2'-class failures) as GAP tests ahead of the allocator fix.

## [0.12.9] - 2026-07-12

- punk::args 0.8.2: fixed stray debug output ("checking tp ... against value ...") printed to the console when parsing multi-element clause arguments (e.g. 'i try ...' style usage against the tclcore moduledoc). Found during G-041 prework probing of if/switch/try/lseq real-vs-model divergence; findings recorded in the G-041 and G-055 goal detail files.

## [0.12.8] - 2026-07-12

- G-045 achieved: punk::args 0.8.1 documents the definition container quoting rules in the punk::args::define help (braced values fully literal; double-quoted values with Tcl backslash semantics at record parse while $ and [] stay literal; the backslash-escaped tstr placeholder idiom). This completes the definition authoring ergonomics goal: @cmd -unindentedfields honoured (0.6.1), -& record continuation (0.7.0), @normalize block-form indent normalization with the help-system definitions as consumer proof (0.8.0 / punk 0.2.5), quoting documentation (0.8.1).

## [0.12.7] - 2026-07-12

- G-045 (increment): punk::args 0.8.0 adds the bare @normalize directive - opts a definition into indent normalization of block-form multi-line field values (structural leading newline dropped, content re-based to the file-style 4-space continuation convention, relative indents preserved, -unindentedfields exempt, no-op on conforming file-style definitions). Intended for constructed (string-built) definitions. punk 0.2.5: the help-system definitions (::punk::help / help_chunks) convert to indented block authoring under @normalize; 'i help' rendering unchanged.

## [0.12.6] - 2026-07-12

- G-045 (increment): punk::args 0.7.0 adds the -& record-continuation token - an unquoted trailing -& on a definition record line continues the record on the next line, assembling byte-identically to the equivalent backslash continuation. Intended for constructed (string-built) definitions, where backslash-newline is consumed by the building code's own quoting. Brace a literal trailing -& value ({-&}); -& elsewhere on a line or inside braced/quoted multi-line values is ordinary data. Backslash-continuation authoring is unchanged.

## [0.12.5] - 2026-07-12

- G-045 (increment): 'i help' usage table alignment fixed (punk 0.2.4) - the Description block's continuations no longer render indented +12 relative to the first line, and the topic argument's help first line no longer renders +4 relative to its continuations. ::punk::helptopic::define_docs authors help text at the left margin with -unindentedfields {-help} on the @cmd and topic lines. Help text content unchanged.

## [0.12.4] - 2026-07-12

- G-045 (increment): punk::args 0.6.1 - the @cmd directive now honours -unindentedfields for -help, so command help authored at the left margin renders its first line flush with continuations in help/usage displays (previously the option was accepted on @cmd but ignored). No shipped definitions used it yet, so existing help rendering is unchanged; the define documentation states where -unindentedfields is valid.

## [0.12.3] - 2026-07-12

- punk::args 0.6.0 synced into the common vfs payload (_vfscommon.vfs): help/usage synopses render small restricted choice sets as literal alternates (module change committed earlier as c309ed47; this ships it in built kits).

## [0.12.2] - 2026-07-12

- G-039 fix: an interactive shell orphaned by its hosting console dying (killed conhost/terminal) previously spun ~2 CPU cores forever — the Tcl 9 windows console driver never delivers the dead-console state to the script level as a fileevent (tclWinConsole.c `ConsoleEventProc` only notifies on buffered data) while its reader thread busy-loops on the persistent channel error. punk::repl 0.5.0 adds `repl::console_watchdog`, a 5s liveness poll (read-only GetConsoleMode via `chan configure -inputmode`) armed only for a tcl9 console input channel on the process-default console on windows: on a dead console it closes the input channel (stopping the driver's reader thread) and finishes the repl via the normal eof path, so the orphan exits cleanly within seconds. Piped, foreign-console and tcl 8.6 inputs are unaffected.

## [0.12.1] - 2026-07-11

- G-062 achieved: canonical project license declared as BSD-2-Clause. `LICENSE.txt` added at the repo root with the standard BSD-2-Clause text (copyright Julian Marcel Noble, 2023-2026); `README.md` names BSD-2-Clause and points at `LICENSE.txt`; root `AGENTS.md` Repo-wide Notes names the license precisely; `punkproject.toml` `[project]` carries `license = "BSD-2-Clause"`. Repo-sweep found no top-level license claims contradicting it (vendored library licenses and module-level generated docs are not project-level claims).

## [0.12.0] - 2026-07-11

- punk::args 0.6.0: command synopses (`s <cmd>`, `i <cmd>`, usage headers) now display small restricted choice sets as literal alternates — an argument whose choice pool (`-choices` plus `-choicegroups` members) has 1–3 members and `-choicerestricted` true renders those words unitalicised joined by `|` (e.g. `after cancel` shows literal `cancel`; a 3-choice option shows `(left|centre|right)`), matching the existing display style of `literal()`/`literalprefix()` type-alternates. Larger or unrestricted choice sets keep the italicised argname/type display and an explicit `-typesynopsis` always wins, so e.g. `textblock::frame -type (choice|<dict>)` is unchanged. Synopsis rendering of type-alternates, `-typesynopsis` passthrough (including documenter ANSI styling), and multi-element clause display gained characterization test coverage (`synopsis.test`).

## [0.11.0] - 2026-07-11

- G-001 achieved: an interactive REPL can be launched against a non-detectable terminal-like device via an `::opunk::Console` subclass — `repl::init -console <spec>` selects the console (channel pair, anchored instance name, or object value), `repl::start`'s input channel defaults to it, and the repl's output (prompts, results, and the code interp's stdout/stderr — diverted via shellfilter junction stacks and emitted per run) routes through the selected console's channels, with eof/size/capability answered by the subclass overrides (punk::repl 0.4.0; base `::opunk::Console` and `punk::console` untouched). Verified end-to-end by child-process driver tests: an `::opunk::SshConsole` socket session whose scripted remote terminal answers `CSI 6n` (size resolved over the socket) and a `::opunk::TkConsole` text widget wired as a live console by new `opunk::console::tk::console` (reflected output channel rendering into the widget + Return-binding input pipe, opunk::console::tk 0.2.0). Process-console behaviours (tcl_interactive prompt gating, stdin reopen on eof, raw-mode re-enable, the windows line re-decode experiment) now apply only when no foreign console is selected; default stdin/stdout repl behaviour is unchanged.

## [0.10.3] - 2026-07-11

- make.tcl build/promotion commands (`project`, `packages`, `modules`, `libs`, `vfs`, `vfslibs`, `bin`, `bootsupport`, `vfscommonupdate`) now warn when `src/` has uncommitted fossil/git changes — artifacts built from dirty src have no committed provenance (G-026 direction). Warn-only by default; a new `-dirty-abort` flag makes the check aborting. Dirt outside `src/` is ignored (`punkboot::utils::vcs_dirty_warnings` gained an optional scope argument, utils 0.2.0). With `<builtexe> src` / `<builtexe> src shell` available for evaluating uncommitted source directly, building is the promotion step the warning treats it as.
- Provenance warnings (dirty-src gate and vendorupdate's dirty source-project check) are presented with a plain column-0 `PROVENANCE-WARNING:` token for automated discovery in redirected output plus ANSI colour for humans, and are recapped at the very end of the run (via a wrapped `::exit`) so they survive scrolling chatty build output. Interactive terminal runs (Tcl 8.7+/9 stdin `-inputmode` probe) additionally get a 3-second ctrl-c grace countdown before a dirty build proceeds; piped/agent/CI runs pay no delay. `make.tcl check` now reports src provenance status and what the build/promotion commands would do.

## [0.10.2] - 2026-07-11

- punk::libunknown bumped 0.1 → 0.2.0 (file renamed `libunknown-0.2.0.tm`), retroactively versioning the register_all_tm and source_pkgindex additions and adopting full `major.minor.patch` form. Verified nothing requires it by exact version or filename: punk_main.tcl and punk::repl glob `libunknown-*.tm` picking the highest by vcompare (the dev copy now also outversions the stale bootsupport/project.vfs `0.1` copies deterministically), bootsupport's include_modules.config lists it by name only. The module header now carries a version-history block in lieu of a buildversion.txt, and `src/modules/AGENTS.md` documents the manual-versioning bump mechanics for agents (same Patch/Minor/Major rules as buildversion-tracked modules).

## [0.10.1] - 2026-07-11

- punk::libunknown: pkgIndex.tcl scripts are now executed in an isolated `source_pkgindex` frame providing the documented `$dir` variable plus `auto_path`/`env` global links (matching stock tclPkgUnknown's contract), instead of at `::` scope with a `global dir`. Fixes any user global named `dir` being silently overwritten whenever a `package require` fell through to the pkg unknown handler, and stops index scripts' helper variables (`ver`, `pkg`, `script`, `_CawtSubDirs`, ...) leaking into the global namespace. Verified against tcllib 2.0's index behaviours (auto_path extension, apply-scoped subindex sweep, critcl-style loaders); regression tests pin the `$dir`/auto_path/no-leak contract.

## [0.10.0] - 2026-07-11

- `dev lib.search` now performs deep module discovery by default: new `punk::libunknown::register_all_tm` registers `.tm` modules at every namespace depth across all tm paths (reusing/populating the per-epoch directory index cache, at most one scan per `package epoch` per interp), so namespaced modules in never-requested subfolders (e.g. `test::*`) appear in search results without `-refresh` (punk::mix::commandset::loadedlib 0.2.0).
- `dev lib.search -refresh` repurposed to mean a genuine filesystem re-scan: it increments the package epoch (invalidating punk::libunknown's scan caches) and re-runs discovery, picking up `.tm` files added/removed on disk and re-sourcing `pkgIndex.tcl` files. Without punk::libunknown active, `-refresh` retains the previous dummy-require deep-walk behaviour.
- lib.search dependency fixes: works in bare interps now — highlight ANSI codes computed only when highlighting (fully-qualified `punk::ansi` with inline require; previously errored on the shell-global `a+` alias even with `-highlight 0`), inline requires for `punk::path` (fallback walk) and `textblock` (table output).

## [0.9.1] - 2026-07-11

- `dev lib.search` `-refresh` help rewritten to document actual semantics (punk::mix::commandset::loadedlib 0.1.1, doc-only): the flag performs a deep tm discovery pass (registering `.tm` modules in never-requested namespace subfolders), it does not re-scan directories already indexed in the current `package epoch` (run `package epoch incr` first for a genuine filesystem re-scan), and tm/auto_path list changes are epoch-invalidated automatically. New characterization test suites pin the underlying punk::libunknown discovery/epoch-cache behaviour (`src/tests/modules/punk/libunknown/testsuites/discovery/`) and the lib.search match + `-refresh` contract (`src/tests/modules/punk/mix/testsuites/loadedlib/`), including a GAP pin of the pkg-unknown handler clobbering a user global `dir` variable (stock Tcl keeps it proc-local).

## [0.9.0] - 2026-07-11

- G-001 (in progress) increment 1: three pluggable `::opunk::Console` backend modules, with the base class and punk::console untouched — `opunk::console::test` (`::opunk::TestConsole`: deterministic channel-pair test double with fixed size and probe-free eof — the console seam the repl/editbuf characterization work needs), `opunk::console::ssh` (`::opunk::SshConsole`: socket-carried terminal sessions — construction-time capability, chan-eof without byte-consuming probes, and size resolved by punk::console's ANSI cursor-report provider *querying over the connection*, proven against a scripted remote terminal answering `CSI 6n` over a socket pair), and `opunk::console::tk` (`::opunk::TkConsole`: a Tk text widget as terminal — widget char dimensions as size, backend eof marker via `opunk::console::tk::set_eof`, verified live under the tk-capable punk91 kit). Subclass values dispatch through existing base-class holders and `console_spec_resolve` virtually. Remaining for the goal: repl launch-time console selection and output-channel parameterization.

## [0.8.2] - 2026-07-11

- G-054 achieved: tclcore moduledoc 0.2.0 — the `string is` class choices shown by `i string is` (and the per-class docids like `i string is digit`) are now harvested from the running interpreter at define time instead of a hand-maintained list, so the documented/parsed class set always matches what the interpreter accepts (previously the static Tcl 9.0 list wrongly accepted `dict` under 8.6 and rejected 8.7's `unicode`). Version notes render on classes that differ across releases; unrecognized future classes get a generic label instead of vanishing. Behavioural parity pinned by `tclcoreparity.test` with expectations derived from the live interpreter — green on Tcl 8.6.13, 8.7a6 and 9.0.3.

## [0.8.1] - 2026-07-11

- `runtime.cmd` powershell payload: cached `sha1sums.txt` fallback backported from the bash payload - `list -remote` against an unreachable server now warns and compares using the previously fetched copy instead of dying on an unhandled download error, and the `fetch` path's pre-existing silent fall-through to a cached copy now announces itself. Also fixed a latent undefined-variable bug creating the runtime folder in the `list -remote` branch.

## [0.8.0] - 2026-07-11

- `runtime.cmd list -remote` parity for the unix (bash/zsh) payload: local-vs-server comparison table (Same version / UPDATE AVAILABLE / not-listed, plus remote-only entries), falling back to a cached `sha1sums.txt` with a warning when the server is unreachable. Both payloads gain a `PUNKBIN_URL` env override of the artifact server base url (mirrors/testing).
- The polyglot's 512-byte label guard caught its first real regression during this change: the payload growth pushed the template's `:exit_multishell` label across a boundary; fixed with a documented byte-alignment spacer comment in `runtime.bash` (workflow recorded in bin/AGENTS.md).
- G-059 achieved: WSL capability probing + native-filesystem staging for driving unix-side tests from Windows (`src/tests/testsupport/wslprobe.tcl`, `wsllinux` constraint). The MULTISHELL sh payload and the runtime.cmd unix payload (`runtime.bash` — never previously executed on real unix) now run green under WSL: active/use/run resolution, env override, stale-active guidance, single-candidate fallback and offline `list -remote` — all staged to the distro's native filesystem, with a guard test asserting the Windows checkout's git state is untouched.

## [0.7.0] - 2026-07-11

- `bin/runtime.cmd` gains a `use` subcommand and an active-runtime selection: which runtime `run` launches is now the per-machine `bin/runtime/<platform>/active.toml` (`active = "<name>"`, shown with a `*` in `list`), resolved as `PUNK_ACTIVE_RUNTIME` env override → `active.toml` → sole installed candidate — with multiple runtimes and no selection, `run` errors listing candidates instead of silently launching the last name alphabetically. The first `fetch` sets the active runtime only when none is recorded; later fetches never steal it.
- `runtime.cmd fetch` checksum parity for the unix (bash/zsh) payload: it now fetches `sha1sums.txt`, skips when the local file already matches, downloads to `.tmp` and installs only on sha1 match (tool detection across `sha1sum`/`shasum`/`sha1`/`openssl`; refuses unverified downloads when no tool is present), and accepts an optional runtime name like the powershell payload. Bug fixes: the powershell "stored hash" message printed an empty variable; the bash MSYS branch had an invalid `interp =` assignment; `.tmp` leftovers and non-runtime files no longer appear as run/list/use candidates.
- The runtime scriptset sources moved to `src/scriptapps/bin/` (alongside `getzig.*`), the proper home for bin-deployed scriptsets; `bin/runtime.cmd` regenerated via the scriptwrap workflow (round-trip pinned byte-identical by the multishell test suite).

## [0.6.0] - 2026-07-10

- G-058: punkshell boot honours statically-linked runtime packages. Kits built on runtimes with static extensions (e.g `tclsfe-x64.exe`: Thread/twapi/sqlite3/tdbc — the punk91 kit) previously lost `package require` access to them in every interp/thread punkshell fabricates: the appended vfs replaces the runtime's own `//zipfs:/app` mount (taking its pkgIndex files with it) and punkshell's boot path replacement discards what remains — the repl code interp failed with `can't find package Thread` and punk::console couldn't load. Now `punk_main.tcl` captures the static baseline at boot (probe-loading each empty-filename `info loaded` prefix in a throwaway interp and recording only packages the load actually provides; configurable denylist for side-effecting/composite inits), and the baseline seeds `package ifneeded <name> <ver> {load {} <prefix>}` mappings in the boot interp, the repl codethread and code interp (punk::lib 0.4.0 `interp_sync_package_paths`/`snapshot_package_paths`, punk::repl 0.3.0), and shellthread 1.7.0 workers — introspectable at the repl via `::punkboot::static_packages`.
- punk::packagepreference 0.2.0: static-vs-bundled resolution is version-aware — requiring a baseline package triggers the package-index scan before resolution so all candidates register and the highest version wins (a static copy is no longer masked by an older bundled dll, and a genuinely newer bundled copy remains reachable); static *registrations* (empty-filename `info loaded` entries) no longer trip the loaded-shared-object same-version pinning. Composite statics whose script layer lived in the runtime's replaced zipfs (e.g. static twapi) provide nothing under the probe and correctly defer to bundled complete copies.
- Tests: `staticseed.test` + `staticpolicy.test` (un-gated, simulated baselines) and `staticruntime.test` (kit integration behind the capability-probed `statickitavailable` constraint). The verification runtime `tclsfe-x64.exe` is pinned in the punkbin artifact repository (win32-x86_64 + sha1sums) so the constraint is satisfiable on other machines.

## [0.5.0] - 2026-07-10

- punk::args 0.5.0 (G-049): new `punk::args::parse_status` - runs a parse attempt and returns a documented parse-status structure instead of raising on validation failure: overall `ok`/`status` (valid|invalid|incomplete)/`scheme`/`message`/`errorcode`/`failureclass`/`badarg`/`form`/`receivednames` plus per-argument `argstatus` entries (class, status ok|bad|unparsed, received count/positions, value-in-effect including `-default` fill). `arg_error` accepts the structure via `-parsestatus` and both its renderers derive goodarg/badarg row marking and choice value-in-effect highlighting from it. Scheme colours now resolve per-render: the documented `-scheme` value `nocolour` takes effect (previously it fell through to whatever colours the last scheme render left behind) and scheme renders no longer mutate the shared colour array (the strike-only goodarg style no longer leaks into later renders). `missingrequiredvalue`/`missingrequiredleader` failures now carry `-badarg`, and `punk::args::parse` gained `-caller` to control the `%caller%` attribution in failure messages.
- punk::ns 0.2.0 (G-049): `i <cmd> <args...>` (cmdhelp) gains `-return dict` - a machine-parsable dict of resolution info (origin/docid/cmdtype/args_remaining) plus the parse-status structure of the supplied argument words. Display fixes: a value failing its `-type` check now badarg-marks the unfillable argument's row (previously only choice violations marked anything), an explicit `-scheme` is honoured on the parse-failure path, and parse-failure messages name the queried command instead of leaking cmdhelp's internal parse source line at top call depth.

## [0.4.15] - 2026-07-10

- punk::args 0.4.2 (G-046): argument resolution no longer expands display-only `${...}` content (`-help` on `@cmd`/`@examples`/argument records, `@formdisplay` bodies) - expansion is deferred to help-display time with its own cache. First use of heavily documented commands is dramatically faster (`punk::ansi::mark_columns` first call ~4.3s → ~12ms; tclcore `::lseq` resolve ~184ms → ~2ms) and `-help` content that calls punk::args-parsing commands (including against its own definition id) resolves cleanly instead of stalling or looping. Also fixed: `@dynamic` multiline `-help` substitutions now align at their insertion column; prefix/alias-normalized choice values keep the same plain-string shape as exact input (`\Deleted`-style values); the `-return string` renderer aligns cmd-help continuation lines under the `Description:` label and its `Example:` line shows the example reference instead of the doc url.
- punk::ns 0.1.4 (G-046): `i <cmd>` for a command whose definition requires arguments (e.g `i string is`, `i punk::args::define`) shows plain usage instead of an internal-looking "Bad number of leading values for punk::args::parse ..." error prefix - and builds the usage table once, not twice (the discarded advisory-parse error no longer renders its own table; display timing for large argdocs stays at pre-change levels).

## [0.4.14] - 2026-07-10

- punk::args 0.4.1: first `i <ensemble>` in a fresh shell now shows subcommand-help markers and synopsis choicelabels for documented subcommands whose argdocs are lazily registered (e.g `i ansistring`); previously the autogenerated ensemble help omitted them until another path loaded the ensemble's registered namespace (`punk::args::update_definitions`).

## [0.4.13] - 2026-07-09

- punk::args 0.4.0 tidy-up (pre G-046): dead developer-test procs removed (`define2`, `parseXXX`, the unused `::punk::args::TEST` definition); the interactive `Get_caller` diagnostics and `lib::tstr_test_one` renamed with leading `__` and unexported; internal parser/renderer helpers moved from the exported `punk::args` namespace into a new internal `punk::args::private` namespace (underscore prefixes dropped); every remaining exported command now carries a PUNKARGS definition, so `i`/`punk::args::usage` render documentation for the whole punk::args API (including `punk::args::lib` utilities and the `helpers`/`argdocbase` ANSI shorthands).

## [0.4.12] - 2026-07-09

- Tcl 9.1b0 runtime compatibility (first `punk91.exe` build on `tclsfe-x64`; the developed baseline remains Tcl 9.0.2):
  - punk::lib 0.3.1: TIP 746 removed `lseq`'s expr-operand behaviour in Tcl 9.1 — `punk::lib::range` (lseq branch) now normalizes `int[+-]int` offset operands itself, so callers like `range 0 [llength $list]-1` keep working (this broke `punk::ansi::grepstr` and with it example-block highlighting and the punk::args examples.test under 9.1). Also aligned the lseq branch with the tcl8 fallback contract: default `by` now infers direction (descending `range 5 1` previously returned empty under tcl9 — silently breaking documented reverse-range callers) and `by 0` returns an empty list (9.1's `lseq ... by 0` changed to return one element). Direct `lseq` expression operands wrapped in `expr` (`lzipn_tcl9b`/`lzipn_tcl9c`/`cols`/`cols2`). `check::has_tclbug_safeinterp_compile` falls back to `interp invokehidden tcl:unsupported:disassemble` — Tcl 9.1 safe interps hide `tcl::unsupported::*` instead of exposing it.
  - punk::args 0.3.2: `zero_based_posns` expr-wraps its `count-1` lseq operand (TIP 746).
  - New `modules/punk/lib` range.test pins the `range` contract (offset expressions, direction inference, `by 0`) under both 9.0 and 9.1.

## [0.4.11] - 2026-07-09

- punk::args 0.3.1: fixed `helpers::example` `-title` path calling bare `[a]` (resolvable only where a global `a` alias exists, e.g. punk shells — errored under plain tclsh); found by the new test coverage.
- punk::args test suite grown to 128 tests across 15 files with four new characterization areas, added before any changes to the definition-parsing/indentation machinery (user-directed tests-first):
  - rendering.test (15): nesting independence (plain + tstr defs), relative-indent preservation (the 2-space help and 1-space choicelabel conventions, single and multiline), `-unindentedfields` (arg ✓; `@cmd` accepts-but-ignores — GAP pinned), constructed (string-built) definition absolute-indent characterization, multiline `${[cmd]}`/`${$var}`/`resolved_def` insertions (aligned, nesting-independent), and `@dynamic` `${$DYN_X}` double substitution (stable across resolutions, provider refresh honoured in help/choices/parse, textblock::frame-style art choicelabels byte-aligned — but multiline `-help` insertions lose insertion-column alignment for lines 2+, GAP pinned: the second-round subst lacks paramindent re-alignment).
  - choicegroups.test (5): group/`-choices` union parsing, cross-group prefixes, duplicate-entry dedup, per-group titled choice tables, and punk::imap4-style `{\Deleted}`/`{$MDNSent}` values with labels — plus a GAP pin: prefix-normalization list-quotes special-character choices (`\De` stores `{\Deleted}` where exact input stores plain `\Deleted`).
  - defquoting.test (3): container quoting rules pinned — braced values fully literal (`$`, `[]`, two-char `\n`, bare backslashes), quoted values get Tcl backslash semantics (`\n` → newline, `\\` → `\`) with `$`/`[]` still literal, and `\$\{...\}` renders a literal `${...}` in tstr-processed blocks.
  - examples.test (6): helpers::example box structure/title/`-tstr` toggle (no-dedent direct-call contract pinned), `${[example ...]}` insertion into `@cmd` help aligned with the auto `Example:` row, `@examples` → `punk::args::eg` with `#<nodisplay>`/`@#<nodisplay>` stripping (ANSI-tolerant), strip_nodisplay_lines unit, and a live tclcore `::lseq` smoke (the heaviest example consumer).
- Investigated punk::ansi::mark_columns first-call slowness (reproduced: 3.94s first call, 60µs thereafter): the argdoc `-help` contains `${[...num_columns_example]}` which itself calls `mark_columns`, whose parse resolves the same still-resolving definition — the in-source TODO already documents the recursion hazard and candidate fixes (reentrancy guard, or deferring `-help` processing out of argument resolution, which would also help general first-parse performance). Recorded for goal consideration rather than patched.

## [0.4.10] - 2026-07-08

- G-040 completed (activated and flipped to achieved same day, user-directed): punk::args 0.3.0 adds `-choicealiases {alias canonical ...}` — aliases accepted exact under any `-choiceprefix`/`-nocase` setting, participating in prefix calculation when prefixing is on, always normalizing to the canonical choice in parse results; deny semantics apply to the matched name (a denied alias needs the full alias; a canonical reached via its alias is exempt from the canonical's own deny entry); usage display folds aliases into the canonical entry as an `(alias: …)` label note. The choice-word matching chain was extracted from `get_dict` into the shared resolver `punk::args::choiceword_match`, now also consumed by punk::ns 0.1.2's `cmd_traverse` — closing the doc-lookup parity gap (deny/reserve lists were honoured by parse but ignored by the walk; the pinned GAP tests flipped to agreement) and normalizing aliases before choiceinfo lookup.
- punk 0.2.3: punk::help adopts the feature — `i help` lists one entry per registered topic (`topics` +alias help, `tcl`, `env` +alias environment, `console` +aliases term|terminal), and topic words accept aliases and unique prefixes with the user-decided minimum-prefix policy: `h`/`he`/`hel`, `e`/`en`, `te`/`ter`, and the reserved `c`/`to`/`tc` all fall through to command lookup. Argless `help` output remains byte-identical to the strict-80 baseline.
- Suites: punk/args 98 pass + 1 skip (5 new alias tests incl. the punk::help policy matrix), punk/ns 28 pass (parity tests flipped, alias traverse + display-fold assertions), punk/lib 21 pass.

## [0.4.9] - 2026-07-08

- punk 0.2.2 (doc-only, from interactive review of the help restructure): `i help` / `i help <topic>` usage tables narrowed to reasonable widths (61-68 columns, previously ~160) by manually folding the helptopic argdoc help texts at ~70 columns and rendering the generated topic-choice grid with `-choicecolumns 2`. Only the argless `help` overview is strictly 80-column (unchanged, byte-identical); usage tables do not yet wrap to terminal width, so punk::args source line lengths directly set table width — automated (potentially language-dependent) wrapping, e.g. via tcllib's text-adjust facilities, is a possible future improvement.

## [0.4.8] - 2026-07-08

- G-037 completed (flipped to achieved): new `make.tcl vfslibs` step (also run inside `make.tcl project`) propagates declared vendored platform-library packages into kit vfs lib_tcl<N> trees — declarations in `src/runtime/vendorlib_vfs.toml` (tomlish-parsed, per-package per-kit with explicit `supersedes` removal so no silent mixed-version provision; punkcheck provenance at src/vfs/.punkcheck, outside kit payloads; malformed entries fail the build naming the entry). G-024's mapvfs toml conversion was not required; the file is recorded as a candidate for absorption into it. Authority is the declaration — undeclared drift (e.g. the vendorlib/kit twapi divergence) is deliberately untouched.
- 8.6 kits upgraded to tcludp 1.0.13 via the new step (the G-036 loose-end decision): the 1.0.13 folder is dual-generation (pkgIndex selects tcl9udp1013.dll or udp1013t.dll), so the single vendorlib_tcl9 source serves punk8win.vfs/lib_tcl8, replacing udp 1.0.12. Verified on rebuilt kits: punksys/punkbi (8.6.13) and punk902z (9.0.2) all load udp 1.0.13 from their vfs; no udp1.0.12 remains under src/vfs.

## [0.4.7] - 2026-07-08

- G-036 completed (flipped to achieved): `punk::lib` 0.3.0 adds `punk::lib::check::has_libbug_udp_threadexit` — version-based detection of the tcludp < 1.0.13 thread-exit bug on Tcl 9 Windows (the per-thread exit handler closes process-global event handles, silently freezing every other udp-using thread's event loop after any udp-loaded thread exits; root-caused and fixed-by-upgrade earlier in 0.4.3). `has_libbug_*` is the new check family for bundled/vendored library bugs; `help tcl` (punk 0.2.1) now scans it alongside `has_tclbug_*`, supports a full `url` reference key for non tcl-core trackers, and no longer errors on a triggered check carrying a reference without a description. Detection is loaded-version based, falling back to the best available registered version without loading the binary. New checkbugs.test covers the classifier matrix, the live check's dict shape/consistency, and a contract test over all existing check procs. Open decisions recorded in the G-036 detail: punk8win (8.6) kit's udp 1.0.12 swap; optional upstream tickets for residual tcludp trunk weaknesses.

## [0.4.6] - 2026-07-08

- punk::ns 0.1.1: five development trace `puts` removed from the command doc-lookup machinery — "PROC auto def"/"ENSEMBLE auto def" from generate_autodef were emitted on STDOUT and polluted `i`/`s` output in script/exec contexts (visible to agents piping `i <cmd>` per the G-015 pattern); "skipping to documented subcommand", "cmd_traverse ensembleparam" and the "cmd_traverse 10 ... review" fallthrough note were stderr noise on the space-form-id, ensemble-parameter and undocumented-tail traverse paths. No functional change.
- punk/ns cmdflow.test: 8 new tests characterizing ensemble `-parameters` handling in the doc-lookup flow (the punk::netbox::man pattern — a leading apicontextid before the subcommand): generate_autodef models each parameter as a `-ensembleparameter 1` leader with @leaders min/max = nparams+1 (single- and two-parameter fixtures), cmd_traverse consumes parameter words and resolves the subcommand's explicit or autodef docid (re-presenting the parameter in args_remaining for the subcommand's own leader parse, matching punk::netbox::man's explicit defs), nested sub-ensembles with their own -parameters re-consume the curried parameter at each level, parse treats -ensembleparameter leaders as ordinary named leaders, and cmdhelp renders a documented subcommand's usage through the parameter.

## [0.4.5] - 2026-07-08

- punk::args 0.2.3: fixed `@opts -any 1` (arbitrary/adhoc option passthrough) — an option not present in the definition crashed parsing with "can't read argname" instead of passing through with its value (real consumers: vendored tomlish definitions declare `-any 1`); silenced a debug `puts stderr` that fired on every failed clause type assignment (visible on any multiform parse miss). Found while building the punk::args comprehensiveness pass.
- punk::args test suite expanded (94 tests, was 71 at start of day): new forms.test characterizes multi-form (@form) parsing and pins the G-041 gaps — parse without -form is effectively form-0-only, -form rejects the documented list-of-forms usage, and the @form -synopsis override is stored but ignored by synopsis rendering; new validation.test covers the -type validation matrix, -range/-minsize/-maxsize, -nocase choice normalization, option -multiple accumulation, required-option enforcement, unknown-vs-adhoc option handling, the -- end-of-options marker, @values -max enforcement, received-key clause ordinals, solos, and default-vs-received distinction.
## [0.4.4] - 2026-07-08

- punk::args 0.2.2 (doc-only): the `punk::args::define` documentation for the -choice* fields now spells out the verified interplay semantics — `-choicerestricted 0` passthrough of non-matching/ambiguous/denied/reserved words, `-choiceprefix` normalization of accepted prefixes to the canonical choice in parse results (with minimal-prefix highlighting in usage display), and the `-choiceprefixreservelist` phantom-entry idiom for per-choice minimum-prefix control. Characterization tests added (choices.test unrestricted-mode tests; new punk/ns cmdflow.test for the cmdwhich/cmdinfo/cmd_traverse docid flow, pinning the parse-vs-doc-walk deny/reserve divergence as GAP pending G-040).
- help system restructured onto a topic registry (punk module 0.2.0, `::punk::helptopic`): each topic (`topics|help`, `tcl`, `env|environment`, `console|term|terminal`) is a handler proc with its own punk::args definition, and the `::punk::help`/`::punk::help_chunks` definitions are (re)generated from the registry — `i help` now renders a documented usage table with topic choices and summaries, and `i help <topic>` shows per-topic documented usage (previously an autogenerated stub). `help topics` is derived from the registry (lists all aliases, fits 80 columns). No-arg overview and command-fallthrough (`help <cmdname>`) output byte-identical to before; 80-column layout preserved; verified on both generations, script and shell subcommands. `help env` without an initialised punk::config (e.g. script contexts) degrades to a one-line notice instead of an error stack. The registry is the intended seam for future subshell-declared topics (punk::config-gated — not yet a goal).

## [0.4.3] - 2026-07-08

- vendored tcludp upgraded 1.0.12 -> 1.0.13 in the tcl9 kit vfs folders (punk9win.vfs, punk9win_for_tkruntime.vfs) and src/vendorlib_tcl9 — 1.0.12's Windows per-thread exit handler closed the process-global tcludp synchronization events, so the first udp-loaded worker thread to die froze every other udp-loaded thread's event loop (root cause of the G-036 wedge; fixed upstream in 1.0.13). Verified with the G-036 regression harness: run-2 syslog workers alive (baseline on 1.0.12: wedged 4/4). Note: punk8win.vfs still bundles udp 1.0.12 (8.6 appeared immune; pending decision). The manual vfs copy was required because `libs`/`vfscommonupdate`/`project` do not propagate vendorlib_tcl<N> platform libraries into kit vfs lib_tcl<N> trees — gap recorded as goal G-037.

## [0.4.2] - 2026-07-08

- piped-stdin exit/quit hang fixed (shellthread 1.6.3, shellfilter 0.2.4): repl/app teardown no longer blocks forever when a shellthread worker's event loop has wedged — unsubscribe's settings-reset send is now async, shutdown_free_threads keeps its timeout armed across all waits, and shellfilter::run honors its -syslog option (default empty -> no runtag log worker thread) instead of a hardcoded 127.0.0.1:514 debug leftover. Syslog remains supported when explicitly configured. Diagnosis and residual root-cause work tracked as goal G-036.

## [0.4.1] - 2026-07-07

- `script` subcommand now matches `tclsh` for GUI scripts: if the script leaves a registered Tk main loop (Tk loaded with a live main window), the event loop is serviced until the last window closes, instead of sourcing-then-exiting immediately. So a Tk script with no explicit `exit` (or one that exits from an `after` callback, like `scriptlib/tktimer.tcl`) stays alive and runs — previously it flashed and died, and `shell` was needed as a workaround (relevant to G-020 GUI automation). Console scripts are unaffected; a script that errors exits at once (no hanging window). Verified on both generations.

## [0.4.0] - 2026-07-07

- `script` subcommand supports `lib:<name>` scriptlib scripts (with or without `.tcl` extension), matching the `shell` subcommand's scheme: `punkexe script lib:hello`, or bare `punkexe lib:hello` via reclassification. The prefix always wins — a literal path beginning `lib:` (pathological; illegal on Windows filesystems) is reachable via `./lib:...`. Not-found errors list the searched locations. Only `.tcl` runs via the script subcommand.
- punk::path 0.2.2: new `scriptlib_resolve` — the shared `lib:` resolution policy (kit-internal `app/scriptlib` first and not externally overridable, then scriptlib dirs relative to the executable), PUNKARGS-documented. app-punkshell still carries its inline copy of the policy (refactor onto the shared proc noted as todo).

## [0.3.1] - 2026-07-07

- `script` subcommand: core-command documentation parity with the interactive shell — app-punkscript now loads `punk::args::moduledoc::tclcore` (catch-guarded, ~40ms) so e.g. `'i list' | punkexe script` renders the doc table instead of "Undocumented command". Verified on both generations; kit startup ~0.33s.
- punk::packagepreference 0.1.1: the moduledoc auto-load success notice moved from stdout to stderr — stdout must stay clean for script/exec contexts (the failure branch already used stderr).

## [0.3.0] - 2026-07-07

- `<punkexe> script` completed (goal G-015 achieved): runs a script file (`script <file> ?args?`, conventional ::argv0/::argv) or the whole of piped stdin (`<commands> | <punkexe> script` — no trailing `exit` needed) in the default punk shell module/alias environment, so one-liners like `dev projects.work *name*` work with no boilerplate. Honest exit codes (0 success, 1 error with errorInfo on stderr, script's own `exit` honoured); no shellfilter transforms or logging side effects; never falls into an interactive shell. Implemented as the lean `app-punkscript` package; the stdin form echoes the script's final result when non-empty. Verified on both Tcl generations (punk902z, punksys).

## [0.2.7] - 2026-07-07

- vendored tomlish 1.1.10: the library API now carries punk::args (PUNKARGS) documentation (from_toml, to_dict, from_dict, to_toml, update_tomlish_from_dict) — the G-014 tomlish precondition is met (documented upstream, re-vendored from a clean checkout); plus the explicit `tomlish::cmdline_main` entry point.
- `_vfscommon.vfs` kit payload refreshed to the current built modules: shellfilter 0.2.3, shellrun 0.1.4, punkboot::utils 0.1.1, punk::console 0.7.1 content catch-up, tomlish 1.1.10.

## [0.2.6] - 2026-07-06

- shellfilter 0.2.3: `shellfilter::run` now guarantees tee-stack removal (stdout/stderr restoration) via try/finally even when its body errors, and the decorative stack-status table rendering in `log::critical` blocks is catch-guarded. Previously an error there (observed: a stale same-version textblock snapshot calling the new punk::ansi `sgr_merge_singles` with old option syntax) aborted the run between channel diversion and restoration, leaving the process's stdout/stderr broken and all subsequent output silently lost.
- runtests.tcl: testinterp tm path order fixed so the bootsupport snapshot wins same-version module ties over vendormodules (verified rule on Tcl 9.0.3, standard and punk::libunknown scanners: `tcl::tm::add` prepends, head of `tcl::tm::list` wins ties, so the last `test_tmlist` element wins); testinterp auto_path gains the parent of `[info library]` so binary packages such as Thread resolve in projects without a root `lib_tcl<N>/<arch>` payload.

## [0.2.5] - 2026-07-06

- `make.tcl vendorupdate` now checks each vendor source project's fossil/git checkout state and warns (non-fatal, once per VCS root) when pulling from a dirty tree — vendored artifacts built from uncommitted source have no committed provenance. Enforcement policy (abort/override, bootsupport coverage) tracked by goal G-026.
- punkboot::utils 0.1.1: the dirty-checkout check lives in the module as `vcs_dirty_warnings` (make.tcl calls it via a guarded lazy require and degrades to skipped if the bootsupport snapshot lacks it); bootsupport snapshots refreshed (punkboot::utils 0.1.1, shellrun 0.1.4, punk::console 0.7.1 content catch-up).

## [0.2.4] - 2026-07-06

- shellrun 0.1.4: added missing `package require punk::args` (the 0.1.3 migration left `runx`/`run`/`runout`/`runerr` calling `punk::args::parse` without requiring it, breaking shellrun in bare interps — including the `src/tests/runtests.tcl` testinterp, which made every source-tree test run fail).

## [0.2.3] - 2026-07-06

- shellrun: removed `-tcl` from `sh_run`, `sh_runout`, `sh_runerr`, `sh_runx` (was silently swallowed with no useful effect because `sh -c` is always an external shell; now rejected with a usage message).
- shellrun: migrated `runx` from the legacy `get_run_opts` parser to `punk::args::parse` with a full PUNKARGS spec; `runx` now honours `-debug` and `--timeout=` (previously accepted but silently ignored).
- shellrun: expanded PUNKARGS documentation for `run`, `runconsole`, `runout`, `runerr`, and `runx` with `@cmd` metadata and `-help` text on all opts/values.
- shellrun: migrated `sh_run`, `sh_runout`, `sh_runerr`, `sh_runx` from `get_run_opts` to `punk::args::parse` with dedicated PUNKARGS specs.
- shellrun: removed the internal `get_run_opts` proc; all callers now use `punk::args`.

## [0.2.2] - 2026-07-06

- punk::console: terminal queries fired while an 8.6-based shell waits at an idle line-mode prompt now fail fast with a discriminable errorcode (`PUNK CONSOLE QUERY HOSTAGE_COOKED_READ`) and emit nothing, instead of timing out (~500ms) and later corrupting input with the swallowed response (idle-reader hostage guard; raw mode and mid-command queries unaffected). Owner-routed queries preserve the owner-side errorcode.

## [0.2.1] - 2026-07-06

- `make.tcl projectversion`: new advisory subcommand verifying `CHANGELOG.md` matches `punkproject.toml` and warning if `src/` has commits since the last project-version bump.
- Clarified project versioning policy: the `make.tcl` command interface is part of the product surface.

## [0.2.0] - 2026-07-05

- Initial tracking of the punkshell project version in `punkproject.toml`.