Completes the user's conversion of punk::mix::base from manual
versioning to the magic-version mechanism: header Meta block + magic
provide block in base-999999.0a1.0.tm, version history relocated to the
new base-buildversion.txt (0.1.2 - same version, mechanism-only
change). The file RENAME itself landed one commit early (ab3e1a45)
via the staged index riding an unrelated commit - net effect is the
repo's preferred move-then-edit sequence, recorded here for clarity.
src/modules/AGENTS.md: punk::libunknown is now the only manually
versioned module; base noted as converted 2026-07-21.
Bootsupport syncs from the rebuild ride along: base-0.1.2.tm restamped
from the converted source; libunknown comment catch-up
(runtime.cmd -> punk-runtime.cmd reference).
Conversion verified: make.tcl modules builds base-0.1.2.tm from the
magic-versioned source under Tcl 9.0.3 and 8.7a6 (the original build
failure was the unrelated pre-c971e6c7c4 zipfs mkzip dotfile defect,
fixed by cli 0.5.2's fallback in ab3e1a45).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
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 (in the modpod source since G-087 stage 3)
whenever a modpod rebuild triggers under such an interp (c:/tcl87 =
8.7a6 is the PATH-default tclsh here). Diagnosed from the eager
punkcheck INPROGRESS remnants (two died-mid-transfer attempts) plus a
3-file minimal repro; the suspected punk::mix::base version-scheme
conversion was coincidental and is innocent.
The fallback reuses the invocation already established for zipfs-less
interpreters (punk::zip::mkzip -base ...), deletes the partial zip a
failed mkzip leaves behind (punk::zip::mkzip refuses to overwrite),
and notes the interpreter + upstream ticket on stderr. Healthy
interpreters keep the zipfs path.
Verified: re-triggered templates-modpod rebuild under 8.7a6 completes
exit 0 via the fallback; the built modpod mounts with identical entry
sets under 9.0.3 and 8.7a6 (incl the .fossil-custom/mainmenu payload
that broke mkzip); punkcheck/mix/lib suites 193 pass under 9.0.3.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Next in the user-confirmed sequence after G-097 (achieved 2026-07-21),
ahead of the 8.6 arc (G-099/G-100/G-101). Contract untouched - the
naming decision, piperepl default-on policy, tcltls battery line and
TIP-741 temp-dll characterization were already recorded in the detail
file while proposed.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
User-approved goal from the modpod characterization session. The modpod module
carries broken tar-era code (undefined tar::readHeader connect path,
::tarjar::connected reference, con/connected + versionpart typos, malformed
set/lindex constructs, debug stderr, unimplemented wrapped get) while the live
zip surface (make_zip_modpod + self-mount stub) has ZERO automated tests
anywhere - the detail file inventories both with line refs.
Preservation-aware by design: tarjar-system artifacts (vendored tarjar-2.4.3.tm
carrying the G-110 .TclSoCache prior art, the _build #tarjar tree, cli
#tarjar-* build branch, libunknown skip-globs) are assess-only with
per-item user-confirmed dispositions - old tarjar material at
c:/tclmodules/main (incl the intermediate tarpack generation) is preserved and
migrated manually by the user, outside this goal.
Acceptance: defects dispositioned + recorded, tarjar grep clean-or-deliberate,
testsuite (wrap -> require from disk + zipfs paths, binary payload, unwrapped
redirect, both -offsettype forms) passing on tcl9 kit + 8.6 (or recorded G-034
limitation), re-vendored into shellspy.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The prior note claimed no .tclsocache code survives, based on a grep whose
output was head-truncated - the user spotted .TclSoCache inside the vendored
tarjar-2.4.3.tm (itself a tar-format .tm). Corrected with the surviving
locations (vendored tm ~L746/L834, _aside + _build copies, unwrapped
loadscript, c:/tclmodules/main - which also holds the intermediate tarpack
generation) and the actual mechanism as design contrast: per-user
~/.TclSoCache keyed by IN-ARCHIVE PATH (not content), warn-but-load-stale on
size/mtime mismatch, retry-once concurrency. Path-key collisions and stale
loads are exactly what content-hash keying eliminates.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
tarjar (modpod's tar-based predecessor) had a shared .tclsocache dll cache -
in-house prior art for the extraction-cache concept, though without content-hash
or collision handling; no code survives in current trees (grep-verified). Old
tarjar material at c:/tclmodules/main is preserve/partly-migrate, handled
manually by the user (explicitly not a goal) - tarjar artifacts are not
free-to-delete cruft.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The G-101/G-105 notes added earlier today referenced G-058 without its
achieved/archive marker - goals_lint flagged both. Marked per the sweep-rule
form (achieved 2026-07-10, archive path).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Two goals split out of the modpod/TIP-741 characterization work, user-approved.
G-109 (libunknown manifest-declared multi-name discovery): a zip-based .tm
declares case-variant and sub-package require names in a manifest at a
well-known internal location, discovered by libunknown at scan time - with
real-disk scan performance preserved (worst-case parity with plain tclsh's
builtin tm handler; libunknown exists to fix slow-drive/large-tree scan cost).
Multi-name .tm are punkshell-only (documented); TIP 590 lowercase single-name
stays the plain-tclsh happy path; extensive-tests bar recorded.
G-110 (extraction cache investigation): whether per-run fresh-temp-dir copies
are necessary or replaceable by a single per-user content-addressed cache
(hash-keyed parent dir). Carries the 2026-07-21 sniff evidence: concurrent
sharing of one copy works; classic Win32 delete refused while mapped; POSIX-
semantics delete succeeds with mappings unharmed (a third option next to TIPs
741/709: unlink-after-load on modern NTFS). Fallback requirement recorded:
punkshell wants a win32/old-Windows fallback (industrial systems on isolated
VMs/old hardware) - capability-probe, classic-safe GC degradation. Precedents
(.NET single-file, critcl cache), sibling-dll sets, AV interplay, and the
cheapest-first implementation surfaces (modpod stub -> suite core patch ->
upstream) recorded.
G-066 note pointed at the new goals.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Characterized (windows tcl 9.0.5) how a binary inside a zip-based .tm (modpod)
loads, per user request. All three loading locations - dll direct in a kit's
zipfs, a standalone .tm on disk, and a .tm nested inside a kit's zipfs
(zip-in-zip, mounts in place, no .tm pre-extraction) - converge on the same
copy-to-temp path: the binary extracts to %TEMP%\TCL<hex>\ with the same
TIP-741 non-cleanup/accumulation flaw (verified with a #modpod-mybintest
wrapping a real tcl9thread307.dll built via modpod::lib::make_zip_modpod).
- G-066 (substantive): the characterization + the multi-package-name design -
manifest-at-well-known-location read by libunknown (not yet implemented);
multi-name .tm are punkshell/libunknown-only, not plain-tclsh loadable, which
is acceptable if clearly documented; TIP 590 lowercase single-name is the
plain-tclsh-loadable happy path; complex mechanisms OK with extensive tests.
- G-103: TIP-741 thread extended - windows flaw now confirmed for BOTH
direct-kit and modpod-.tm binary loads.
- G-105: linux leg BLOCKED - local WSL is tcl 8.6.14 (no zipfs); needs a linux
tcl9 runtime. Windows done; linux expectation (no accumulation) unverified.
- G-067: binary modpod .tm as a publish payload; cross-refs.
Temp reclaim also done this session: 15,596 orphaned %TEMP%\TCL* dirs removed.
Notes only; no contract changes. goals_lint clean.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Ran the copy-to-temp characterization on the suite's own tclsh90szip.exe 9.0.5:
loading a dll out of a zipfs mount extracts it to %TEMP%\TCL<8hex>\<dll>, a new
dir per load, none removed on exit. 'info loaded' reports the //zipfs:/ path so
the temp copy is invisible to script. Unbounded accumulation confirmed in the
wild: 15,609 such dirs in this machine's %TEMP% spanning ~5 months of kit runs.
Windows = the known TIP-741 flaw, no in-runtime cleanup in 9.0.5. Linux/WSL and
qemu guests still to characterize (unix can unlink-while-mapped, so expect no
accumulation - verify).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Self-contained-but-loads-binaries (the tclkit model) resolved: tclkitsh
binaries are self-contained via an embedded Metakit/zip VFS yet DYNAMICALLY
linked to libc - that dynamic libc is what makes dlopen/'load' work; extensions
reach Tcl via the stubs table handed through the interp (no host symbol export
needed). "Self-contained" and "static libc" are orthogonal.
- G-105 (Notes were empty): the dynamic-libc first-linux-target choice is the
tclkit model; fully-static-musl 'load' limitation stands; compiled-in
alternative is G-058.
- G-101 (Notes were empty): the prebuilt tclkit86bi/tclkit-win64-dyn runtimes
it replaces are dynamic-libc Metakit kits; loadable binary batteries imply a
dynamic-libc runtime; container choice is orthogonal to libc linking.
- G-103: temp-copy dll cleanup wart the family runtimes inherit - Tcl core
TIP 741 (DRAFT, windows-framed: TclpFinalizeLoad; extract-to-temp preserved
vs TIP 709). Requirement (user): test on ALL target platforms (windows,
linux/WSL, G-060 guests) that temp-dll/dir accumulation does not happen or is
characterized. Applies to G-101's Metakit kits too.
- G-060: one-line cross-platform test item pointing at the G-103 requirement.
Notes only; no contract changes. goals_lint clean.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
bi (batteries-included) kits should carry tcltls (core.tcl-lang.org/tcltls) as
a suite-built battery with a zig-built OpenSSL/LibreSSL backend. TLS is already
a consumer dependency (punk::imap4, punk::netbox) shipped today only as vendored
prebuilt binaries (tcltls1.7.23/tls2.0b2) - building it retires those and
advances G-004.
Recorded as a battery line, not a separate goal, and explicitly non-gating:
the crypto-backend build (the real work - no OpenSSL/LibreSSL in the tree yet)
lands when convenient; tcltls stays a vendored binary in the interim. Aim is
to validate the general build/test/kit processes, not rush a releasable product.
Notes capture the feasibility crux (thin extension over a crypto lib built under
zig-only) and the G-107 test-step follow-on.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
All six acceptance clauses satisfied (suite tclsh90s.exe 9.0.5, zig 0.16.0,
machine superbee): thread 143/0 + tclvfs 51/7 GATED with tracked dispositioned
baselines, deterministic across consecutive full runs; tcllib 62328/54 WITH
tcllibc accelerators engaged + tklib 796/18 RECORDED with artifact paths in the
detail file; tk opt-in step with a recorded first census (9865/27, native-
dialog files excluded by default); per-library policy/testargs/notfiles
overridable at invocation; no-totals fails the step in gate AND record mode
(verified); evidence-summary shape documented for the G-103 metadata consumer.
Flips G-107 active -> achieved and archives it (index entry to GOALS-archive.md,
detail file to goals/archive/). Reference sweep: G-103 Notes gain the thread-
payload-refresh forward-pointer (the suite now derives thread naming from the
checkout; the punk905 kit payload still carries the old 3.0.1 copy) and the
evidence-summary-shape pointer; G-108 Notes record that its constrained-test
runs reuse G-107's test_gate generalization + -notfiles knob.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The Tk opt-in census surfaced that the full suite cannot run unattended in
the batch/suite-built shell: winDialog.test (72 cases) and winMsgbox.test (19)
pop real OS dialogs whose message-injection automation does not drive them
here, so each blocks waiting on a human. The first attempt hung there and the
engine correctly reported no-totals/fail.
Adds a dedicated file-exclusion mechanism: tools/test_gate.tcl gains -notfiles
(a Tcl list of test-file globs passed to the driver as one -notfile list
value - a flat -testargs cannot carry a multi-file list through the option
chain, since -notfile takes a single list arg), and build905.zig gains a
per-library -Dtestnotfiles-<lib> knob defaulting to each LibSuite's
default_notfiles. tk defaults to excluding winDialog.test + winMsgbox.test so
its census completes hands-off; -Dtestnotfiles-tk="" re-includes them for a
manual interactive run. The exclusion is recorded in the evidence summary
(notfiles key). Empty for every other library (no behaviour change).
Verified: tk census now completes unattended (9865 run / 8127 passed / 1711
skipped / 27 failed, result pass; log shows winDialog/winMsgbox in the driver
skip list); thread gate still PASS with notfiles {} recorded.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The external libraries the suite builds and ships now run their own
testsuites under the suite-built shell as named recipe steps - evidence for
the exact zig-built combinations nobody upstream tests.
tools/test_gate.tcl generalized into the shared engine: -mode gate|record
(gate = parsed totals diffed against the tracked dispositioned baseline,
the core-gate mechanism; record = run must complete, failures recorded not
fatal), -driver selection (tcllib-family trees run via their own
support/devel/all.tcl - sak's underlying driver, standard aggregate totals,
per-file child interps), -summaryfile/-library line-record evidence
summaries for the G-103 artifact-metadata consumer (shape documented in the
goal detail file), file-error evidence capture (@+ blocks, runAllTests
error-exit sections). A run with no parseable totals FAILS the step in
every mode - infrastructure breakage must never masquerade as recorded
results. Core gate semantics unchanged; the core gate step now also writes
out/testreports/tclcore.summary.
build905.zig: steps test-thread test-tclvfs test-tcllib test-tklib test-tk
+ composite test-libraries (tk excluded - opt-in: interactive desktop, maps
windows, tens of minutes; intended before publishing bi-family artifacts).
Per-library -Dtestpolicy-<lib>=gate|record|skip and -Dtestargs-<lib>
invocation overrides - no recipe edits needed. Baselines staged in both
flows (suite.tcl item list + bootstrap recipe_items).
suite.tcl: test action takes -steps (default test-gate) and a -zigargs
passthrough (both actions); passes -j1 so combined test invocations
serialize (zig runs independent steps concurrently; thread's suite is
timing-sensitive).
Census results (suite tclsh90s.exe 9.0.5, zig 0.16.0): thread 143/0 failed,
EMPTY baseline, GATE PASS x3 identical consecutive runs; tclvfs 51/7
failed, all dispositioned (3x tcl9 glob-nomatch semantics through vfs
handlers, 4x dead-FTP network tests), GATE PASS x3 identical; tcllib WITH
tcllibc accelerators engaged (E tcllibc markers, crc32-critcl-* variants
running) RECORDED 62328/54 failed; tklib RECORDED 796/18 failed (widgetPlus
drift under Tk 9). No-totals failure path verified in both modes.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Found by G-107's first library-test census: build_tclthread.zig compiled
thread trunk 3.0.7 sources with -DPACKAGE_VERSION="3.0.1", so the dll
self-reported the wrong version, and the pkgIndex declared only capital-T
Thread 3.0.1 - the thread testsuite's 'package require -exact thread 3.0.7'
could never load the package (nothing previously load-tested thread at all).
The version now derives from the tree's AC_INIT (new build_common helpers
acInitVersion/stripChars) and flows into every version-derived name: dll
tcl9thread<nodots>.dll, install dir lib/thread<version>, PACKAGE_VERSION/
PACKAGE_STRING defines, rc resource fields, and a pkgIndex declaring BOTH
'Thread' and 'thread' (threadCmd.c provides both; the testsuite requires
lowercase, existing punkshell consumers the capitalized name). Dead
zip-related consts riding the stale version removed.
Note for the next kit/vfs refresh: the punk905 kit payload
(src/vfs/punk9wintk905.vfs/lib_tcl9/thread3.0.1/) still carries the old
self-consistent 3.0.1-stamped copy - refresh to the derived naming then.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Index entry + detail file for a debug tier of suite_tcl90: build debug-flavoured
runtimes (NDEBUG off / tcl::test compiled in; deeper tier adds TCL_MEM_DEBUG for
the script-level memory command), named with a trailing -dbg<n> suffix, and run
the core suite's normally skipped C-test-command-constrained tests in record
mode (never gated). Kit-family debug members coordinate with G-103; the
TCL_MEM_DEBUG all-or-nothing linking constraint drives the feasibility item.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
New goal from the 2026-07-21 discussion: the libraries the buildsuites compile
and ship (thread, tclvfs, tcllib+tcllibc, tklib, tk) get their own testsuites
run under the suite-built shell as named recipe steps with per-library policy -
gate (core test-gate mechanism, dispositioned baselines), record (report
artifacts under out/testreports/), or opt-in (tk). Context carries the source
survey, test_gate.tcl generalization notes, G-102 cost knowns, and the G-103
metadata / 8.6-arc relationships.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The buildsuite commits added .zig-cache/ and zig-pkg/ to .gitignore (zig 0.16
cache and fetched-package extraction dirs beside the suite manifest) but the
derived fossil ignore-glob was not updated - fossil extras listed ~12k
zig-pkg/.zig-cache files under src/buildsuites/suite_tcl90. Same derivation
rules as the existing zig-cache/zig-out entries.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Records the 2026-07-21 discussion: make.tcl's zip-kit extract+merge_over+mkimg
flow already unifies runtime-zip and vfs-folder payloads, so the bi family is
isomorphic to a .vfs kit without its main.tcl fauxlink. Where the equivalence
breaks (tcl_library provenance, stock-boot path setup, dll-from-zip loading) is
the goal's design content; generators stay separate (bootstrap hermeticity,
release axis) and meet at the attached-image contract - the concrete payload
contract promised to G-101.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Clarify that ad-hoc working-tree snippets should use bin/punk91 src script rather than live src shell mode.
Assisted-by: harness=pi; primary-model=unknown; api-location=unknown
The tcl-whatis and tcl-nslist skill descriptions contained ': ' mid-string,
which is illegal in a plain YAML scalar and broke frontmatter parsing in
strict harnesses. Wrap both description values in double quotes (all four
copies: .agents/skills and .claude/skills).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Layout fold ("fold stale layout fixes into this"): both layout stores
(src/project_layouts/vendor/punk/project-0.1/bin and the modpod-templates
equivalent) had shipped a stale 2025-era runtime.cmd and a pre-G-096
getzig.cmd - the getzig rename had missed this channel entirely, so
generated projects still received old-named, stale utilities. Both renamed
via git mv AND refreshed to byte-copies of the current bin/punk-runtime.cmd
+ bin/punk-getzig.cmd; the layout src/runtime/Readme.md references updated
in both stores (the layout bin/runtime/ data-dir READMEs are gitignored -
updated on disk only, like the root bin/runtime/README.md); residual
old-name scan of the layout trees clean. The built modpod zips (root
modules/ + bootsupport templates-0.2.0.tm) carry old names until their
normal re-vendor cycles. Layout paths join the goal Scope per the approved
fold.
Flip mechanics: detail file to goals/archive with Status achieved
2026-07-21 (verification in its Progress section: full rename sweep, pins
retargeted and passing 26/25/1/0, drift findings recorded), GOALS.md entry
removed, GOALS-archive record appended (amended Scope + Acceptance
verbatim), reference sweep: G-103 scope parenthetical (both tiers) +
Relationships and the G-099/G-103 sequencing notes annotated with the
achievement; G-096's archive gains a note recording the layout-channel gap
and its closure. goals_lint clean (75 active-index goals, 31 archived).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Root cause of the punk86-kit runtests failure: a punk-kit interpreter's
boot pre-registers its kit-stamped package versions, and an unversioned
require satisfiable by such a registration never fires package unknown -
the freshly-added bootsupport tm path is never scanned. The old kit's
punk::args 0.1.0 won the require, then the versioned punk::path 0.4.0-
pulled a modern module whose load-time argdoc needs punk::args::define:
invalid command name.
Floors added: punk::args 0.12- (runtests.tcl toplevel), punk::lib
0.5.1- (both 8.6 lpop guards; punk::path 0.4.0- already floored in
discovery.tcl). Feature-era minimums, not latest pins - their purpose is
rejecting fossils so package unknown resolves the bootsupport copy.
Verified: the floors alone resurrect even the OLD fossilized punk86 kit
(discover-only + a 35/35 suite run); native 9 and 8.6 unaffected.
concurrency.test child preamble floors its punkcheck require (0.6-):
kit-hosted children boot with kit-stamped punkcheck PRELOADED, and a
stale kit (pre-vfscommonupdate _vfscommon) made children fail on
'invalid command name punkcheck:🔒:acquire' - the floor turns that
into a loud version-conflict in the child log.
src/tests/AGENTS.md records the floor rationale and the kit-hosted
child-suite requirement (vfscommonupdate-current kit content).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The remaining pre-policy punkshell-own bin utilities take the punk- prefix
(bin/AGENTS.md policy, G-096); external-tool wrappers (dtplite/sdx/kettle)
and getpunk stay unprefixed. No behaviour changes ride along.
- runtime scriptset: sources/toml renamed (src/scriptapps/bin/punk-runtime.*),
the eight ps1 usage/guidance messages now say punk-runtime.cmd, the bash
payload's filename-derivation comment example updated; REWRAPPED ->
bin/punk-runtime.cmd (checkfile clean) with the .ps1 twin regenerated as a
byte-copy. The roundtrip + checkfile pins retarget the new names and pass.
- tclargs + bits: sources/tomls renamed; bin artifacts moved BYTE-FAITHFULLY
(git mv, no rewrap): a pre-rename drift check showed both carry
pre-template-refresh wraps (477-byte delta = the template's 2025
wmic-deprecation timestamp fix; runtime was in sync). Regenerating them is
a behaviour change and is deliberately not smuggled into the rename -
recorded in the goal. Fresh .ps1 twins cut from the current .cmds; stale
untracked bits.bat/bits.ps1/tclargs.ps1/runtime.ps1 removed.
- selfsign experiments: SELFSIGN.PS1 -> punk-selfsign.ps1 (case normalized),
wrap toml + cert-create script renamed; the four untracked bin experiment
scripts renamed on disk; the exported .cer is not a script and keeps its
name.
- runtimebash_wsl.test: functional staged-filename sites updated (bash
./punk-runtime.bash; solo-fixture cp) along with comments.
- Docs: bin/AGENTS.md (examples, policy paragraph records the sweep, runtime
contract section, pin pointer corrected to runtimecmd_roundtrip.test),
src/scriptapps/AGENTS.md, root AGENTS.md, README.md, src/README.md,
untracked bin/runtime/README.md, G-060 reference, libunknown comment (src
copy; root modules/ + bootsupport copies catch up on their re-vendor
cycles).
Suites green: scriptwrap + binscripts 26 run / 25 passed / 1 constraint-skip
/ 0 failed - the roundtrip pin passes against the renamed artifact.
Discovered and recorded for user decision (goal Progress): the project
LAYOUT stores ship stale utility copies - runtime.cmd (2025-era payload) and
getzig.cmd (the G-096 rename missed this channel) - so generated projects
still receive old-named utilities; folding the layout channel into this
goal's scope vs a layout-sync mechanism is flagged.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
First of the confirmed sequence G-097 -> G-103 -> 8.6 arc: the bits/runtime/
tclargs scriptsets and selfsign experiment scripts take the punk- prefix per
the bin/AGENTS.md policy, ahead of G-103 extending the runtime scriptset.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The skills existed but sessions never leaned on them - grep+read wins
the tool-selection contest when descriptions are phrased as information
needs ("use when you need to know how to call...") rather than task
triggers, and grep 'succeeds' without revealing which of the repo's
coexisting module copies is loaded. Changes to convert them from
available options into workflow steps:
- SKILL.md descriptions rewritten task-first ("Use BEFORE reviewing,
testing, editing..." / "Use BEFORE writing any new Tcl helper..."),
with run-it-first wording, the multiple-module-copies rationale, and
cost stated up front (one sub-1s side-effect-free invocation).
- Invocations shortened to lib: script-path form (verified working,
~0.45s each): punk91 src script lib:developer/whatis|nslist.
- New caveat in both: prefer the Bash tool over PowerShell (attached
console can flip colour/terminal detection in punkshell children).
- AGENTS.md Repo-wide Notes entry converted from reference note to
DIRECTIVE: whatis/nslist first, grep reserved for full-text searches.
- .claude/settings.json: allow entries for both scripts (Bash +
PowerShell, punk91/bin/punk91/./bin/punk91 prefixes) so first use is
prompt-free.
.claude/skills copies re-synced from canonical .agents/skills
(diff-verified byte-identical).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The two child-spawning tests (two-writers, lock-holder) proceed on the
children's done-marker files, but a just-finished child can still hold
its redirected stdout log open for a beat after writing the marker - on
Windows the -cleanup tree delete then hits 'permission denied'
(observed as a ~1-in-3 flake under the slower Tcl 8.6 interp, exposed
by the new native-8.6 runtests support; same race existed under Tcl 9).
cleanup_workspace retries the delete briefly (10s deadline) instead of
flaking. 8/8 consecutive clean runs under 8.6.17 post-fix; Tcl 9.0.3
punkcheck subtree 84/84.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The runtests toplevel guard loads the bootsupport punk::lib snapshot -
this re-vendor puts the parity-fixed compat family there (the
testinterp guard already resolved the dev module via prefer-latest).
punkproject.toml 0.17.4 + CHANGELOG entry.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The validated compat family is brought across for the harness: guarded
'package require punk::lib' when ::lpop is absent, at both boot levels -
the toplevel two-stage boot (right after punk::args, whose parse needs
lpop at runtests.tcl's own argument parsing) and the singleproc
testinterp (before the shellrun preload, whose runx parses via
punk::args::parse). No-ops on Tcl 8.7+/9: -discover-only stays ~0.35s.
Native 8.6.17 verification: -discover-only works; punkcheck subtree
84/84, punk/lib subtree 43/43+1 gated skip, runner discovery suite
22/22, and a multiproc (-singleproc 0) probe 49/49.
Tcl 9.0.3 regression per the src/tests/AGENTS.md runner-change
contract: full suite in both modes with -report json,
runtests_parity.tcl reports PARITY: ok (files=96 total=1064
passed=1047 skipped=16 failed=1 - the sole failure is the documented
pre-existing exec-14.3 baseline).
src/tests/AGENTS.md records the supported-interpreter change and the
still-broken punk86-kit runtests route (kit/dev module skew - use a
native 8.6 tclsh for 8.6 legs).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Reviewed against the real 8.7/9 lpop via a 24-case differential (results,
final list state, error messages, errorcodes) - the compat proc was
correct on every success path; all divergences were error-path:
- out-of-range errors said 'tcl_lpop index ...' with no errorcode; now
the builtin's exact message and {TCL VALUE INDEX OUTOFRANGE}
- variable-access failures named the internal upvar link variable and
flattened detail (missing array element misreported as 'no such
variable'; unset var lost its TCL LOOKUP errorcode). Fixed by probing
with a plain read in the caller's frame BEFORE creating the upvar
link - through a link, or once a link merely exists, Tcl downgrades
the error class.
Finding: the assumed 8.6 divergence does not exist - 8.6's core index
parser already accepts TIP-176 M+N arithmetic, so compat::lpop now has
zero known divergence from the builtin (doctools note corrected).
compat.test extended (3 -> 10 tests): a differential parity matrix that
runs wherever the builtin exists (self-updating pin against future core
changes), static pins for values/errors/errorcodes that also run under
8.6 where no builtin exists, an unconditional M+N arithmetic pin, and
an 8.6-only pin that punk::lib installs lpop as the compat alias.
Green under Tcl 9.0.3 (9 pass/1 gated skip) and native 8.6.17
(9 pass/1 gated skip).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Two naming tiers mapped by tooling: immutable fully-qualified ARTIFACT names
in punkbin vs short stable WORKING names consumed by mapvfs and typed by
users, with 'runtime use' and mapvfs as the mapping. Runtime artifacts:
tclsh<dotted-patchlevel>[-punk[-bi]]-r<N> (dotted patchlevels required,
always-present assembly-revision tail, no punkshell version or patch revision
in the name - metadata is authoritative); working names drop the -rN. Kits
flip the release axis to the punkshell version: major-only typable working
names (punk9/punk8/punk9bi), artifact names punk<major>-<punkshellversion>;
the dev-matrix style (punk905 etc) remains a recognized owner-controlled
third style. Hyphen separators (no punk- prefix or PATH collision in these
namespaces), '+' banned for punkbin URLs, ordering by metadata not filename
sort. Resolves the acceptance's tcl-patchlevel/punk-version/separator
questions; the acceptance's pre-decision tclsh905punk.exe example is noted
as superseded by the dotted -punk shape.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
User confirmation 2026-07-21, noted in the three affected goals: the bin
naming sweep runs first (runtime.cmd -> punk-runtime lands before G-103
extends the scriptset's list/use), the kit family + naming decision +
punkbin publication next, and the suite_tcl86 fork is taken only after the
family assembly steps exist in suite_tcl90 (fork-timing logic per
G-102-before-G-099), giving G-101 a concrete payload contract.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The archived verification is corrected to the accurate history: the earlier
scrubbed passes proved each pipeline step green individually while the
single-invocation composite was failing on the decorative echo banner steps
(a pipeline-exit misread had over-claimed it). With the banners removed the
clean-clone composite now runs to a true exit 0 with all four shells landing
(incl tclsh90szip/tclsh90sprzip) and every smoke green; the echo-spawn hazard
joins the environment findings. The sprzip install residual is marked resolved
(user application released the file; copy completed).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The five decorative 'echo --Installing package X--' addSystemCommand steps
(2024-era) spawned 'echo' as a program - a cmd BUILTIN that only resolved when
a coreutils echo.exe happened to be on PATH. In the PATH-scrubbed bootstrap
environment the spawns failed and took the tm installs, install-libraries and
(transitively) make-zipfs down with them - the reason the zip-wrapped shells
never landed in scrubbed composite runs while every package step was green.
zig's own step reporting makes the banners redundant.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Acceptance verified in full (evidence in the detail file's Progress
section): concurrency suite green on Windows Tcl 9.0.3 (runtests) and
Tcl 8.6.17 (direct tcltest drive) plus a linux WSL tclsh 8.6.14 run
from a native-filesystem staging - two-writer no-corruption/no-loss
with a clean polling reader, lock timeout naming the holder, dead-pid
and age stale-breaks, non-interactive duplicate-INSTALLER recovery.
Real builds under the re-vendored 0.6.0 run with zero lock timeouts,
zero tripwire warnings and unchanged decisions.
Index entry moved to GOALS-archive.md; reference sweep: G-027's
punkcheck note updated to achieved tense, G-028 gains the
advisory-lock name-the-blocker precedent pointer. Detail file moves to
goals/archive/ in the adjacent commit (pure rename).
punkproject.toml 0.17.3 + CHANGELOG entry.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Per the G-095 note, the advisory protocol is only effective once all
writers carry it - bootsupport re-vendored immediately after landing.
Verification under 0.6.0: 'make.tcl modules' + full 'make.tcl project'
exit 0 with zero lock timeouts, zero tripwire warnings, no spurious
recopies, all kit installs skip 'no change detected', and no
.punkcheck.lock / .punkcheck.tmp.* leftovers anywhere in the tree.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Saves are write-temp-then-atomic-rename with bounded retry for Windows
sharing violations; load_records_from_file retries transient open
denials during a writer's replace window (parse errors stay strict -
they mean real corruption). Readers gain torn-read immunity with no
reader-side protocol.
New punkcheck::lock namespace (PUNKARGS-documented): advisory sibling
lockfile <punkcheckfile>.lock via open {WRONLY CREAT EXCL}, contents
naming holder pid/host/installer/timestamp, channel held open for the
duration, in-process reference-counted re-acquisition (sequential
make.tcl installers over one root must not self-deadlock). start_event
acquires; event end / installtrack destroy releases; flushes and
chokepoint saves outside an event take a transient lock. Contention
retries with backoff to -timeout (env PUNKCHECK_LOCK_TIMEOUT, default
15s) then errors naming the holder; stale locks break on provably-dead
holder pid (twapi process_exists / kill -0, capability-gated) or age
(-staleage / PUNKCHECK_LOCK_STALEAGE, default 300s).
Deferred flushes merge own records (INSTALLER by -name, touched
FILEINFO by -targets) into freshly-loaded file state under the lock
instead of wholesale overwrite; an mtime/size tripwire warns when a
non-protocol writer (e.g. older vendored punkcheck) touched the file
between load and save.
Duplicate-INSTALLER recovery: clean actionable error when stdin is
non-interactive (punkcheck::lib::stdin_is_interactive - strict twapi
GetConsoleMode on the real stdin handle on Windows, -inputmode probe /
test -t 0 on unix); the interactive prompt gains the synced-targets
deletion-wedge caveat. default_excludefiletail_core adds .punkcheck.*
so protocol siblings are never installable content.
Consumers verified without edits: cli and five make.tcl sites end their
events (lock released at end); make.tcl's vfs/bin kit flows destroy
their installers on every path (destructor releases).
New concurrency.test (6 tests): two child installer processes with a
polling reader (no parse errors, both installers survive, no duplicate
INSTALLERs, no lock/tmp leftovers), lock timeout naming the holder then
success after release, dead-pid and age stale-breaks against planted
lockfiles, non-interactive duplicate-INSTALLER error via pipe-stdin
child, protocol-sibling exclusion pin.
Matrix: Windows Tcl 9.0.3 84/84 (runtests), Windows Tcl 8.6.17 84/84
(direct tcltest drive - the runtests harness itself needs lpop under
native 8.6), linux WSL tclsh 8.6.14 84/84 (native-FS staged run).
punk/mix subtree unaffected (150 pass / 1 known constraint-skip).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Acceptance evidence recorded in the archived detail file: PATH-scrubbed
clean-clone demonstration (c:/T/g102, env probed tclsh/fossil/git-free) ran
'zig build bootstrap' end-to-end - zon-pinned sources materialized, runtime +
piperepl variants built, smokes green (static/zip shells, Tk tooltip, tcllib,
tcllibc accelerators via critcl+zig cc), demo shell's tcl::build-info identical
to the fossil-flow build; scrubbed test gate 69552/56075/13468/9 with 8
baselined + fCmd-30.3 proven a SystemDrive scrub artifact (single-file rerun
with the var: GATE PASS), matching the same-day fossil-flow delegated gate
(69552/56040/13503/9 GATE PASS). Environment findings (critcl target probing,
windows Run-step stdio, testsuite base-env needs, firewall consent on first
socket listen) recorded in the detail file for future suite work.
Flip mechanics: detail file to goals/archive with Status achieved 2026-07-21,
GOALS.md entry removed, GOALS-archive record appended (Scope+Acceptance
verbatim), pending-tense G-102 references in G-099/G-103/G-104/G-105 annotated
with the achievement and the settled driver shape. goals_lint clean (77
active-index goals, 29 archived).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Follows straight from G-094's landing: the single save chokepoint
(installtrack save_working_recordset) is in place and the G-094
verification itself logged a transient concurrent-save failure with a
second session active in this checkout - the exact exposure this goal
closes.
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Three deltas surfaced by the PATH-scrubbed clean-checkout demonstration:
- build.zig: default-entry re-export of build905.zig's build so the documented
plain 'zig build stage|bootstrap' works from the tracked suite dir (the
recipe keeps its versioned filename; staged flows still invoke it via
--build-file).
- critcl_zig.config + recipe: settings move under an EXPLICIT target
(win32-x86_64-zig) and the tcllibc step passes -target win32-x86_64-zig.
critcl otherwise chooses its build target by probing PATH for compilers -
with a gcc visible the config defaults engaged (why G-098's flow worked),
but the scrubbed environment fell back to critcl's builtin msvc 'cl'
target and all 25 accelerator modules failed with "couldn't execute cl".
Verified green in BOTH environments (scrubbed and normal PATH).
- build905.zig comptime gate: the stale 0.13-dev minimum becomes the actual
suite pin (0.16.0) with a message pointing at punk-getzig and the README
(acceptance: the recipe records the zig version it is written against).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
The whole pipeline now lives in the zig recipe; Tcl remains the tool exactly
where a tclsh is guaranteed to exist - run by the FRESHLY BUILT suite shell
as build steps:
- New steps (native-host builds): smoke (static shell patchlevel; zip shell
patchlevel + attached tzdata/autoload), tklib / tcllib (their own installers,
batch, -pkg-path as a cached output dir installed into the prefix, + package
smokes incl withdrawn-Tk tooltip), tcllibc (critcl driving zig cc via the
tracked critcl_zig.config, module list parsed from tcllib's version.tcl,
pinned-zig dir prepended to PATH, + accel smoke), test-gate (tools/
test_gate.tcl - the parsed-totals gate vs expected_test_failures.txt,
-Dtestargs passthrough).
- Package landing moves into the recipe proper: thread pkgIndex.tcl
(build_tclthread), tk pkgIndex.tcl (build_tk), and the tclvfs script package
incl .in-template substitution + dll (build_tclvfs_shared
install_tclvfs_package) - all hooked to the install step.
- Step scripts: tools/test_gate.tcl, tools/suite_smoke.tcl, tools/pkg_smoke.tcl
(run under the built shell; assertions live in Tcl).
- Two windows-childproofing findings encoded: built-shell Runs execute the
CACHED artifact (no ../lib/tcl9.0 beside it) so each such Run gets
TCL_LIBRARY pointed at the source tree's script library (the zip smoke
deliberately excluded - its attached library is the thing under test); and
cached (non-side-effect) Runs spawn with stdout IGNORED, leaving the tcl
child without a stdout channel ('can not find channel named "stdout"',
critcl's logging) - such runs capture stdout to give the child a real pipe
while keeping caching.
- suite.tcl: thinned to the fossil dev flow - fetch live checkouts (git kind
now handled in the main loop for critcl's pin), stage the recipe, delegate
to the staged steps (run_zig_steps; 'test' delegates to test-gate). Default
-steps and the bootstrap default grow to the full pipeline short of
test-gate. critcl_zig.config + expected_test_failures.txt join both staging
lists. Docs updated (README two-flow section, sources.config two-surface
header, overlay section); .gitignore learns .zig-cache/ and zig-pkg/ (zig
0.16 extracts fetched packages into zig-pkg/ beside the manifest).
Verified (fossil flow): full pipeline PASS - suite_smoke static+zip, tklib
tooltip:2.0.4 under withdrawn Tk, tcllib md5/sha1/cmdline, tcllibc:2.0 with
md5-critcl-accel:1; 'suite.tcl test -testargs {-file http.test}' delegates to
the gate step (528/519/9/0 GATE PASS).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Acceptance verified in full (evidence in the detail file's Progress
section): characterization suites pass with one recorded pin flip,
lifecycle.test pins batch/OO field equivalence (-targets_cksums incl)
and o_record_list coherence across sequential events, the retired-trio
grep is clean, and a full make.tcl build completes with skip/copy
decisions unchanged against pre-refactor .punkcheck state.
Index entry moved to GOALS-archive.md; reference sweep: G-095 notes gain
the chokepoint-achieved pointer, the transient concurrent-save evidence
observed during verification, and the inherited MAX_PATH rider pointer
(G-087/G-094, unaddressed); G-027's punkcheck note updated to achieved
tense. Detail file moves to goals/archive/ in the adjacent commit (pure
rename per the doc-restructure rules).
punkproject.toml 0.17.2 + CHANGELOG entry (punkcheck ships in kits; the
in-shell punkcheck API surface changed with the retirement).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
