Tree: 59bdb1ec49

master

v0.1a

Branches Tags

${ item.name }

Create tag ${ searchTerm }

Create branch ${ searchTerm }

from '59bdb1ec49'

${ noResults }

Commit Graph

932 Commits (59bdb1ec49bb6fe4bcf1b9a72b0cfa521e94270f)

Author	SHA1	Message	Date
Julian Noble	768dab399f	G-033: proj: package-mode scope prefix - visitor binary resolves dev/src against the cwd's project ... punk_main.tcl: an optional proj: prefix on the package-mode string (e.g 'proj:internal-src') rebinds the dev/src path blocks to the punk project containing the current directory instead of the executable's own project. Root discovery is ::punkboot::proj_root_find, a lean boot mirror of punk::repo::find_project/is_project_root in Tcl builtins only (git/fossil control markers via 'file exists', unwise- path exclusion, ./src plus src/modules|src/vfs|src/scriptapps| punkproject.toml candidate check; .git accepted as file OR dir for worktree checkouts - recorded divergence). The proc stays defined post-boot like the G-129 mount procs. The prefix sits outside the ordered dash-list, whose block order remains the same-version tie-break dial; an invalid mode remainder is not consumed (whole arg falls through as subcommand/script as today); bare 'proj:' reads as internal and warns. Discovery outcome is always reported on stderr (detected root + effective precedence - never a silent rebind), a failed walk-up warns and leaves dev/src unbound (no exe-relative fallback rebind), and proj: without a dev/src block warns it has no effect. Under proj:, dev resolves <root>/modules{,_tclN} and <root>/lib{,_tclN} only; src feeds the discovered root into the existing src-mode machinery unchanged (tm paths, auto_path, package prefer latest, inline #modpod scanner). Exe-relative src/dev launches are untouched. punk::args::moduledoc::punkexe 0.1.1: the packagemode help text drafted in the goal file is now live as the packagemode leader of (script)::punkexe (@leaders max 2), with proj: wording in the @cmd overview; rendering/parse wiring remains G-032. Verified on a rebaked punk91 kit, standalone copy outside any source tree: proj:internal-src from tomlish (fossil root, builds no executable) loads tomlish 999999.0a1.0 from its src/modules with kit-first ties; proj:src flips the ordering; from punkshell the visited-root modpod scan registers 6 modpods; no-project and no-root-using-block warnings fire and proceed; bin/punk91 src unchanged. Docs: bin/AGENTS.md launch-package-modes proj: contract, src/vfs/AGENTS.md punk_main contract line, ARCHITECTURE.md entry point. punkshell 0.29.0 (minor - new launch invocation) + CHANGELOG. Claude-Session: https://claude.ai/code/session_0156PuejSCGjgeGb7jiABrDU Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	3 days ago
Julian Noble	faa459f572	punkzip PROVENANCE.md: relative sibling-checkout convention replaces the dev-machine absolute path ... The Origin section now names the upstream by its remote and expects the working checkout at ../zig/punkzip relative to the punkshell repo root, stated as a layout convention with its fallback spelled out (clone the remote there; the pinned Vendored state commit identifies the exact upstream state in any clone). The re-vendor procedure now runs entirely from the punkshell repo root (git -C ../zig/punkzip checks, relative cp sources), making it copy-pasteable on any machine reproducing the sibling layout and independent of the punkshell checkout's own directory name (survives the G-022 rename). PROVENANCE.md was the only operational document carrying the machine-local path - the c:/repo/jn absolute references in goals, archives and CHANGELOG are historical narrative and stay as-is. G-135 names this file's prose as the src/tools vendored-tree convention, so the relative form is set while there is one instance to copy. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	3 days ago
Julian Noble	2c1dec100e	CRLF hygiene ahead of the fossil catch-up: glob the vendored sdx patches, LF two strays ... The catch-up checkin aborted on CRLF in files outside crlf-glob. Five real offenders, in two classes needing opposite treatment: - src/buildsuites/suite_tcl86/patches/* (three files) is vendored sdx source by Matt Newman and Jean-Claude Wippler plus the patch derived from it. Upstream CRLF, and the patch must stay byte-exact to apply. Added to crlf-glob, which is what .fossil-settings/AGENTS.md prescribes for a payload tree that trips the prompt - explicitly in preference to reaching for --no-warnings. - src/modules/oolib-buildversion.txt and src/buildsuites/suite_tcl86/src/main.zig are authored punkshell files (4 lines each, fully CRLF) that the root AGENTS.md LF preference governs. Converted rather than globbed; globbing src/modules would suppress the warning across authored territory. punkzip's hamlet.txt was a false positive in the first scan - already covered by the crlf-glob entry for its fixtures directory, where it is deliberately CRLF byte-exact test data. Converted in git first so the fossil checkin records identical content; a content fix applied only fossil-side would desynchronise the two mirrors in the middle of a catch-up. Note oolib-buildversion.txt is build-generated, so a future regeneration may reintroduce CRLF there. Claude-Session: https://claude.ai/code/session_01YNjnq6oDzecknLg7AuWZgU Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	4 days ago
Julian Noble	03306be068	argparsingtest 1.3.0: opt, cmdline and tepam are optional, not module-level requires ... The module hard-required opt, cmdline and tepam at load time, so on a punkshell build without tcllib 'package require argparsingtest' died outright with "can't find package opt". A harness whose whole job is comparing parser libraries is useless precisely on the runtime you wanted to measure, and the module already had the right pattern for this: argp, parse_args and argparse degrade to a not_loaded row rather than taking the module down with them. The three tcllib parsers were the odd ones out. Reproduced before the change with punk module paths only and an emptied auto_path, which is where tcllib's pkgIndex-based packages come from. Every parser this module compares is now optional. Their roster entries become {optional opt}, {optional cmdline} and {optional tepam}, so discover_parsers reports not_loaded when the package is absent and nothing in the harness calls them. punk::args and Tcl 8.6- remain hard requires - the harness itself is built on those. The guard differs by package because the dependency does: cmdline is only ever called INSIDE proc bodies, so the catch-guarded require at the top is the whole story - the wrapper procs stay defined and are simply never invoked. opt and tepam are used at DEFINITION time: tcl::OptProc and tepam::procedure are the commands that create the wrapper procs. Their seven definition sites (four opt, across opts/tkstyle/tclstyle/sandwich; three tepam) are each additionally guarded, so with the package absent the proc does not exist at all. Those sites test 'package present' rather than wrapping the definition in a bare catch. Both tolerate absence, but a presence test still lets a genuine error in a spec raise loudly on a runtime that does have the package, where a catch would swallow it. The pre-existing argp registration site keeps its original catch idiom - working code, not part of this change. Pinned by module_loads_without_optional_parser_packages: a child interp with an emptied auto_path that asserts the module loads, that each absent package's parser reports not_loaded, and that the opt and tepam wrapper procs genuinely do not exist. Where a runtime supplies one of the three as a .tm on a module path the test degrades to asserting the module still loads, which is the main claim anyway. Tests that name parsers explicitly (first_call_library_backed, first_call_positional_styles, sandwich_strictness) now filter through a loaded_parsers helper, and the tclstyle cmdline_typed status assertion derives its expectation from package presence: those parsers can now legitimately be missing. Verification: 26/26 pass under the 32-bit luck-test2 runtime with no skips, 24 pass + 2 skip under native Tcl 9.0.3. With opt and tepam hidden, the built 1.3.0 module loads, reports both as not_loaded, and compare lists them in its "not loaded:" footer instead of the whole module failing to load. Claude-Session: https://claude.ai/code/session_01Q82VuBXzgeKsAwMYHrbpP1 Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	4 days ago
Julian Noble	7280739204	Project icon: punk1.ico -> punkshell.ico, artwork from the src/assets/logo master ... The four build-consumed copies are renamed and their artwork replaced with src/assets/logo/punk-mark.ico (all four byte-identical to the master): src/runtime/punk1.ico -> src/runtime/punkshell.ico src/vfs/_vfscommon.vfs/punk1.ico -> .../punkshell.ico src/vfs/mkzipfix.vfs/punk1.ico -> .../punkshell.ico src/vfs/punk8_statictwapi.vfs/punk1.ico -> .../punkshell.ico Recorded as delete+add rather than renames: the content is entirely different, so git's similarity detection finds no link and --follow will not trace through. The pre-2026-07 artwork keeps its own lineage under src/assets/logo/legacy/, now the only place the punk1 name survives. No code referenced the old name - every mention was documentation or goal prose, checked before renaming. G-057's contract lines are updated for the new path (Scope in both tiers, plus Goal and Acceptance): pure path substitution, no semantics changed. Its Context gains a dated note recording the rename and artwork swap, and its _vfscommon.vfs override-detection wrinkle now names punkshell.ico. goals_lint clean. Adds src/assets/logo/project-default/, a frozen monochrome placeholder - a blank screen wearing the crest - held as the single copy for seeding generated projects. Unbranded so it reads "replace me" rather than as a product identity. Bezel and crest are ONE union path (drawn as two shapes the bezel's top edge cuts a visible line across the mohawk) and the screen is filled rather than outlined (a solid mass survives 16px where nested thin strokes merge). It shares only the crest with punk-mark.svg and must not be re-derived from it: tracking the real mark would churn every generated project on every logo tweak. Also corrects a defect introduced in 91fa500e. The Ownership bullet in src/assets/logo/AGENTS.md said "do not repoint build consumers at this directory on your own initiative" and then, in the next sentence, instructed re-copying them by hand after regeneration - so whichever half an agent read first would win. It now states the dated fact and names G-057 as the owner of future propagation. Distribution machinery does not belong in an AGENTS instruction, and drift wants a check rather than a standing manual obligation. Stale punk1.ico copies were also swept from the untracked build trees (src/_build x12, bin/test, bin/test2, scratch, testsdx) so the next bake cannot merge both names into a kit payload. Claude-Session: https://claude.ai/code/session_01YNjnq6oDzecknLg7AuWZgU Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	4 days ago
Julian Noble	91fa500e24	Project logo: two-colour bracket mark, master + web set (src/assets/logo) ... Replaces the busy pixel-art punk1 artwork - unreadable below ~32px - with a flat two-colour mark: a mohawked terminal plate enclosing Tcl's [ ] command substitution, Emerald #0FA36B on ink #03120C. Drawn on a 64x64 grid to hold at 16px and verified by rendering at true pixel size on both light and dark plates, rather than judging it large and hoping. src/assets/logo/ is the icon MASTER, distinct from the build-consumed copies at src/runtime/punk1.ico and src/vfs/_vfscommon.vfs/punk1.ico. Those are deliberately untouched and remain a user-directed swap. punk-mark.svg/.ico outline plate (primary) alternative/punk-mark-solid.* solid plate web/ favicon set legacy/ retired punk1 artwork, moved from repo root make-ico.ps1 keeps the .ico reproducible: rsvg-convert for true-alpha rasterizing (never browser screenshots - they bake in a background and lose the alpha the icon depends on), then a hand-assembled ICONDIR. PE icons carry 16/24/32/48/64/128 as uncompressed BGRA DIBs plus a PNG-compressed 256, keeping that entry ~5KB instead of ~265KB - each entry becomes a separate RT_ICON in every embedding kit. Regeneration is deterministic, byte-identical across runs. The web set inverts those tradeoffs. favicon.ico is 16/32/48 all-PNG (2.5KB, fetched on every cold page load) and is web-only: all-PNG is precisely what GDI+ and tklib's ico package cannot read. web/icon.svg carries a prefers-color-scheme block serving the outline plate to light tab bars and the solid plate to dark - something an .ico structurally cannot do - verified by rendering under forced dark. The apple-touch and PWA rasters are deliberately opaque, since iOS ignores transparency and composites against black. Two decoder limits are recorded in the new AGENTS.md as properties of the readers, not defects in the files: legacy GDI+ silently returns the 128px frame when asked for 256 (WIC, which Explorer and the taskbar use, reads all seven), and tklib's ico package parses DIB entries only. Root and src AGENTS.md gain Child DOX Index entries. Both converted to LF per the root line-ending policy, which explicitly declines diff-minimisation as a reason to preserve CRLF. No punkproject.toml bump: these assets are a non-shipped surface until a build consumer is repointed. Fossil still tracks the three legacy files at their old root paths and needs the rename at the next catch-up checkin. Claude-Session: https://claude.ai/code/session_01YNjnq6oDzecknLg7AuWZgU Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	4 days ago
Julian Noble	53c1b30745	make.tcl-generated outputs: bootsupport punk::zip 0.3.0 promotion ... Bootsupport promotion of the G-126 accelerator module (superseded 0.2.0 pruned by the punkcheck-managed step). make.tcl's zipfs-less bake extraction path now carries the accelerator-capable punk::zip. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	4 days ago
Julian Noble	df6642e189	punk::zip 0.3.0: punkzip extraction accelerator + parity suite (G-126); 0.28.4 ... The final G-126 acceptance clauses: - punk::zip::accelerator (auto|none|<path>): auto probes env(PUNKZIP_EXE) then a punkzip beside [info nameofexecutable]; resolution cached until reconfigured - unzip hands whole-archive member WRITING to the accelerator only when the call is one it serves identically (globs {*}, no excludes, default -overwrite/-mtime/-verify, ascii member names) and re-stamps mtimes with punk::zip's local-time convention afterwards, so the two engines produce identical trees; preflight refusals, member selection and the returned names always come from punk::zip's own reader (the G-124 floor, unchanged); accelerator failure falls back to pure Tcl silently; last_unzip_engine/last_accelerator_note expose the per-call decision - parity suite zipaccel.test (9 tests): identical names, bytes and mtimes through both engines over the G-124 shapes and whole-kit extraction of the file-relative tclsh90b4_piperepl.exe (841 members); listings engine-independent; punkzip porcelain agrees with punk::zip member dicts on every shared field (mtime excluded by design - UTC vs local convention); pure path proven when the binary is absent; selective calls proven to run pure. Implicit directories (kits store no dir entries) are excluded from mtime comparison - neither engine stamps them - recorded benchmark (in the goal detail): accelerator fixed cost ~9-13ms, crossover ~15-20 members, material (>= x1.7) from ~50; x2.3-2.6 at kit scale under a native tclsh and x3.8-4.9 kit-hosted (punk91.exe 21.9s -> 5.7s) - the bake-from-kit case the goal motivated - full runtests at recorded baseline under tclsh9.0.5-punk: 1186/1211, 1 failed = the pre-existing exec-14.3 baseline Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	4 days ago
Julian Noble	25b3dcc667	make.tcl-generated outputs: layout sync copies of the tool subcommand ... Thin-layout sync (part of the bootsupport run): the project-layout make.tcl copies (vendor/punk/basic, vendor/punk/project-0.1, and the punk::mix::templates modpod source tree) pick up the G-126 'tool' subcommand from d76a7907 via the punkcheck-managed sync step. No hand edits. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	4 days ago
Julian Noble	91d3594570	punkcheck 0.6.1: installtrack self-heals damaged persisted event history ... Field failure (2026-07-27, second dev machine): 'make.tcl bake punk9_beta' died in the vfslibs phase with eventcollection ... key '$eventid' already exists in collection because that machine's src/vfs/.punkcheck predates punkcheck 0.3.2. The 0.3.2 note fixed the WRITER (a braced template wrote the literal strings '$eventid' etc into event headers) but files written before it keep the damage, and the modern installtrack constructor - which reloads persisted events into a collection keyed by -id - collided on the duplicate literals and aborted the whole bake. - punkcheck 0.6.1: the constructor reload now self-heals instead of aborting: an EVENT record with a missing or duplicate -id is reconstructed under a synthesized unique id ('damaged-id-...') with a stderr warning naming the file and the offending id. Synthesized ids match no INSTALL record's -eventid reference, so healed events age out through normal -keep_events pruning. Reload also tolerates missing -source/-targets keys. - oolib 0.1.4: the collection duplicate-key error no longer misnames the object as 'col_processors' (stray copy-paste in the generic class). - regression test installtrack_damaged_event_history_selfheal seeds a real two-event file, rewrites both -id values to the braced literal {$eventid} (as the damaged files carry it - the record loader evaluates records, so only a braced literal survives as the string), and asserts the reload succeeds with both events kept under distinct keys (one healed). Suite: 30/30. - before/after proof against the released artifacts: punkcheck 0.6.0 (exact-pinned) reproduces the field abort on the crafted file; 0.6.1 heals it. Bootsupport promoted to punkcheck-0.6.1 + oolib-0.1.4 (superseded copies pruned), so make.tcl bakes get the fix. Remedy already applied on the affected machine (file deletion); with this fix, remaining old .punkcheck files anywhere self-heal with a warning instead of failing builds. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	4 days ago
Julian Noble	d76a7907ea	make.tcl tool subcommand: build vendored zig tools into bin/ (G-126); 0.28.3 ... New surface for the vendored first-party build tools under src/tools: make.tcl tool list|info|build|test ?<toolname> ...? ?-test 0|1? - discovery is convention-based (a src/tools directory carrying build.zig is a tool), so the G-128 tool will appear with zero make.tcl edits; per-tool records are read tolerantly from build.zig.zon (version, minimum_zig_version floor) and PROVENANCE.md (upstream, vendored commit) - list/info report installed-binary state (current|stale|absent vs bin/<name><exe-suffix>) and the resolved toolchain - toolchain resolution: PUNK_ZIG=<path> overrides; otherwise bin/tools/zig* is scanned and the LOWEST release satisfying the floor wins - deterministic as newer toolchains get unpacked beside the pinned one; dev builds count only when their base version exceeds the floor - build runs the tree's own 'zig build test' as an EXIT-CODE gate (expected stderr warnings do not fail it), then builds ReleaseSafe and installs to bin/; nothing installs when the gate or build fails; -test 0 skips the gate; build caches are deliberately left in place between runs for rebuild speed (git- and fossil-ignored) - zig stays OPTIONAL: packages/bake never require the step; without a suitable toolchain, list reports the state and build/test exit nonzero with bin/punk-getzig.cmd guidance - full subcommand accompaniments per src/AGENTS.md: punk::args argdoc (SUMMARIES/HELPTEXTS/bespoke passthrough definition), SUBGROUPS ("vendored tools"), known_commands, degraded-mode dispatch, plain help text, bootsupport-staleness exemption, workflow text (new OUT-OF-BAND SUBSYSTEMS section also covering buildsuite), project version 0.28.3 + CHANGELOG entry - piped characterization (zig-independent): src/tests/shell/testsuites/punkexe/maketcltool.test - 4/4 green - verified live: 'tool build punkzip' gated on the vendored tree's 130/130 suite and installed bin/punkzip.exe (ReleaseSafe, v2.3.1, 1.1MB) over the stale Nov-2024 v2.1.0 binary; list state flipped stale -> current; PUNKBOOT_PLAIN degraded dispatch and tabled help verified G-126 Progress updated: the acceptance's build-step clause is satisfied. Remaining: punk::zip fast path, parity suite, benchmark. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	4 days ago
Julian Noble	1cdfb88693	G-126: vendor punkzip v2.3.1 into src/tools/punkzip/ (upstream 0882f037) ... Vendored per the goal's developer decision: build.zig, build.zig.zon, LICENSE + LICENSE-time-dot-zig, README.md and src/ (sources and test fixtures) from the clean upstream checkout at commit 0882f0373eb2c983142e713c814b91aa82cfcea7 (punkzip v2.3.1). Upstream doc/ (article mirrors), reference/ (hwzip 2.4 C source) and repo-local agent/porting notes are deliberately not vendored. - src/tools/punkzip/PROVENANCE.md: origin, vendored commit, re-vendor procedure (clean-upstream-only), G-063-style licence records (public-domain hwzip lineage, MIT musl time.zig, zero dependencies) - src/tools/AGENTS.md: new DOX child - vendored-tools boundary, upstream-first editing policy, pinned-toolchain verification; indexed from src/AGENTS.md - .fossil-settings/crlf-glob: cover the vendored fixtures (hamlet.txt is CRLF byte-exact test data the compression tests pin sizes against; git stores it verbatim - core.autocrlf false) - verified from the vendored location with the pinned bin/tools/zig-x86_64-windows-0.16.0 toolchain: zig build test 25/25 steps, 130/130 tests green; build outputs (.zig-cache/, zig-out/) were already gitignored at any depth Goal Progress updated: the acceptance's vendoring clause (origin, state, re-vendor procedure, licensing) is satisfied. Next: the make.tcl tool build step. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	4 days ago
Julian Noble	d2f7dc6f0e	make.tcl-generated outputs: punkboot::utils 0.5.0 propagation (bootsupport + _vfscommon) + layout make.tcl sync ... Batched punkcheck-managed build outputs from make.tcl modules/bootsupport/ vfscommonupdate for the G-133 change-set: bootsupport and _vfscommon.vfs gain utils-0.5.0.tm (0.4.0 pruned), thin-layout make.tcl copies and the modpod templates payload re-synced. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	4 days ago
Julian Noble	583dd5e0b2	G-133: bake payload/target consistency checks - advisory arch scan + smoke-require probe ... - punkboot::utils 0.5.0: binary_arch_classify (PE Machine field, ELF e_machine, Mach-O thin+universal - honest unknowns), platform_expected_binary, platform_discriminated_segment (canonical <os>-<cpu> tokens + vendor spellings, extensible namespace variables), vfs_binary_arch_report (per-directory exemption, exempt subtrees counted unread) - make.tcl: advisory binary-arch scan at the G-125 gate seam (recapped BUILD-WARNINGs naming file/found-arch/kit-target, capped at 8 per kit with honest total; cross-target kits included); post-build smoke-require probe running the freshly built artifact via its tclsh subcommand with drained stdin (failures recapped naming kit/package/actual error; cross-target skip with stated reason; undeclared kits run nothing new); mapvfs.config entries accept a 5th smoke-require element (mapvfs_parse/mapvfs_kit_outputs carry smokerequire); 'check' reports scan ACTIVE/UNAVAILABLE + declared smoke matrix; workflow K11 - mapvfs.config: smoke-requires declared - punkluck86 {Thread} (the 2026-07-27 incident construction), punk91ix86 {Thread iocp}, punkshell902 {Thread} (cross-target skip demonstrator) - tests: binaryarch.test (generated header fixtures only - no committed binaries; punkluck86 case reproduced; iocp pair + win-x64 exemptions; real-tree sweep with known-real findings filtered), maketclpayloadcheck.test (piped check characterization, ESC-free per G-113) - docs: src/AGENTS.md + bin/AGENTS.md state what the checks do and do NOT guarantee; src/runtime/AGENTS.md entry grammar; ARCHITECTURE.md bake section - REAL FINDING on the scan's first sweep: zint.dll 2.13.0 in punk8win.vfs lib_tcl8/ is 32-bit (PE i386, confirmed with file(1)) and can never load in the x64 tcl8 kits punk86/punkbi/punksys that carry it - those bakes warn until the payload is fixed - punkshell 0.28.2 Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	4 days ago
Julian Noble	f4d1091d61	win32-ix86 verification kits repointed at minimal payloads punk8min.vfs / punk9min.vfs ... The full punk8win/punk9win_for_tkruntime payloads shipped x64-only binary packages into the 32-bit kits (user-observed): a higher-versioned x64 Thread 2.8.12 shadowed the LUCK runtime's working 32-bit 2.8.5 on plain package require and broke the repl (tsv machinery), and x64-only dlls (tcludp etc) failed with "couldn't load from current filesystem". Measured: forcing -exact Thread 2.8.5 loaded fine, proving the backport's zipfs dll loading works and the failure was purely wrong-arch version shadowing. New minimal payloads: both startup fauxlinks (root main.tcl + the undroidwish app/main.tcl hook) plus directory fauxlinks to MULTI-ARCH packages only - those carrying a win32-ix86 platform subdir: iocp-2.0.2 (both majors, from punk8win.vfs / punk9magicsplat.vfs) and twapi-5.0b1 (tcl8, from punk86.vfs). No binary duplication in VCS (G-004); the pure-tcl module/lib set arrives via the dll-free _vfscommon.vfs overlay. Verified: both kits boot, resolve punk::args from _vfscommon, load Thread (2.8.5 own-arch on LUCK / 3.0.5 static on sfe), iocp 2.0.2 (ix86 dll), twapi (4.7.2 stable beats the linked 5.0b1 beta under prefer-stable on LUCK; 5.2.0 static on sfe), udp 1.0.11 on LUCK (own archive copy, no longer shadowed); piped shell sessions evaluate cleanly on both with zero tsv-unavailable noise. Remaining specimen quirk (out of payload scope): punkluck86 segfaults at process exit AFTER a completed shell session (exit 139) - repl-teardown x 2019-era runtime interaction; the tclsh subcommand exits 0 with all four packages loaded. punkexe subtree 69 tests 0 fail. Claude-Session: https://claude.ai/code/session_01UEgomWq6kA6c4A8GswqqGW Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	4 days ago
Julian Noble	191d8ff490	G-129: kit boot derives its payload mount from the runtime's mount table (punkshell 0.28.0) ... Boot (punk_main.tcl + project_main.tcl, identical copies): - zipfs presence keys on tcl::zipfs::mount - present in every supported zipfs generation - not tcl::zipfs::root (8.7/9-era; absent from the androwish/ undroidwish 8.6 backport, whose ::zipfs ensemble also lacks a root subcommand). - New factored ::punkboot procs (zipfs_mount_pairs, zipfs_kit_mountbase, path_within) derive where this executable's attached archive actually mounted: the mount-table entry whose archive file is [info nameofexecutable] (the backport reports it without the windows drive prefix - handled), fallback the mount containing the boot script. //zipfs:/app stays the expected modern answer, not an assumption. Procs remain defined post-boot for tests. - A mounted archive that cannot be attributed to the executable is reported on stderr instead of silently configuring no internal paths (the old failure shape). - The starkit quick-search line [zipfs root]/app/tcl_library (an error on the backport) uses the derived base; internal_paths gains the derived base when no zipfs volume root covers it; the internal-only simulate-kit branch recognises a script under the derived mount as kit-internal. make.tcl (necessary companion fixes - Scope extension flagged in the goal file): - Runtime capability probe keys has_zipfs on tcl::zipfs::mount (was root, which refused zip assembly for the backport; the block also rode a body-less 'if {![catch ..]}' that only worked because piped-stdin tclsh continues past command errors). - merge_over resolves relative fauxlink targets against the link file's own directory (the fauxlink module's symlink semantics) - identical for the pre-existing root-level links, required for the new nested one. Payload + matrix: - punk8win.vfs gains an app/main.tcl fauxlink beside root main.tcl: the undroidwish family auto-runs app/main.tcl INSIDE the attached archive (embedded /app/main.tcl string in the exe) and ignores a root main.tcl. Inert in the metakit kits sharing the payload (punksys/punk86/punkbi verified). - mapvfs.config: punkluck86 (tclsh8.6.10-luck-zip) + punk91ix86 (tclsh9.1b0-tclsfe), both zip kits targeting win32-ix86. Tests + docs: - New shell/testsuites/punkexe/kitmountpoint.test (4 tests): backport exe-path attribution, modern-32 //zipfs:/app, baked punkexe regression pin, unattributed-mount stderr report + no foreign paths. Self-gates on the untracked verification runtimes. - bin/AGENTS.md "Kit-wrappable runtime requirements (G-129)" + src/AGENTS.md pointer, src/vfs/AGENTS.md boot contract, src/tests/shell/AGENTS.md suite doc, ARCHITECTURE.md boot-chain bullet. - Thin-layout + templates-modpod copies synced (make.tcl modules; _config boot mains refreshed into the project-0.1 store layout per the 17250a8c precedent). Verified: punkluck86 + punk91ix86 boot and resolve punk::args from inside their mounts with externals stripped; punk905/punk902z (zipfs) + punksys/punk86 (starkit) probe byte-identical before/after; both zip offset conventions (mkimg + concat) mount on the backport; full suite 1171 tests with only the documented exec-14.3 baseline failure. Claude-Session: https://claude.ai/code/session_01UEgomWq6kA6c4A8GswqqGW Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	5 days ago
Julian Noble	8a2e216490	build outputs: mapped-file replace helpers into bootsupport, _vfscommon and the thin layouts ... Regenerated by 'make.tcl modules' + 'bootsupport' + 'vfscommonupdate'. The thin-layout make.tcl copies and the templates modpod carry ::punkboot::replace_possibly_mapped_file and ::punkboot::mapped_file_hint, so generated projects get the same behaviour - a project whose own bootsupport carries a zip-based .tm meets exactly the same windows constraint. Note the templates modpod copy in this commit was itself replaced by the new delete-then- place path: the bootsupport run reported "in-place overwrite refused - replaced by delete-then-place" and completed, where the previous three runs this session failed and needed a hand copy. The fix was exercised by the change-set that introduced it. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	df83b21a24	bootsupport: replace mounted modpods by delete-then-place, and explain the windows catch-all (punkshell 0.27.4) ... 'make.tcl bootsupport' failed three times this session replacing the zip-based punk::mix::templates modpod, always with: error copying "<src>" to "<dst>": invalid argument Root cause, demonstrated rather than inferred. A zipfs mount MEMORY-MAPS its archive (tclZipfs.c CreateFileMappingW + MapViewOfFile), and windows refuses to overwrite a file with a user-mapped section open - ERROR_USER_MAPPED_FILE (1224). Any holder blocks it: a running punk shell that loaded the modpod, a src-mode session, or the build itself. Reproduced directly: the same file copies fine unmounted, fails while zipfs-mounted, and copies fine again once unmounted. The message is a dead end BY CONSTRUCTION, which is why two earlier attempts at characterising this (in commit messages and in the archived G-124 detail file - first "a modpod make.tcl has itself mounted", then "a transient lock") were both wrong in different ways. Tcl_WinConvertError maps only Win32 codes 0..267 and sends everything above them to errorTable[1], which is EINVAL - so EVERY high-numbered windows error surfaces as "invalid argument", naming neither cause nor remedy. Byte-identical code in 8.6 and 9.1b1. FIX. ::punkboot::replace_possibly_mapped_file tries the ordinary copy and, on failure, 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 (measured: delete succeeds while mounted, the mount keeps serving its old content afterwards). Replacement content is staged to a sibling <target>.punkboot-new first, so a mid-sequence failure can never leave the target missing; a 'broken' return - unlinked and unrestorable - is reported as a build failure naming the file to restore by hand. This replaces a narrower prior workaround that only recognised the already-identical case and otherwise recorded FAILED. DIAGNOSIS. ::punkboot::mapped_file_hint fires when a genuine failure carries the catch-all message and explains what it actually means and which process class to look for, instead of leaving "invalid argument" bare. It stays silent for unrelated errors. Verified on the real failure: the bootsupport run that has failed all session now reports "(in-place overwrite refused - replaced by delete-then-place; target was memory-mapped, e.g a mounted modpod)" and completes, with no hand-copy and no rerun. Helper-level checks cover the unmapped path (plain copy, unchanged), the mapped path (recovers, content correct, holder's mapping intact, no temp left behind) and both hint branches. Full suite 1167 tests, 1146 passed, 1 failure - core/tcl exec-14.3, the documented baseline. Documented in src/bootsupport/AGENTS.md, including the general warning that 'invalid argument' from ANY windows file operation in Tcl means "some windows error above 267" and must never be read at face value. Also drafts G-132 (user-approved): the errno catch-all is upstream's defect, not ours, and deserves reporting. The goal is deliberately shaped so an AGENT DRAFTS AND THE DEVELOPER SUBMITS - no account use, no web form, no API - following the G-039 precedent, with the submission-ready text staged in TEMP_REFERENCE and the finding, reproducer, eventual ticket URL and disposition kept in the tracked detail file, because TEMP_REFERENCE is one 'git clean -xdf' from gone. The report's claim is deliberately narrow: the errno conversion is plainly wrong, while the related fact that 'file copy -force' over a mapped file succeeds on unix and fails on windows rides along as context rather than as a second demand. punkshell's workaround is explicitly not removed if upstream fixes it - an upstream fix dates it rather than deletes it. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	176f2ad702	punk-runtime list/use stop offering support files as runtimes (punkshell 0.27.3) ... 'bin/punk-runtime.cmd list -platform win32-ix86' listed a LUCK.url shortcut and a tclsh8.6.10-luck-zip.cfg build request as if they were selectable runtimes. Both payloads already filtered support files - .txt/.toml/.tm/.tmp/.log - so the mechanism existed and only the recognised set was too narrow. It was also duplicated across NINE filter sites in three syntaxes (a powershell array, a bash case pattern, a bash grep -E regex), and two of them had already drifted: the "not a selectable runtime" guard omitted .log and matched *_BUILDCOPY without a trailing wildcard. The set is now defined ONCE per payload - $PunkRuntimeSupportExt / Test-PunkRuntimeSupportName in the ps1, punk_support_ext_re / is_punk_support_name in the bash - and every site consults it, so the two payloads cannot drift apart. Their listings are required to agree byte-for-byte and now do so by construction. Recognised support files, by category: docs/text (.txt .md .html .htm .pdf .url), config/metadata (.toml .json .xml .yml .yaml .ini .cfg), dev leftovers (.tm .tmp .log .bak .old), checksums/signatures (.sha1 .sha256 .asc .sig) and archives (.zip .tar .gz .bz2 .xz .7z), plus any *_BUILDCOPY* name. Anything else is still treated as a runtime candidate - deliberately, so a stray binary is visible rather than silently hidden. Verified: the win32-ix86 listing drops from 6 entries to the 4 real runtimes; the bash helpers produce the identical four (the polyglot routes msys bash to the ps1 payload on windows, so the bash side was exercised directly). Re-wrapped per bin/AGENTS.md with checkfile ERROR-free; runtimecmd_roundtrip (byte-identity), runtimecmd_checkfile, runtimecmd_freshness (10/10, both payloads against the fixture server) and runtimebash_wsl (8/8) all pass, as does the full suite at its documented baseline. bin/runtime/README.md now states what a platform folder is FOR (a store tier keyed by the platform a runtime is for, not by this machine) and what is tolerated in it, with the recognised-support-file table and the warning that anything else will be offered as a runtime. It is force-added to git: /bin/* is ignored and a ! negation cannot re-include a file inside an excluded directory, so git add -f is the only mechanism - recorded in the .fossil-settings/AGENTS.md force-tracked one-off list, whose fossil counterpart is an explicit 'fossil add' at the next catch-up checkin. Also records in G-101 a survey of TEMP_REFERENCE/androwish for the zipfs-on-8.6 question the developer raised (non-contract findings; pointer added to TEMP_REFERENCE/AGENTS.md per its contract so the survey is not repeated). The undroidwish zipfs backport is ADDITIVE - generic/zipfs.c (6586 lines) + zipfs.h + a library/zipfs1.0 dir + one object line in the win and unix Makefiles - so it is carryable as a suite patch rather than a fork; it is C rather than C++ like mk4tcl; Tcl_StaticPackage registration is gated on ZIPFS_IN_TCL / ZIPFS_IN_TK, i.e. designed to compile into tcl or tk; and it is Christian Werner's, under the tree's Tcl-style license.terms. Its capability delta against Tcl 9 zipfs is recorded (no 'root', no ::zipfs ensemble, and not registered in child or thread interps), as is the exe-path mount convention that motivates G-129. Provenance of our specimen is established: LUCK is a CGI repacker that selects packages out of a prebuilt vanillatclsh binary's ZIP, so bin/runtime/win32-ix86/tclsh8.6.10-luck-zip.exe is a repack of androwish's own build - a working existence proof of zipfs-on-8.6 rather than a claim. Three open questions are recorded rather than guessed: whether it applies to 8.6.18 (the tree is 8.6.10), whether it builds under the pinned zig, and where the boot-time self-mount hook lives - stock tclAppInit calls Tclzipfs_Init only under #ifdef TCL_TEST, so it is in the vanillatclsh/undroidwish application main, which is also the hook a dual-container autodetecting boot would need. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	57b73d9f98	build outputs: punkboot::utils 0.4.0 + gate wording into bootsupport, _vfscommon and the thin layouts ... Regenerated by 'make.tcl modules' + 'bootsupport' + 'vfscommonupdate'. make.tcl reaches the predicate through a guarded require against the bootsupport snapshot, so the third tcl-library convention only takes effect for a bake once this lands; 'make.tcl check' confirms the gate ACTIVE and now names all three locations. The thin-layout make.tcl copies and the templates modpod carry the reworded K10 key note, so generated projects describe the gate the same way. The bootsupport templates modpod again failed to copy in place with 'invalid argument' and again succeeded on retry seconds later - the transient lock noted in the previous two build-output commits, not a make.tcl mount this time. Copied by hand, then 'make.tcl bootsupport' and 'vfscommonupdate' re-run to reconcile; all three copies share one sha1. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	20ee103120	boot-precondition gate learns the exe-path-mount tcl library convention (punkshell 0.27.2) ... The G-125 gate refuses a kit whose merged .vfs supplies no tcl library. It knew two conventions - tcl_library/ for zipfs-attached kits and lib/tcl<major>.<minor>/ for starkit-style ones - which covered all ten kit trees this project builds. A third-party runtime under evaluation (Lean Undroidwish Construction Kit, 32-bit windows, Tcl 8.6.10) uses a third: its attached archive mounts at the EXECUTABLE'S OWN PATH rather than at //zipfs:/app, so [info library] is <exe>/tcl8.6 and the library sits at tcl8.6/ in the root of the tree. The gate reported that runtime unbootable - a false refusal on a directory that is unmistakably a complete Tcl library (auto.tcl, clock.tcl, encoding/, history.tcl, init.tcl, package.tcl ...). Found while investigating whether our bake infrastructure can handle that kit; the androwish/undroidwish zipfs backport for 8.6 mounts this way generally, so it is a family of runtimes, not one artifact. punkboot::utils 0.3.0 -> 0.4.0 adds tcl<M>.<m>/ at the vfs root as a third recognised location; the report's 'checked' list and the not-found reason name all three, as do 'make.tcl check' and the workflow K10 key note. Glob note worth keeping: the root tcl[0-9]* pattern also matches the tcl8/ and tcl9/ MODULE trees this project's own kits carry. They hold no init.tcl, so the existing qualification test rejects them without needing a narrower pattern - now pinned by bootlib_module_tree_not_a_library so a later tightening of the glob has to argue with a test rather than look like an obvious cleanup. Verified: the LUCK runtime's extracted tree now reports ok=1 locations=tcl8.6, and all ten assembled src/_build/*.vfs trees still pass with unchanged detected locations (tcl_library for the six zipfs kits, lib/tcl8.6 for punk86/punkbi/punksys, lib/tcl9.0 for punkshell902). Full suite 1167 tests, 1146 passed, 1 failure - core/tcl exec-14.3, the documented baseline. One of the gate's own punkexe pins failed first time round and was right to: re-wrapping the 'check' report split "...that / cannot initialise" across lines. The pattern now matches across the break rather than the assertion being loosened. Note this does NOT make such a runtime bakeable yet - src/vfs/_config/punk_main.tcl still probes zipfs with 'info commands tcl::zipfs::root' (a 9-era command the 8.6 backport lacks) and assumes the //zipfs:/app mount point, so a kit built on it would not find its own payload. That is separate work; this change only stops the gate refusing a tree that has a perfectly good library. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	b549198cba	build outputs: punk::libunknown 0.2.3 into bootsupport and _vfscommon ... Regenerated by 'make.tcl modules' + 'bootsupport' + 'vfscommonupdate'. punkshell's own boot paths load punk::libunknown from bootsupport, so the pkgIndex namespace fix only reaches a running shell once the snapshot carries it. Two catch-ups ride along, neither authored here: argparsingtest 1.0.0 -> 1.2.0 in _vfscommon (from commit 4bd0ab13 - the previous change-set deliberately held this back while that module was still uncommitted work in progress), and the _vfscommon copy of the templates modpod, which was lagging the copy committed with the G-125 build outputs in de049fba. All three copies (modules/, bootsupport, _vfscommon) now share sha1 9115a3a37992. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	ec781d7019	punk::libunknown 0.2.3: pkgIndex.tcl scripts run in the global namespace again (punkshell 0.27.1) ... A third-party 8.6 kit under evaluation (Lean Undroidwish Construction Kit, 32-bit windows) could not load its own bundled twapi 4.7.2 inside punkshell: P% package require twapi invalid command name "twapi::set_scriptdir" ("package ifneeded twapi 4.7.2" script) twapi's pkgIndex.tcl defines 'namespace eval twapi { proc set_scriptdir ... }' when the INDEX is sourced, and its ifneeded script calls that command at require time. The command existed - as ::punk::libunknown::twapi::set_scriptdir. punk::libunknown 0.2.0 moved index sourcing out of 'namespace eval ::' into source_pkgindex, a proc of the punk::libunknown namespace, to stop each index's helper variables leaking into the global namespace and to stop a user global named 'dir' being clobbered. That fixed the frame, but a proc body executes in ITS OWN namespace, so an index's RELATIVE 'namespace eval foo' started creating ::punk::libunknown::foo instead of ::foo. Stock tclPkgUnknown is a proc in ::, so indexes have always had global resolution; the ifneeded script, evaluated later at global scope, then looked for a command that was never there. The body now runs via apply with :: as its namespace argument, which gives both properties at once: a local frame (index helper variables still discarded, user 'dir' still safe) and the global resolution context indexes are written against. Not twapi-specific: any index that defines commands or namespaces at source time for its own ifneeded scripts to use hits this, and the pattern is common in third-party indexes. punkshell's own vendored twapi 5.0b1 happens to avoid it (self-contained 'apply' ifneeded scripts), which is why the tree never surfaced it. Verified on the LUCK 8.6.10 runtime end-to-end - 'package require twapi' at the punkshell prompt now returns 4.7.2 with ::twapi::scriptdir correctly set - and pinned generically by pkgindex_relative_namespace_is_global in the discovery suite, whose new fixture copies twapi's pattern so no third-party kit is needed. The pin was proven to bite: driven against the pre-fix body the same fixture fails with 'invalid command name' and leaves ::punk::libunknown::pklu_nsindex behind; against 0.2.3 it resolves and leaves nothing. Full suite 1165 tests, 1144 passed, 1 failure - core/tcl exec-14.3, the documented baseline. Manual-version bump per src/modules/AGENTS.md: file renamed, Meta line, manpage_begin and provide-block updated, version-history line appended. The rename sweep converted the two goal Scope references (G-035, G-109) from the exact filename to the glob form src/modules/punk/libunknown-*.tm that GOALS.md's own Scope-authoring guidance recommends, so the reference stops drifting at every bump. Also records in G-035 the mechanism behind its "avoid mixing .tm and pkgIndex.tcl provision" folklore, which this investigation identified (non-contract Notes). Stock ::tcl:™️:UnknownHandler returns early when $satisfied is set and therefore never falls through to ::tclPkgUnknown, so a require satisfied from a .tm suppresses the whole library scan. Measured with a synthetic fixture on both 8.6.10 and 9.0.3: with mixpkg 1.0 as a .tm and mixpkg 2.0 in an auto_path library under 'package prefer latest', stock returns 1.0 and reports only 1.0 from 'package versions', runs no index side effect and registers no sibling package; the libunknown chain returns 2.0 and does all three. libunknown already neutralises it - its handler still computes $satisfied but the early return is commented out - and that unconditional fallthrough is exactly what the epoch cache pays for. The folklore is therefore substantiated for stock Tcl with a named failure mode (silent older-version selection) and neutralised under the punkshell chain, with stock behaviour still reachable in any interp that has not run punk::libunknown::init. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	de049fbad4	build outputs: punkboot::utils 0.3.0 + G-125 make.tcl into bootsupport, _vfscommon and the thin layouts ... Regenerated by 'make.tcl modules' + 'bootsupport' + 'vfscommonupdate' for the G-125 change. make.tcl loads punkboot::utils from bootsupport through a guarded require, so the boot-precondition gate only becomes ACTIVE once the snapshot carries vfs_boot_library_report - before this propagation 'make.tcl check' correctly reported the gate UNAVAILABLE. The thin-layout make.tcl copies and the templates modpod pick up the gate and the K10 workflow key note, so generated projects get the same refusal behaviour. The bootsupport templates modpod again could not be replaced in place - 'BOOTSUPPORT module update FAILED: ... templates-0.2.0.tm (invalid argument)'. Unlike the G-124 occurrence a plain tclsh failed on the first attempt too and succeeded on a retry moments later, which points at a transient lock by one of the running shells rather than at make.tcl having the file mounted. Copied by hand, then 'make.tcl bootsupport' re-run to reconcile the punkcheck record. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	b7c632b670	G-125 achieved: a kit that cannot boot is not deployed (punkshell 0.27.0) ... The kit machinery already knew when it had failed to extract anything from a source runtime, and under G-122 that notice became a recapped BUILD-WARNING - but the build still assembled an artifact, DELETED the deployed kit and copied the new one over it, closing running instances of the kit first. Field-observed 2026-07-26: an msys2-hosted bake of punk91 produced a 49,501,792-byte bin/punk91.exe that failed at startup with "application-specific initialization failed: Cannot find a usable init.tcl", by which point the previously working kit was already gone. G-124 removed the particular cause; this removes the class. A build tool should not make a working executable worse. A bake now refuses any kit whose merged .vfs supplies no tcl library: the kit 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. The gate sits immediately after the _vfscommon.vfs + <kit>.vfs merges and before the kit-type assembly switch, the last point at which no build product exists; a refusal appends to failed_kits, ends the punkcheck event FAILED and continues, so the punkcheck records stay consistent with what is on disk. The merged src/_build/<kit>.vfs tree is deliberately left behind - it is the evidence for the refusal. It gates on the BOOT PRECONDITION, not on the extraction step having run. Some .vfs folders legitimately supply their own tcl library (src/vfs/punk8_statictwapi.vfs, punk9test.vfs and mkzipfix.vfs do today), and those must keep building - so the check reads the merged tree. It is structural and executes nothing, which also makes it valid for cross-target kits this host could never run, and costs 0.63ms per kit (measured over 200 iterations against the real src/_build/punk905.exe.vfs), so it runs unconditionally. The predicate is punkboot::utils::vfs_boot_library_report (0.2.0 -> 0.3.0), called through a guarded require modelled on the existing get_src_provenance_warnings: a stale or missing bootsupport snapshot degrades the gate to a NOTE rather than failing every kit. Both branches were observed live - before the bootsupport propagation 'make.tcl check' reported UNAVAILABLE, after it ACTIVE. Scope was extended at activation (user-approved) to put the predicate in that module rather than private to the build script, because a gate that cannot be exercised cannot be characterized, and punkboot::utils is already where make.tcl's provenance gate helper lives. What qualifies as a tcl library: tcl_library/init.tcl (zipfs-attached kits) or lib/tcl<major>.<minor>/init.tcl (starkit-style), AND at least one companion file beside it (tm.tcl, package.tcl, auto.tcl, clock.tcl, history.tcl, word.tcl) or an encoding/ directory. The companion test is load-bearing: every punkshell kit carries lib/BWidget1.10.1/init.tcl, so a check that looked for any init.tcl anywhere would pass a kit with no tcl library at all. lib/tcllib2.0 is correctly not matched by the lib/tcl[0-9]* glob. Verified against all 10 assembled trees in src/_build - both conventions are represented (tcl_library for punk902z/punk905/punk91/punk9_beta/punk9bi_beta/punkmagic, lib/tcl8.6 for punk86/punkbi/punksys, lib/tcl9.0 for punkshell902). The no-extraction BUILD-WARNING was reworded from "will not initialise unless src/vfs/<x> supplies one" - a consequence the build then ignored - to "will FAIL the boot-precondition gate unless src/vfs/<x> supplies one". It stays a warning rather than becoming the failure, because a .vfs that supplies its own library needs no extraction. Verified 2026-07-27 end-to-end against a temporary fixture kit (a mapvfs entry pointing a zip kit at a payload-free runtime, plus a vfs with no tcl library), with sentinel files standing in for a previously deployed kit and a previous build product. Across the failing run bin/punkgatefixture.exe (sha1 2a0700eb...) and src/_build/punkgatefixture.exe (sha1 12c50567...) were byte-identical with unchanged mtimes, and the kit was reported under FAILED KITS with the cause named. Adding tcl_library/{init.tcl,tm.tcl} to that same fixture vfs - extraction still yielding nothing - made it build and deploy normally, which is the other half of the criterion. The fixture was removed afterwards and mapvfs.config restored byte-identical to its backup. Real-kit bake through the gate: punk905 builds, deploys and boots. Characterization: 12 tests in modules/punkboot/utils/testsuites/utils/bootlibrary.test (both conventions, the companion-file requirement and its BWidget decoy, near-miss reporting, missing/empty trees, and a sweep asserting every assembled src/_build/*.vfs tree still passes so the gate cannot fail kits that boot today) and 2 in shell/testsuites/punkexe/maketclbootgate.test (the ACTIVE/UNAVAILABLE report, ESC-free per the G-113 piped policy). Existing punkexe suites pass unchanged. Documented in src/AGENTS.md with the remedy when a kit is refused, in ARCHITECTURE.md beside the provenance gates, and in the embedded 'make.tcl workflow' data flow as key note K10 (verified under the default and PUNKBOOT_PLAIN=1 paths, within the 100-column budget). The achieved flip archives the entry and the detail file, and sweeps the live tier: G-127 (its output location inherits the never-written guarantee as long as the gate stays upstream of the artifact write, and its payload/target mismatch reporting should share this vocabulary), G-028 (the sibling "why bin/<kit> was not updated" surface in the same deploy step) and G-101 (a new container type must land its library where the gate looks, or extend the location list). Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	4bd0ab1354	argparsingtest 1.2.0: parse_args/argparse timing wrappers, plus a sandwich calling style ... compare's tables listed parse_args and argparse under "not loaded:" on the one runtime carrying both - bin/runtime/win32-ix86/tclsh8.6.10-luck-test2.exe (argparse 0.61, parse_args 0.5.1). Neither was missing. Both sat in optional_externals, the probe-only list whose entries can never be better than not_wrappable, and compare bucketed not_wrappable into the same footer line as not_loaded - so a package that was present and working read as absent. Timing wrappers now exist for both, and they move out of optional_externals into the style rosters under a new {optional <pkg>} marker that generalises the former hardcoded argp case: loaded when the package was present at module load, not_loaded otherwise. getopt and cmdr stay probe-only. compare now reports not_wrappable separately - a "no timing wrapper:" table line, and a not_wrappable key in the dict and json forms - so the two conditions cannot be conflated again. Availability is per-runtime and the two do not travel together (surveyed 2026-07-26): the 32-bit luck-test2 kit has both, the native Tcl 9.0.3 install has argparse only, punk902z and punk91 have parse_args only, punk905 has neither. The tests constrain them separately for that reason. Style support was measured rather than assumed. argparse reaches tk style only via -mixed (its default mode rejects leading positionals as excess arguments) and its switches need an explicit -argument in the element spec or their values fall through as positionals. parse_args cannot consume leading positionals at all, so tkstyle parse_args is unsupported. Both wrappers call their library by fully qualified name: the procs deliberately share the package name, so an unqualified call would resolve to the wrapper itself - infinite recursion in argparse's case. The fourth calling style, sandwich, is 2 mandatory leading positionals, THEN the 10 keyed options, THEN 3 mandatory trailing positionals. The 2/3 split is asymmetric on purpose - an even split could be satisfied by a parser that merely halves the positional block. All three candidates support a valueless flag, so the argvec exercises -join as a solo flag alongside the -flag value pairs. Wrappers: manual_switch, punkargs, punkargs_by_id, punkargs_parsecache, opt and argparse, returning {{p1 p2} {p3 p4 p5} optsdict} - grouped rather than the flat {p1 p2 opts} of the other positional styles, because positionals now sit on both sides of the options. punk::args and opt model that arrangement - @leaders/@opts/@values are three separate sections, and tcl::OptProc fills positional slots in declaration order - and both reject options placed ahead of the leaders, as does the hand-rolled baseline. argparse reaches the style only through -mixed, which permits switches ANYWHERE, so it parses the argvec correctly without being able to enforce it: the identical spec also accepts all five positionals trailing. It is timed for the parse, not credited with expressing the constraint, and both halves of that are pinned by tests (sandwich_strictness, sandwich_argparse_is_permissive) rather than left as a silent difference. Its flag convention differs too, opts being reported in each parser's natural form: a supplied bare flag yields an empty-string value and an omitted one is absent from the dict, where the others report -join as 1 or 0. cmdline, tepam, argp and parse_args are unsupported N/A rows in this style, each verified to fail on its argvec. cmdline is the one worth naming: cmdline::getoptions returns WITHOUT error having parsed nothing at all - every option left at its default and the whole argument list untouched - so a check that only looks for a raised error would wrongly credit it. Verification: 25/25 pass under the 32-bit runtime with no skips, which is the only configuration that exercises every new test; 23 pass + 2 skip under native Tcl 9.0.3, where parse_args is absent. first_call times the new parsers in fresh child interps, confirming these binary extensions load a second time in-process. Built artifacts are not in this commit - modules/ is git-ignored. modules/ argparsingtest-1.2.0.tm was deployed via 'make.tcl modules' so the dev shell can see it: a plain 'package require argparsingtest' there resolves a deployed stable version in preference to the 999999.0a1.0 source module, so source edits are invisible until built. Claude-Session: https://claude.ai/code/session_01Q82VuBXzgeKsAwMYHrbpP1 Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	39c6e9a025	build outputs: punk::zip 0.2.0 into bootsupport + _vfscommon, thin-layout make.tcl sync ... Regenerated by 'make.tcl modules' + 'bootsupport' + 'vfscommonupdate' for the G-124 change. make.tcl loads punk::zip from bootsupport, so the reader only reaches the zipfs-less bake path once the snapshot carries it. Also carries two version propagations pending from earlier work, not authored here: punk 0.2.7 -> 0.2.8 and punk::platform 0.1.0 -> 0.2.0 (the latter was added to include_modules.config in 5c1e6927 but never snapshotted). Their source buildversion files are unchanged by this change-set. The templates modpod is rebuilt because the layout make.tcl inside it changed. Note for next time: 'make.tcl bootsupport' cannot replace a bootsupport modpod that the running make.tcl has itself mounted - it failed twice with 'BOOTSUPPORT module update FAILED: ... templates-0.2.0.tm (invalid argument)' while a plain tclsh copied the identical file without complaint. Copied by hand, then re-run to reconcile the punkcheck record. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	78d9bf238a	G-124 achieved: dependency-free zip reading in punk::zip (punkshell 0.26.0) ... punk::zip could write a zip and split an executable-prefixed one, but nothing in the tree could READ a zip without zipfs, vfs::zip or tcllib's zipfile::decode - a system package punkshell does not vendor, with exactly one consumer: make.tcl's zipfs-less kit extraction. Measured at drafting: msys2's /usr/bin/tclsh8.6 has no tcllib, so a bake driven from that host could not lift the runtime's own zip (its tcl_library) into the kit and the artifact did not boot. punk::zip 0.1.1 -> 0.2.0 adds three commands, stock Tcl only. archive_info reports where the archive begins inside its host file and which offset convention it uses; members walks the central directory and returns one dict per entry without decompressing anything; unzip extracts all or part of it with every member's crc32 verified, directories materialized, and the writer's 2MB whole-member-vs-streaming threshold mirrored. All three accept a plain .zip or a zip attached to an executable or script prefix, so "what is inside this kit / runtime / modpod" is now answerable on a runtime without zipfs. The offset derivation is stated once and shared: offsetbase = cdiroffset - diroffset. Positive means an external preamble with archive-relative offsets, zero means the offsets are file-relative (or there is no preamble); dataoffset is the split point either way. Because every surface reads the ORIGINAL file at that base, the file-relative shape stops being a special case - the broken split-off intermediate is never produced. tclsh90b4_piperepl.exe is the store's one file-relative artifact: the old split-then-tcllib path still fails on it with exactly "Bad zip file. Bad closure.", and it now extracts all 841 members crc-verified in 735ms. extract_preamble consumes that machinery instead of carrying its own scan, and two defects went with the refactor: it read 28 bytes of a 30-byte local file header, and it derived the file-relative base from the FIRST central directory record only (its own '#todo! loop through all cdr file headers'). The shared walk takes the minimum over ALL records and then validates that a PK\3\4 really sits there. Its five puts stdout debug lines are gone. Refusals are decided for every selected member before any file is opened, so an unsupported archive cannot leave a half-populated directory: encrypted entries, zip64 size/offset fields, unknown compression methods, crc mismatch (partial file removed) and member paths that would escape the target are each named. Store and deflate only, mirroring the writer's own primitives. make.tcl's zipfs-less extraction now calls punk::zip::unzip on the original runtime and no longer creates the extracted_<runtime>.zip intermediate. No punkshell code requires zipfile::decode any more. Two categories of hit remain and are deliberate: tcllib's own zip/decode.tcl shipped as a LIBRARY inside some .vfs payloads, and one guarded optional probe inside the vendored third-party ooxml1.10/ooxml.tcl. First tests punk::zip has ever had (src/tests/modules/punk/zip/), covering the three archive shapes from one source tree, the introspection output, the named refusals, and the mkzip -> read round trip - the first coverage that what punkshell WRITES is readable. Verification, all 2026-07-26. Reference host for the dependency-free claim: msys2 tclsh8.6 8.6.12, probed as zipfs / vfs::zip / zipfile::decode all ABSENT - suite driven there harness-free, 28 passed, 2 skipped (the zipfs-gated cross-checks only), 0 failed. 30/30 under Tcl 9.0.3, 28 + 2 skipped under native 8.6. Full source-tree suite after the change: 1142 tests, 1 failure - core/tcl exec-14.3, the documented baseline. Independent cross-check: 868 members of tclsh905.exe extracted byte-identical against the same archive mounted with tcl zipfs. Bake demonstration: 'bake -confirm 0 punk905' driven by the msys2 tclsh8.6 (host=msys-x86_64, target=win32-x86_64) produced extracted_tclsh905.exe/{tcl_library,lib} with no intermediate .zip, and the deployed kit boots with tcl_library at //zipfs:/app/tcl_library, 95 encodings, msgcat 1.7.1, http 2.10.2 - and reads itself with the new reader. Trap worth keeping: zlib stream inflate's 'get' returns only what the stream has already produced, so one get per put silently truncates a large member to the first few chunks (measured 327680 of 5048890 bytes on both 8.6.12 and 9.0.3). The inner drain loop is load-bearing, and the >2MB streaming test exists to catch its removal. Standing question answered and pinned: zipfs does NOT misidentify punk::zip::mkzip directories. It keys on the trailing slash (which punk::zip::walk always emits), not the stored permission bits, exactly as the current cli-999999.0a1.0.tm wording says - the older, less precise wording survives only in frozen module snapshots under src/vfs/mkzipfix.vfs and src/vfs/project.vfs. The real difference from zipfs mkzip is that zipfs writes no directory entries at all. The achieved flip archives the entry and the detail file, and sweeps the live tier: G-034 (the reader is its extract-rather-than-mount alternative, plus the measured finding that make.tcl on 8.6 cannot mount its own templates modpod), G-101 (the read half of a zipfs-less 8.6 container is solved), G-125 (the gate now has a named failure reason to report), G-126 (the member-dict keys and fixture set its parity suite must reproduce, and a back-pointer to G-128 whose only recorded bridge was this goal), G-128 (archive_info answers "is this artifact safe to move the overlay of?" in one call). One acceptance clause is reported for review rather than claimed silently: "NO zipfile::decode requirement remains anywhere in the shipped tree" is satisfied for punkshell's own code, with the vendored-third-party residue named above. The archived detail file's diff is a rename plus the flip's evidence write-up. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	5c1e692700	build outputs: thin-layout make.tcl sync + include_modules.config punk::platform entry ... Regenerated by 'make.tcl modules'/'packages' after the G-122 make.tcl changes - the punkcheck-managed thin-layout copies (src/project_layouts/vendor/punk/{basic,project-0.1} and the punk.project payload inside the punk::mix templates modpod) plus the pending punk::platform entry in the bootsupport include_modules.config copies. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	04d2a51b67	G-122 achieved: make.tcl bakes by target platform, not host personality (punkshell 0.25.0) ... 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 (an msys2/cygwin-runtime tclsh, which 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; 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 tier is a first-class mapping entry. The linux tclkit entry now declares linux-x86_64 and bakes to a suffixless bin/punkshell902. 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 type uses. Both mount identically (archive-start-relative offsets). Fixes exposed by the above: the runtime capability probe looked for a suffixless filename and so reported find-fail for every runtime on windows; with real capabilities arriving, the extraction try-list was found to test capability presence rather than value, which would send a plain zipfs runtime down the kit path where a failed sdx mksplit replaces raw_runtime with the un-split original; BUILDCOPY names now follow the runtime's own filename; runtime map keys strip only .exe (file rootname would eat the last dotted segment of tclsh9.0.5); and native windows command lines run through a guard suppressing msys2's posix-path argument rewriting - without it taskkill /PID <n> reached taskkill as taskkill C:/<msysroot>/PID <n>, so the sweep could find a running kit but never replace it. punk::platform folds the msys2 family (mingw64/mingw32/ucrt64/clang*) 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 normalize_os; cygwin-x86_64 added to the platform table; 'help platforms' documents the fold. make.tcl check reports the derivation on one line; bakelist rows carry target=<platform> for non-default targets and the detail block names the tier; workflow text gains key note [K9]. New characterization suite maketclplatform.test (5 tests, two self-gating on discovering and probing a real msys/cygwin tclsh). Full suite 1112 tests: 1093 passed, 18 skipped, 1 failed = the documented exec-14.3 baseline. Verified: msys2 tclsh8.6 bakelist output identical to native Tcl 9; a changed punk91 baked and deployed from that host with the sweep killing a running instance msys ps cannot see; a real cross-target linux kit baked from windows to a 24MB ELF; a zipfs-less 8.6 bake producing a bootable zip kit. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	f4031b209f	G-101 pre-activation survey: reference inventory + sdx wrap.tcl provenance capture ... Notes-only on the goal; the new tracked material is the sdx patch capture. G-101 gains four sections and an extended Scope line naming the surveyed trees and bin/sdx.kit: a reference-material inventory giving the metakit boot chain in reading order with file:line anchors (kbskit kitInit.c pre-init -> boot.tcl raw mk::select bootstrap -> setupvfs.tcl -> mk4tcl.cpp Mk4tcl_Init, with KitCreator's storage-switch variant as the comparison reference for scoping cookfs/zipvfs); sdx provenance; the self-mount shadowing gotcha; and metakit-under-Tcl-9 feasibility. Two findings bear on the container choice: 8.6's core zlib cannot do mk4vfs streaming (kbskit boot.tcl states it - forces whole-blob decompression, relevant to a battery-carrying kit), and TEMP_REFERENCE/metakit is UNPATCHED upstream while KitCreator carries eight 8.x-era fixes. Self-mount shadowing verified live rather than inferred: inside a running tclkit [file type [info nameofexecutable]] is 'directory', so a kit cannot read or copy itself as a file and cannot be handed to sdx as -runtime (wrap.tcl LoadHeader is a plain file read). make.tcl:4995-5005 already works around this with _BUILDCOPY; recorded as a requirement to preserve. Container-agnostic - zipfs has the same shape via its /app anchor. src/buildsuites/suite_tcl86/patches/ captures the bin/sdx.kit delta into tracked form. bin/sdx.kit is tracked and is what make.tcl resolves for the 'kit' wrap path, but it is NOT stock sdx: unwrapping it and stock 2011-03-17 sdx gives 64 identical files except lib/app-sdx/wrap.tcl, whose entire difference is a two-line binary-mode read of tclkit.inf before stringfileinfo patches the PE version resource. That evidence previously existed only in gitignored scratch/ and TEMP_REFERENCE/ copies - one git clean from gone. Stock baseline, patched file and diff are all byte-exact (hashes in the README); the patch round-trips to the patched file byte for byte under 'patch --binary'. The stock baseline is kept in full because upstream (svn://svn.equi4.com/sdx/trunk) no longer exists. Both .tcl files are CRLF and must stay CRLF - vendored fidelity artifacts, exempt from the repo's LF-on-edit preference. Scope-limiting facts recorded rather than assumed: the patched path is DORMANT for present builds (fires only with -runtime + a .exe runtime + a tclkit.inf in the source .vfs, and no tclkit.inf exists in this repo), and the fix itself is UNVERIFIED - plausible as a CRLF-corruption fix but no reproducer was recorded with it. Confirming or retiring it is part of this goal's characterization work, alongside the bootstrap circularity noted in the goal (sdx.kit is itself a metakit starkit, so wrapping a suite-built 8.6 runtime currently needs a prebuilt third-party tclkit of exactly the kind G-101 exists to replace). Of the two proposed acceptance riders, this commit discharges the first (capture the patch before any sdx rebuild). The second stands: the end-to-end demonstration must wrap a SEPARATE COPY of the runtime so the shadowing path is exercised rather than accidentally avoided. A back-pointer to these findings was added to TEMP_REFERENCE/AGENTS.md so the survey is not repeated; that tree is VCS-ignored, so it is not part of this commit. Claude-Session: https://claude.ai/code/session_014UkdFjhuEdfCNDp4f8V6Xe Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	f89bd58c31	G-100 achieved: suite_tcl86 dependency set (Tk 8.6, tklib, tcllib + tcllibc) and an 8.6 runtests census ... Extends the 8.6 buildsuite with the remaining pieces punkshell's Tcl 8 testing depends on, all built from source with the pinned zig 0.16.0 toolchain only. Tk 8.6.17 (build_tk86/build_tk86.zig) is a RE-DERIVATION from the staged 8.6 win/makefile.vc + rules.vc, not a port of suite_tcl90's build_tk.zig: - no generic/tkPkgConfig.c in 8.6 -> no CFG_* defines at all - rules.vc TCL_MAJOR_VERSION==8 arm -> tk86t.dll + lib/tk8.6 (vs tcl9tk90.dll) - @TK_WIN_VERSION@ substitutes $(DOTVERSION).0.0, not the tcl-style <dotversion>.<releaselevel>.<patchlevel> formula - 8.6 TKOBJS carries no tkIcu/tkImgListFormat/tkImgSVGnano/tkWinGDI/tkWinIco/ tkWinSysTray; XLIBOBJS is a first-class makefile macro here (the 9.0 derivation had to recover those files by hand) USE_TCL_STUBS + USE_TCLOO_STUBS against libtclstub86.a is the stock rules.vc arrangement and keeps the dll loadable by both suite shells. tkMain.c is compiled twice per makefile.vc, the ansi copy forced with -UUNICODE/-U_UNICODE (verified the pinned zig cc defines neither for x86_64-windows-gnu, so the 8.6 core's ANSI/TCHAR arrangement is matched exactly). tklib 0.9 and tcllib 2.0 install via their own installer.tcl (batch, packages only, explicit -pkg-path so the installer's tcl_pkgPath-derived default can never write off-tree); tcllibc is built by critcl 3.3.1 driving zig cc (tracked critcl_zig.config, target pinned win32-x86_64-zig - critcl otherwise probes PATH and falls back to its builtin msvc target). critcl compiles against its own bundled headers chosen by the driving shell's version, which is why one config serves both generations. New evidence tools, run on every build: tools/tk_smoke.tcl maps a real toplevel with classic + themed widgets, updates, then destroys it - a load-only check passes even with no usable display tools/accel_bench.tcl same payload through sha1's critcl and pure-Tcl implementations with the digests compared: 4MB -> critcl 5.7ms vs pure-tcl 2736ms = 477x Test tiers (G-107 policy pattern): test-tcllib/test-tklib on the record tier inside test-libraries, plus an opt-in test-tk (winDialog/winMsgbox excluded by default - their message-injection automation blocks on a human under a batch shell). First censuses on tclsh86ts.exe 8.6.18: thread 142 run / 0 failed GATE PASS tklib 796 run / 0 failed record tcllib 64259 run / 41 failed record (names in the summary artifact) tclvfs 51 run / 4 failed GATE PASS after baseline extension The tclvfs baseline grew vfsFtp-1.1..1.3: they sit behind an 'ftp' constraint needing tcllib's ftp package - absent at the G-099 capture, satisfied now that the suite installs tcllib - so they run and reach the retired ftp.tcl.tk, the same external-service class already baselined for vfsUrl-1.2/2.1. punkshell runtests census (both runs 2026-07-26, identical flags, default single-process sequential - 8.6 has no tcl::process so -jobs re-convoys): native Tcl 8.6.17 1003 total / 986 passed / 10 skipped / 7 failed suite tclsh86ts 1003 total / 987 passed / 10 skipped / 6 failed runtests_parity.tcl finds exactly ONE difference: punkcheck concurrency_two_writers_no_corruption failed in the BASELINE run and passed on the suite runtime - a full-suite-load flake (6/6 twice in isolation on both runtimes). Zero suite-runtime-only failures. Tk-dependent suites ran unskipped (repl consolebackends 3/3, including the tkconsole backend on the suite Tk). Two defects fixed in passing: - scrubTclEnv never cleared TCL<major>_<minor>_TM_PATH, so suite shells silently preferred machine/repo tm modules over what the suite had just installed (a repo-built sha1-2.0.4.tm shadowed tcllib's 2.0.5, md5 2.0.8 shadowed 2.0.9) - any census through such a shell measures the machine. The scrub list now covers TCL8_0..8_7 + TCL9_0/9_1 in all three sites (per-step, suite.tcl process-level, bootstrap nested build). suite_tcl90 has the same exposure via TCL9_0_TM_PATH and is deliberately untouched here (out of this goal's scope) - pointers recorded in G-108 and G-116. - build.zig.zon's .fingerprint was hand-authored, so 'zig build' from the tracked suite dir died with 'invalid fingerprint' - the no-tclsh bootstrap flow could never have run. Corrected to zig's computed value (its low half is a checksum of the package name); 'zig build stage' now verified. Goal flipped to achieved 2026-07-26 and archived; reference sweep updated G-005, G-101, G-105 and ARCHITECTURE.md, and the archive-time xref pair G-101 <-> G-116 (whose only bridges were the archived G-100/G-103) gained a direct Related pointer. No project-version bump - arm's-length build tooling, not shipped shell behaviour. Assisted-by: harness=claude; primary-model=claude-opus-5[1m]; api-location=anthropic.com	5 days ago
Julian Noble	a20a64a2a1	G-099 achieved: suite_tcl86 zig buildsuite for Tcl 8.6 (runtime, thread 2.8, tclvfs, core-test gate) ... Builds a Tcl 8.6.18 windows runtime from core-8-6-branch sources with the pinned zig 0.16.0 toolchain only (no MS toolchain), forked from suite_tcl90 (duplication-with-a-note posture). Products (stock threaded-release naming - rules.vc keeps the 't' SUFX on 8.6): bin/tcl86t.dll + tcl86t.lib shared core (static zlib compiled in; no zlib1.dll) bin/tclsh86ts.exe static shell (kit-class; dde/registry static) bin/tclsh86t.exe dynamic shell against the dll lib/libtclstub86.a stub lib for extensions lib/tcl8.6/... script library tree (dde/reg dlls inside) lib/tcl8/{8.5,8.6}/... tm modules; lib/thread2.8.13, lib/vfs1.4.2 Companions: thread 2.8.13 (thread2813t.dll, TEA installed shape from the checkout's own pkgIndex.tcl.in) and tclvfs 1.4.2 (vfs142t.dll) build and load; registry 1.3.5 / dde 1.4.6. No zipfs in 8.6 - exe+dll+on-disk lib tree only, resolved hermetically (exe/dll-relative, TCLLIBPATH/TCL_LIBRARY unset in all child runs). Recipe: build86.zig + build_common.zig + build_zlib86/build_tclthread86/ build_tclvfs86 helpers; tools/ and src/main.zig copied verbatim (suite-agnostic). Two load-bearing 8.6 findings recorded in the README DERIVATION and recipe comments: MP_FIXED_CUTOFFS (the in-core libtommath ships no bn_cutoffs.c) and TCL_BROKEN_MAINARGS + console subsystem for the gcc-class main() entry. Test gates (suite.tcl test -> zig build test-gate/test-libraries) parse all.tcl totals (exit codes untrusted) and gate against tracked dispositioned baselines. First census + two-consecutive-run deterministic PASS (machine SuperBee): core 46525 run / 2 failed (filename-16.12/16.13 admin-share self-globbing) thread 142 run / 0 failed (empty baseline) tclvfs 51 run / 2 failed (vfsUrl-1.2/2.1 external ftp.tcl.tk dependency) Goal flipped to achieved 2026-07-26 and archived; reference sweep updated G-005, G-100 (now unblocked), G-101 and ARCHITECTURE.md. No project-version bump (arm's-length build tooling, not shipped shell behaviour). Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	5c34951441	build outputs: bootsupport punk::platform-0.1.0.tm + templates-0.2.0.tm modpod sync ... make.tcl bootsupport pulled the newly-included punk::platform; the templates modpod refresh used the established virgin-tclsh direct copy (cd6aec8f precedent) because the in-process punkcheck copy fails while the running interp holds the target modpod mounted ("invalid argument") - snapshot now byte-identical to the built modules/punk/mix/templates-0.2.0.tm (carries the G-121-era layout make.tcl payload). Claude-Session: https://claude.ai/code/session_01Jz7wkUsknJzyuJ3tgMaL2t Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	d9061b90ce	bakelist/bake: loud missing-store-folder diagnosis; punk::platform into bootsupport (punkshell 0.24.1) ... Field-driven follow-ons to G-121 from driving make.tcl with an msys/cygwin-runtime tclsh (scoop msys2 tclsh8.6, tcl_platform(platform)=unix): - bakelist: when the derived per-platform runtime store folder does not exist, a stderr WARNING now shows the platform-canon derivation (tcl_platform os/platform -> canon, observed mingw64-x86_64 under a MINGW64_NT MSYSTEM env) and the header line carries (FOLDER MISSING) - every row reading runtime=missing now self- diagnoses its non-per-entry cause. bake's "No executable runtimes found" exit gains the same note when the folder is absent. Native-host output unchanged (maketclbakelist.test 4 PASS). - make.tcl shell 'help platforms' degraded to "punk::platform package not available" on ALL hosts (not msys-specific): the module was absent from the bootsupport snapshot. Added punk::platform to src/bootsupport/modules/include_modules.config (pure Tcl, deps only Tcl's own platform package); verified the full canon table renders via a piped 'help platforms' under tclsh90 src/make.tcl shell. - G-013 Notes: recorded the msys raw-mode field finding (cursor placed at row 1 after each command under the msys pty class) as a raw-mode-default blocker dimension - console/repl work, deliberately outside the make.tcl host/target catering goal being drafted. Claude-Session: https://claude.ai/code/session_01Jz7wkUsknJzyuJ3tgMaL2t Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	c172912f4b	G-121 achieved 2026-07-26: flip + archive record + reference sweep ... Acceptance verified in full (evidence recorded in the detail file's Progress section): bakelist under plain tclsh + punk902z with per-entry presence and deployed state; selective 'bake punk91' refreshing only that kit's _build/bin artifacts and punkcheck records (stat + punkcheck diffs, negative grep clean); unknown-name no-build errors; bare bake kit-set parity; help/workflow contract updates; maketclbakelist.test 4 PASS; punkshell 0.24.0 minor bump. Index entry moved to GOALS-archive.md. Reference sweep: no pending-tense live-tier mentions existed; mapvfs.config pointer updated to the archive path; forward-pointing notes pushed as back-pointers to the live goals they serve (G-024 parsed-model reader swap, G-023 scheme-expansion entry point, G-115 per-kit payload detail, G-057 single-kit icon verification, G-028 deployed -state reporting sibling, G-101 kit-type non-assumption). Detail-file move to goals/archive/ follows in the adjacent commit as a pure rename (root AGENTS.md Doc Restructures). Claude-Session: https://claude.ai/code/session_01Jz7wkUsknJzyuJ3tgMaL2t Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	7d56aa00ff	build outputs: thin-layout make.tcl sync copies (G-121 bakelist + selective bake) ... make.tcl packages run after the G-121 make.tcl change - punkcheck-recorded sync of src/make.tcl into the vendor layouts (basic, project-0.1) and the punk::mix::templates modpod payload copy, per the established sync channels. Claude-Session: https://claude.ai/code/session_01Jz7wkUsknJzyuJ3tgMaL2t Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	0abd07ecd7	G-121: make.tcl bakelist + selective bake (punkshell 0.24.0) ... One parsed-mapping model (::punkboot::lib::mapvfs_parse / mapvfs_kit_outputs / mapvfs_match_outputs) now backs the kit surfaces - the bake machinery, the new bakelist report and the bake/bakelist argdoc kit-name choices all consume the model, never the file format, so the G-024 toml conversion can swap the reader underneath. - bakelist ?kitname ...? (informational SUBGROUP): kit name, type, runtime with presence, vfs folder, deployed state of bin/<kit> vs src/_build/<kit> (current|stale|absent|nobuild; size/mtime fast path, byte-compare when in doubt), anomalies as trailing key=value notes (runtime=missing, vfs=missing, rtrev=r<cur><r<max> via the shared runtime_materialization_check core). Name filtering doubles as per-kit detail (resolved paths/sizes/mtimes). - bake ?kitname ...?: selective kit assembly - unknown names error before any build listing the configured names (flag-order hint included); selected-but- unbuildable kits (missing runtime/vfs) fail fast; runtime BUILDCOPY/caps probe/vfs cksums/vfslibs phase all narrow to the selected kits so other kits' _build/bin artifacts and punkcheck records stay untouched. Bare bake unchanged (verified full-set run). Deprecated alias vfs mirrors the values. - argdocs: SUBVALUES/VALUES_SYNOPSES mechanism, kitname choices enumerated from the mapping at define time (-choicerestricted 0, degrade-safe), block-form help values; SUMMARIES/HELPTEXTS/SUBOPTS/SUBGROUPS/known_commands/plain gethelp/workflow text updated per the update contract; degraded-mode (PUNKBOOT_PLAIN) dispatch collects trailing kit names. - mapvfs.config todo comment retired to point at G-121. - piped characterization: src/tests/shell/testsuites/punkexe/maketclbakelist.test (full-report row/vocabulary + ESC-free, filtered detail, unknown-name errors on both surfaces with the no-build guarantee) - 4 PASS; maketclcolour.test still 3 PASS. - selective-refresh verification (recorded in the goal file): forced punk91 rebuild + deploy via 'bake -confirm 0 punk91'; stat diffs over src/_build and bin show only punk91.exe, punk91.exe.vfs, raw_tclsfe-x64.exe and bin/punk91.exe changed; .punkcheck diffs mention only punk91's records (negative grep for all other kit/vfs names clean). - docs: src/AGENTS.md work-guidance bullet; ARCHITECTURE.md build-entry subcommand list refreshed (bake/bakehouse naming + bakelist, stale in-flux G-112 mention dropped); CHANGELOG 0.24.0 + punkproject.toml minor bump. Claude-Session: https://claude.ai/code/session_01Jz7wkUsknJzyuJ3tgMaL2t Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	f5e3ed1ce0	colour policy: unix-class 8.6 terminals detected via the tty channel signature (-mode) ... Second 8.6 tty rung: on unix-class Tcl 8.6 (linux/WSL/mac AND msys2/cygwin- runtime builds, which report tcl_platform unix even on a windows machine - e.g the scoop msys2 usr/bin/tclsh8.6), real ttys are isatty-gated into the tty channel type and expose the serial/tty options (-mode etc) that pipes/ files lack - a zero-exec, dependency-free discriminator exactly parallel to -winsize. Verified: WSL 8.6.14 pty mode=tty colour on ansistrip=0 vs piped tcl86-plain stripped; msys2 8.6.12 piped tcl86-plain (interactive -mode presence per field report); windows batteries + punkexe suites 54/0 unchanged. Together with the 0.23.2 windows console-encoding signature this completes dependency-free 8.6 tty detection - the planned G-122 probe-ladder goal (procfs/test -t/twapi rungs) is superseded before drafting: channel signatures cover every case without exec or optional packages. Also confirmed by byte-diff: 8.6 and 9.x piped help renders are IDENTICAL (modulo native-8.6 zipfs warnings) - no 8.6-specific rendering defect exists; the top-level help's wide Description indent is a cross-version cosmetic of that one constructed definition, present on 9.x too. punkshell 0.23.3 + CHANGELOG; src/AGENTS colour bullet updated. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	ac7dcdb02a	colour policy: never wrap utf-16 console channels; 8.6 windows console auto-detected by encoding signature ... Defect: the G-113 ansistrip transform was pushed onto Tcl 8.6 windows CONSOLE channels, which encode utf-16 ('unicode') - at byte level ESC arrives as 1B 00 so sequences passed through unstripped and the hold-back carry deferred/ dropped real content, seriously mangling interactive punk86/punksys 'src/make.tcl help' layout (user-reported). Chunk-boundary handling itself is sound: a rig proved strip output invariant across chunk sizes 1..4096 over byte-transparent encodings; the failing dimension was encoding, not cutting. (Comparison recorded: shellfilter::chan::ansistrip is the complement - it decodes per the channel encoding and buffers undecodable tails, but assumes line-buffering for sequence splits.) Fix is detection, not utf-16 stripping: only 8.6 windows console channels report -encoding unicode (pipes/files get the system encoding; hidden-console probes confirmed across bawt tclsh86 8.6.13, punk86 8.6.17, punksys 8.6.13; tcl9 consoles report utf-16 but are caught by -winsize first). The policy now treats that signature as tty: mode=tty, colour ON, no transform - interactive 8.6 windows colour with no env var and no twapi. The transform push is now PER CHANNEL and skips utf-16-class channels ('> file' from an 8.6 console wraps stdout only - stderr console keeps its ANSI); ansistrip_pushed is the wrapped-channel list, shell pops exactly those, policy/check lines print ansistrip=stdout+stderr|stdout|0. Verified: hidden-console 8.6 x3 mode=tty ansistrip=0; half-redirected punk86 console run ansistrip=stdout; piped 8.6/9 batteries unchanged (zero ESC, force re-colours); punkexe suites 54/0 (maketclcolour pins updated to the channel-list format). punkshell 0.23.2 + CHANGELOG; src/AGENTS bullet updated. 8.6 unix/WSL stays plain-by-default pending the probe-ladder goal. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	a646f1089f	punk-runtime use accepts suffixless names; make.tcl bake materialization staleness guard ... use wart fixed (both payloads + rewrapped bin/punk-runtime.cmd, checkfile ERROR-free, scriptwrap roundtrip + binscripts freshness suites green 35/0): candidate lookup falls back to <name>.exe when the exact name is absent - exact names (incl. extensionless unix artifacts in foreign -platform staging folders) still match first. Verified live in a throwaway platform folder via the .cmd (ps1 payload) and the bash payload driven directly (msys exe-magic already masked the wart under msys bash; the fallback serves real unix shells). make.tcl: new ::punkboot::runtime_materialization_warning (warn-only, local, no network) - the kit loop compares each wrapped working runtime's beside-toml revision against the highest -r<N> artifact revision in the same folder and emits a recapped BUILD-WARNING naming the repair command. Verified live: toml demoted to revision=1 fires the warning once + recap (and would have caught today's stale-r1 punk9_beta incident at the first bake); current state silent; no spurious rewraps. Pre-family runtimes (no toml) and direct -r<N> references stay silent. Docs: bin/AGENTS.md runtime-manager contract (suffixless use + guard note), src/AGENTS.md Work Guidance bullet; G-121 Notes records the guard as bakelist's acting-path complement; G-031 Notes records the make.tcl-shell punkexe-surface question with the boot-core vehicle recommendation. punkshell 0.23.1 + CHANGELOG. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	232f2f6096	contextual sweep: deprecated make.tcl subcommand references (user-directed, incl. contract lines) ... Each site read in context, not grep-replaced; bakehouse vs bake chosen per meaning, and missing invocation arguments supplied where the reference describes an unattended flow: - G-005 Approach + Notes, G-006 Acceptance + body: 'make.tcl project' -> 'make.tcl bakehouse' (the full consumer build these goals integrate with; no flags - conceptual references on a clean checkout, not recipes). - G-057 Acceptance: 'make.tcl project' -> 'make.tcl bake' (icon embedding lives in the kit-wrap stage and must hold for a bare bake; bakehouse inherits). - G-028 Context DELIBERATELY UNCHANGED: dated 2026-07-07 incident narrative - the command actually run; history stays verbatim (alias keeps it readable). - src/runtime/mapvfs.config todo: 'vfs' -> 'bake' wish, re-annotated - G-104 achieved WITHOUT the single-kit-by-name/list-kits item (still open). - src/runtime/vendorlib_vfs.toml header: vfslibs now runs as a phase of bake (G-112 fold), not 'as part of project'. - shellexit.test + staticruntime.test rebuild comments: 'project' -> 'bake' AND stale '(interactive y confirmation)' / bare vfscommonupdate corrected to '-confirm 0' (piped y retired since G-030) - the missing-arguments class. - punk.project layout store vfs READMEs (lib + modules): 'make.tcl vfs' -> 'make.tcl bake'; modpod template twins propagated via the punkcheck layout->modpod sync (make.tcl modules run), never hand-edited. Variant-phrasing sweep confirmed non-hits: G-115 '(vfs assembly)' names the src/vfs payload tier (name retained), G-047 'make.tcl/project.new' is the dev project.new command. Archive tier + CHANGELOG history untouched. goals_lint + goals_xref refs clean. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	c6909c3c02	goals-system v3: goals_lint stale-versioned-reference warning + Scope version-reference guidance ... New informational pass in goals_lint (stem-sibling repair heuristic): a live-tier reference to a version-suffixed filename that is MISSING while a same-stem sibling exists on disk warns, naming the sibling(s) - the drift signature of manually-versioned module bumps (libunknown), magic-scheme conversions (config) and stamped bootsupport/vendored renames (fauxlink), exactly the classes fixed in the preceding hygiene commit. References with no stem-sibling stay quiet (proposed goals legitimately name future artifacts); archive tier, globs, URLs, absolute paths and 999999.0a1.0 exempt; warnings never fatal (Scope is contract text - the fix stays a directed edit). Verified: clean on the current tree (incl. correctly-silent existing legacy artifacts) and against a synthetic fixture exercising warn + all five silent classes with per-file dedupe. GOALS.md: front matter bumped v2 -> v3; version-section policy broadened per user direction - bump for ANY change that could affect how an LLM interprets or uses the system (contract, grammar, tooling behaviour, output shape); liberal bumping endorsed. Format section gains Scope version-reference guidance (glob or magic-version filename for stamped files; exact versions only in dated prose). goals/AGENTS.md Verification describes the new check; src/modules/AGENTS.md manual-versioning bullet gains bump step (4): the doc reference sweep. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	81f69d913f	G-099 active: suite_tcl86 born on the G-104 contract - sources pinned, staging live, derivation started ... Activation gated on the G-112 rename core + G-104 contract (both achieved today); score re-run recorded. First increment: - suite_tcl86/: sources.config (source + self-description records - lists and describes itself via 'make.tcl buildsuite list/info' with no make.tcl edits), suite.tcl fork (folder-derived naming, {tcl tclthread tclvfs} requireds, build86.zig recipe file), build86.zig pending-derivation skeleton (staging succeeds, recipe fails with a clear message), build.zig.zon with real zig-fetch hashes, README with the DERIVATION record. - Thread 2.8 selection: thread-2-8-branch @ e14fa771ae (tagged thread-2-8-13, 2.8.13); configure.ac TCL_MAJOR_VERSION==8 guard confirms the pairing. - 8.6 source records in BOTH flows: dev refs (core-8-6-branch -> tcl86, thread-2-8-branch, tclvfs trunk 1.4.2-era) + zon per-checkin pins (tcl 1d1d5cbd91 = 8.6.18 era). - First staging run through 'make.tcl buildsuite build suite_tcl86' (a real driver run through the G-104 surface): checkouts staged from the shared hermetic store, TCL_PATCH_LEVEL 8.6.18 read from the tree. - makefile.vc derivation findings recorded incl. two Context corrections: 8.6.18 DOES generate tclUuid.h; zlib is prebuilt-import in stock nmake (zig-only recipe will compile compat/zlib statically). Core-test census + baseline deliberately pending until a built tclsh86 exists (recorded in Progress). Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	a04979eb38	build outputs: thin-layout make.tcl copies synced (bake/bakehouse era) ... The bakehouse acceptance run's thin-layout sync refreshed the three layout make.tcl copies (basic, project-0.1, templates modpod payload) with the G-112/G-113/G-104 make.tcl - punkcheck-managed, batched per the build-output commit carve-out. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	a949414bf8	make.tcl window doc sweep: bake/bakehouse vocabulary across workflow references ... Root README, src/README, src/AGENTS.md (full-build guidance, provenance command list incl. bakehouse's default-on dirty gate, punk-exe-hosted builds, verification), src/modules + src/lib + src/vfs + src/tests + src/tests/shell + src/runtime AGENTS.md workflow references, and the rebuild-instruction comments in scriptexec.test/tclshcmd.test. G-104's G-058 mention gains its archive marker. Historical CHANGELOG entries untouched; goal-file contract texts (G-005/G-006/G-057 Acceptance mention 'make.tcl project') left for proposal-first user approval - the deprecated alias keeps them executable. Layout make.tcl copies are punkcheck-synced build outputs - they refresh via the normal sync channels in the acceptance build, never hand-edited. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	889d3cd1c7	G-104: make.tcl buildsuite list/info/build + sources.config self-description contract ... buildsuite group added as a declared passthrough (punk::args + fallback paths): list discovers suites (dirs carrying suite.tcl; _build excluded), info renders the manifest-derived detail, build forwards driver args untouched streaming output and exiting with the driver's status (punk-kit-hosted runs launch the driver via the kit's script subcommand, selected by the G-058 marker). Self-description = manifest option: sources.config gains description/product/ doc/zigpin records parsed by both suite.tcl (strict; pinned-zig default now reads zigpin - one source of truth) and make.tcl (tolerant). suite_tcl90 seeded with its records. Copied trees appear automatically (verified with a throwaway copy). Contract documented in new src/buildsuites/README.md + AGENTS.md, indexed from src/AGENTS.md. CHANGELOG extended under 0.23.0. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	221c2c4fff	G-112 rename core: bake/bakehouse stage-true vocabulary, deprecated aliases, vfslibs fold ... vfs -> bake (kit assembly; vfslibs phase folded in so a bake cannot ship stale binary libs), project retired -> bakehouse (packages + bake, -dirty-abort default ON, refusal names the granular developer flow). project/vfs remain as deprecated aliases: one-line stderr notice + pre-dispatch mapping; alias argdocs mirror their targets so flags parse identically. vfslibs survives as a standalone narrowing (in-session decision recorded in the goal file). SUBOPTS/SUMMARIES/HELPTEXTS/SUBGROUPS/known_commands, end-of-run guidance, plain gethelp and workflow text rewritten to the two-persona model (groups: build & bake / promotion gates / source maintenance / deprecated aliases). Verified: tabled + PUNKBOOT_PLAIN help; dirty-tree refusal via bakehouse and via the project alias; alias help renders. punkshell 0.23.0 + CHANGELOG. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	e03ce1d64a	G-113: make.tcl terminal-aware colour policy (tty probe, force overrides, zero-ESC piped) ... Startup policy: NO_COLOR > PUNK_FORCE_COLOR/FORCE_COLOR > stdout -winsize tty probe (Tcl 8.7+/9); Tcl 8.6 fails OFF. Piped/unknown non-forced runs push an ansistrip transform on stdout+stderr - zero ESC bytes from every emitter incl. module-side (punk::ansi's colour_disabled deliberately keeps non-colour effects; design finding recorded in the goal file). shell pops the transform. sgr_cache cleared for punk-exe-hosted pre-loaded interps per the punk::ansi contract. define_global_ansi fully gated; raw-SGR literals routed through new ::punkboot::sgr gate (fixes a malformed \x1b\31m). check reports the policy. make.tcl converted CRLF->LF. Verified: tclsh90 + punk91 kit piped zero-ESC (check/help/workflow/modules); force overrides re-colour piped output; NO_COLOR always suppresses colour; 8.6 tcl86-plain; real Windows console mode=tty (hidden-console evidence); WSL pty mode=tty vs piped-plain (tclkit-902-Linux64). New suite punkexe/maketclcolour.test 3/3. src/AGENTS.md bullet revised; punkshell 0.22.2. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago
Julian Noble	3808ec0914	build outputs: libunknown 0.2.2 promoted (bootsupport + _vfscommon payload) ... make.tcl modules + bootsupport + vfscommonupdate after the G-120 fix; punkcheck managed copies batched per src/AGENTS.md build-output commit carve-out. Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com	6 days ago