tee_to_pipe declared the transchan 'clear' op; the Tcl core delivers 'clear'
before every write when declared (the G-145 defect class - see
goals/archive/G-145-piped-usage-ansi-remnants.md) and the handler discarded
o_encbuf, the held partial multi-byte character. The following chunk then began
with orphan continuation bytes and could never decode (the tail-backoff only
trims the end), so whole chunks accumulated in o_encbuf and successive clears
discarded them - eating contiguous multi-KB ranges of multibyte-dense output on
the unix repl console stack: 'i <cmd>' help tables displayed with missing
paragraphs, fused mid-word seam lines, or truncated bottom rows (WSL and native
linux ttys; windows unaffected). Isolation repro: 393/600 lines lost with
clear-per-write vs 0/600 without.
- tee_to_pipe: 'clear' no longer declared in initialize, so the core never
delivers it (same shape as the G-145 fix in ::punkboot::ansistrip); the
method is kept as a state-preserving no-op with a warning comment.
- flush repaired in tee_to_pipe, tee_to_log and logonly: all three referenced
an undeclared o_buffered variable (copy-paste from the o_buffered-bearing
ansiwrap/ansistrip classes) and threw on every invocation. Each now holds an
undecodable partial char in o_encbuf for the next write and emits any
decodable remainder per its class semantics.
- Verified: transform isolation harnesses clean at all chunk sizes with
clear-per-write; WSL end-to-end (linux kit visitor mode on a script(1) pty)
shows all previously-missing define/list table content present;
'make.tcl modules' mint OK.
- shellfilter-buildversion 0.2.4 -> 0.2.5; punkproject 0.49.3 -> 0.49.4 with
CHANGELOG entry; G-090 Context records the finding as audit evidence (no
characterization tests exist yet for these transforms - G-090 scope).
Claude-Session: https://claude.ai/code/session_01QgaxV27VZkmEec7oNbEVFc
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
- GOALS.md: "Goals-system: v2" front matter; canonical "Goals system versions"
section; Scope structure rule (elements separated by '; ' at top level,
commas only inside parens/braces); Child DOX Index note in root AGENTS.md
- goals_lint.tcl: new finding for top-level commas in Scope lines
(paren/brace depth-aware)
- goals_xref.tcl: repo-local configuration via goals/xref.conf (path roots,
stop words, file extensions, reference containers, module dirs,
bare_path_roots, prefix_anchors); defaults byte-identical to previous
behaviour (verified on this tree)
- goals/AGENTS.md: cross-repo goal references drop the G- prefix; xref.conf
documented in Verification
- Scope delimiter normalisation across GOALS.md, GOALS-archive.md and all
detail files, live and archived: 224 lines, top-level commas only;
conservation verified (goals_xref report byte-identical pre/post, diff
hunks are Scope lines only). Archived tier converted under explicit user
approval; sanctioned exception recorded in GOALS-archive.md.
Assisted-by: harness=opencode; primary-model=openrouter/moonshotai/kimi-k3; api-location=openrouter.ai
Both user-approved 2026-07-18 from the runx -teelog / runtests watch-mode
work (a65f9e5b) and the -singleproc 0 completion (218fbb1c).
G-090 records the probe evidence deliberately not addressed in that work:
float-placement starvation of a tee under the float-locked capture
junction, insert_transform re-instantiation churn, parked-worker adoption
never re-applying settings (silent cross-target log misrouting), the
per-file teehandle worker-thread leak in unwatched singleproc runs, and
the tcludp Tcl 9 receive-fileevent first-datagram-only behaviour.
G-091 is the phase-2 parallel test plan over the phase-1 child spawn:
tpool -jobs N, per-child tmpdirs, serial group for console-sensitive
suites, parity + >=2.5x acceptance (floor ~90s, multishell.test bound).
Assisted-by: harness=claude; primary-model=claude-fable-5; api-location=anthropic.com
Julian Noble