[fdaviddpt]

Curate pass over trap.d/ (#1670/#1723 procedure). 9 fragments read together, each decided on its own merits.

Promoted (2)

Both merged into .claude/jit-context/paths/00-manual/hooks.md's existing "one awk trap, measured on both engines" section -- they are the same class of lesson (an awk-writing gotcha specific to common.sh/common-awk.sh), so a reader already in that section finds them beside their siblings rather than in a new file.

• Part of A ~ Bash rule matches words inside a heredoc body, so a payload that merely mentions a blocked command is refused #432 -- a dynamic regex string needs a doubled escape byte to put one literal backslash inside a bracket expression; a /literal/ regex does not. Cost one common-awk.sh extension 205/343 tests, confirmed by reverting.
• Part of Heredoc stripper (#432) hides real commands from block rules when a <<WORD string is not a heredoc opener #442 -- in is an awk reserved word and compiles silently as a parameter/array name on both engines, no syntax error, just garbage downstream. Cost 211/374 tests, fixed by a rename alone.

No match pattern changed on either merge (body text only, into an already-firing file), so no new firing proof is needed and rebuild-tsv.sh was not run.

Declined, already fixed (2)

• trap.d/424.jit-misses-generic-words-v-escape.md and half of trap.d/424.rebuild-tsv-v-escape.md (the GENERIC_WORDS_FILE/wf site) -- both routed through ENVIRON in e0f8d39 (Ship a slim release branch to the Anthropic directory (port claude-remember's release-branch system) #437/Ship a slim release branch for the Anthropic plugin directory (#437) #438), which this pass found while checking the claim against HEAD rather than trusting the fragment's own age.
• trap.d/466.privacy-page-claims-vs-code.md -- all four claims are now stated correctly in PRIVACY.md, landed in docs: PRIVACY.md says what the hooks really write (#466) #468 and docs: PRIVACY.md, the four claims the second audit round caught (#466) #470.

Declined, still true -- filed (4)

Each checked against HEAD (bda0282) before filing, per the decline-of-a-code-claim rule:

• Part of trap.d/424.rebuild-tsv-v-escape.md (the layerdir/bytesof() site) -- rebuild-tsv.sh's bytesof() still reads layerdir through -v, not ENVIRON #475
• trap.d/461.release-check-crash-reads-as-pass.md -- A check_release_tree.py crash reads as a pass in test-release-branch-437.sh and -461.sh #476
• trap.d/461.release-equivalence-builds-head-not-source-ref.md -- Release-branch equivalence test builds HEAD, not $SOURCE_REF -- gap on workflow_dispatch with an older ref #477
• trap.d/459.awk-pipe-failure-masked-by-or-true.md -- detect_heredoc_lines() in test-no-heredocs-459.sh: an awk crash and a clean file render identically #478

Declined, agent-facing lesson (1)

• trap.d/437.wrong-worktree-edits.md -- forgetting a per-call cd <worktree> && prefix. Not promoted: no ERE can distinguish "this call already has a same-call cd prefix" from "it does not" (no lookbehind in this awk dialect), so the only matchable rule would fire on every edit/paste/batch call in the whole repository, correctly-prefixed ones included. The supertool write op's own [branch: ...] receipt already names where a write landed, checkable after the fact on exactly the calls that need it.

Disagree with any of this

Say so on the PR -- every decision above is a sentence, not a verdict, and this pass has no standing to insist on any of them.

trap.d/ ends this pass empty (README.md only). Nothing deferred.

[AI-generated]

[fdaviddpt]

Maintainer review (tick wait+review step, skills/manager/phases/review.md):

Content: clean. Both promoted lessons (dynamic-regex bracket escaping, in as a reserved word) are well-sourced, body-text-only edits into hooks.md's existing awk-trap section -- no match pattern changed, so no rebuild-tsv.sh run needed, consistent with the PR's own claim. The four declined, still true items were each checked against HEAD before filing (#475-#478), all carry filed-by-loop + a lane + a priority label. Blast radius matches the description exactly (trap.d/ deletions, two 00-README.md additions, hooks.md addition).

CI finding (below-bar, not this diff's fault): the one real failure -- hooks (windows-latest, shard 2/4), job #112276374977 -- is test-run-all-timing.sh's FAIL: the slower suite is listed before the faster one (slow_line=10 fast_line=9). This diff touches only trap.d/ and two 00-manual docs, nothing under tests/ or tests/run-all.sh, so the failure is unrelated to this PR's content: a tie-break flake where the fixture's "slow" (sleep 1) and "fast" suites rounded to the same integer wall-clock second on a loaded Windows runner, making the slowest-first sort order ambiguous. Not blocking this release by the ranking table in skills/manager/phases/findings.md (a test-harness tie-break, not a shipped-code defect) -- recorded here as the receipt rather than filed as an issue. Re-run the failed leg before merge; the fragment leg shown as superseded/cancelled is expected (a later run of the same name superseded it, per pr_green.py).

Decision: content ready-to-merge; merge blocked only on a green re-run of the windows hooks shard-2/4 leg.