Skip to content

verify-before-stop: TEST_CMD_RE recognizes the pack's own entrypoints - #76

Merged
F-e-u-e-r merged 1 commit into
mainfrom
hook-testcmd-coverage
Jul 24, 2026
Merged

verify-before-stop: TEST_CMD_RE recognizes the pack's own entrypoints#76
F-e-u-e-r merged 1 commit into
mainfrom
hook-testcmd-coverage

Conversation

@F-e-u-e-r

Copy link
Copy Markdown
Owner

Owner-requested follow-up to PR #74: the hook blocked its own repo's sessions twice today while genuine verification ran — bash hooks/test-*.sh and python3 .github/checks.py were invisible to TEST_CMD_RE. Three command-position alternates added (generalized run-all.sh, test-*.sh suites, checks.py), BLOCK_REASON breadth clause, option-prefix forms documented as a known limit. The gtg checks/run-all.sh convention is untouched (gate-before-commit.sh consumes it; both READMEs teach it).

Two-sided: suite 23/23 on the fix; pre-fix hook fails exactly at the first new allow fixture. cat of a test script still blocks (negative guard).

Gate: single pre-commit round, three families — TRIPLE PROCEED (grok reproduced the suite + stress-probed command-position negatives; luna/sol nits folded or kept-with-reason; false-allow direction explicitly adjudicated as the cheap one for a presence heuristic).

🤖 Generated with Claude Code

https://claude.ai/code/session_01GhLu6NCGzM9HxnqNb87MAx

Observed live twice (2026-07-24): the hook blocked its own repo's
sessions while genuine verification ran — bash hooks/test-*.sh and
python3 .github/checks.py were invisible to TEST_CMD_RE, whose only
script-shaped alternate was the literal bash checks/run-all.sh (the gtg
target-project convention — kept; gate-before-commit.sh consumes it and
both READMEs teach it). Three alternates added inside the existing
command-position group: bash \S*run-all.sh (generalizes the literal,
covers template/run-all.sh), bash \S*test-*.sh, python3? \S*checks.py.
BLOCK_REASON gains a breadth clause; Known limits documents the
option-prefixed forms (bash -e) that stay unmatched. False-allow
tolerance is deliberate and now commented: this is a presence heuristic
and false-block is the expensive direction here.

Tests: +4 fixtures (three two-sided allows + cat-of-a-test-script
negative), suite 23/23; pre-fix hook fails exactly at the first new
allow. Gate (grok-4.5 high + gpt-5.6-luna ultra + gpt-5.6-sol max,
single pre-commit round): TRIPLE PROCEED — grok reproduced the
two-sided run and stress-probed negatives; nits folded (overmatch
comment, option-form Known-limits line) or kept-with-reason (fixture
ordering, .bak-suffix tolerance).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GhLu6NCGzM9HxnqNb87MAx
@F-e-u-e-r
F-e-u-e-r merged commit a5d17b6 into main Jul 24, 2026
3 checks passed
@F-e-u-e-r
F-e-u-e-r deleted the hook-testcmd-coverage branch July 24, 2026 12:44
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant