Regression Issue Fix — Hard intake gate
When to use
User asks: "fix issue #N" / "read and fix issue #N" / "address the
regression in issue #N" — for any GitHub issue that affects a
Phase 1 per-doc auto-extractor: L1.ic_name, L6.fsm_states,
L4.regmap, L8.timing_waveform, L11.otp_content, etc.
Step 2 — Confirm the fixture FAILS on current code
Before writing any fix:
cd vibe-ic-marketplace/plugins/vibe-ic
python3 -m pytest tests/test_phase1_fixtures_regression.py -k <project> -v
The new fixture must fail (the picker returns the verifier's
reported actual value, not the expected). If the test passes
on current code, the issue is stale or already fixed — comment on
the issue and stop. Do not push a no-op release.
Step 3 — Write the fix
Edit the picker / extractor in
programs/phase1_one_shot_runner.py (or the relevant L-doc
generator). Constraints:
- Run the full fixture suite after each iteration:
pytest tests/test_phase1_fixtures_regression.py - Add the new fixture's expected name to
tests/test_phase1_fixtures_regression.py::_EXPECTEDonce the fix passes.
Step 4 — Self-verify, push, close with core-closed
The core agent now SELF-VERIFIES and CLOSES the issue. Before
closing, run the FULL plugin test suite the CI way and reproduce
the original failure to confirm it is gone (the 本機驗證
evidence below).
# 1) self-verify the CI way (reproduce + full suite)
python3 -m pytest tests/test_phase1_fixtures_regression.py -k <project> -v
cd vibe-ic-marketplace/plugins/vibe-ic && python3 -m pytest # full suite
# 2) push the chip-AGNOSTIC fix + version bump
git push origin main
# 3) close + tag core-closed (the field-audit target marker)
gh issue close <N> --repo vibeic/vibe-ic
gh issue edit <N> --repo vibeic/vibe-ic --add-label core-closed
Post a 繁體中文 close comment in the canonical 5-section shape
(see below), then close + add the core-closed label.
New backlog lifecycle (replaces the retired
wait-for-verification limbo): the default terminal state of a
fixed issue is CLOSED with core-closed. The field agent audits
CLOSED core-closed issues against the real benchmark every tick
— on success it adds field-verified (terminal); if the fix is
inadequate it gh issue reopens, removes core-closed, and posts
counter-evidence, putting the issue back in the core agent's queue.
wait-for-verification is RETIRED — never apply, poll, or
cross-check it.
Core close-comment shape (5 sections, 繁體中文)
Core agent 已推送修復:<sha>
**問題**:<重述 field-agent 的問題>
**根因**:<root cause>
**修法**:<chip-AGNOSTIC fix + files>
**本機驗證**:<gates/tests run + result, e.g. N/N PASS>
Core agent 已自行驗證並關閉此 issue(已加 core-closed 標籤)。field agent 複查若發現未完整,請 reopen 並補反證。
What this skill enforces
- Verbatim input or no fix. Issues that omit the input snippet are bounced back to the verifier with a templated ask. No guessing rounds.
- Fixture lands first, fix lands second. A regression that
doesn't reproduce in
programs/tests/fixtures/phase1_local/cannot be fixed reliably; the fixture is part of the issue, not a downstream artefact. - Self-verify then close with
core-closed. The core agent self-verifies (reproduce + full CI suite) and closes the issue, taggingcore-closed. The field-agent's real-benchmark audit is the safety net: it addsfield-verifiedon success orgh issue reopens (removingcore-closed) with counter-evidence on failure.wait-for-verificationis retired.
Cross-references
- Issue template:
.github/ISSUE_TEMPLATE/picker-or-extractor-regression.yml - Fixture suite:
programs/tests/fixtures/phase1_local/+tests/test_phase1_fixtures_regression.py - Sibling skill (general bug→backlog):
skills/community-backlog-submit
Smallest fix of the original problem (Charter C10, owner 2026-11-14)
Owner: "fixing PR/issue 不要越搞越複雜,用最小的方式、最小的改動去修正原來的問題." Before a fix, state the original problem in one sentence; keep main's existing path and change only what that problem needs. Review asks only: is the original problem fixed, is anything worse than main, and is the fix general and no-cheating? A defect found in behaviour main already has is a separate known gap, written down, never folded into the same PR. After about two SEND BACKs, or when a PR keeps adding new mechanism (parsers, analyses, inventories, readers) to answer counterexamples, stop and restart from main as the smallest fix of the original problem. A smallest fix enters the plugin or a skill only if it is also general and no-cheating (owner 2026-11-14): it holds for any design meeting the same condition, names no benchmark, problem, design or testbench, and takes nothing from a testbench, reference answer, scorer or score. A design-specific fix belongs in that design, not in the product.