You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@magpie.apache.org by "Grove-ovo (via GitHub)" <gi...@apache.org> on 2026/09/12 02:40:50 UTC
[PR] fix(pr-management-triage): include top-level reviews in F5a cooldown [magpie]
Grove-ovo opened a new pull request, #1202:
URL: https://github.com/apache/magpie/pull/1202
## Summary
- Include submitted top-level review bodies in F5a's 72-hour author-response cooldown, so feedback left without an inline thread prevents a premature triage ping. Fetch `body` in the existing `latestReviews` query and use `submittedAt` for ordering.
- Select the newest feedback across all three sources before testing maintainer status. Preserve the strict time boundaries and F5b's original comment-only ping sources.
- Add 11 regression cases and synchronize the related rule documentation, spec and eval counts. Align the existing decision-table eval prompt with its production action and explanation requirements; clarify terminal rendering so full URLs retain their form and an empty `NO_COLOR` still selects plain text.
## Type of change
- [x] Skill change — `skills/pr-management-triage/`; eval fixtures updated
- [ ] Tool / bridge contract
- [ ] Python package
- [ ] Groovy reference implementation
- [ ] Cross-cutting (RFC, AGENTS.md, sandbox, privacy-LLM)
- [x] Documentation — related rule summaries, eval inventory and F5a spec
- [ ] Project template
- [ ] CI / dev loop
- [ ] Other
## Test plan
- [x] Final `prek run --all-files` passes (Python 3.13.5, prek 0.5.2), including links, generated files, workspace ruff/mypy/pytest and spec validation.
- [x] Normal commit hooks pass for the final submission; no hooks were bypassed.
- [ ] Python/Groovy-specific invocation checks — not applicable; no implementation files changed.
- [x] Entire affected skill eval suite: **51/51 passed**, zero failures, manual cases or runner errors, in one complete run with Gemini 3.1 Pro (High) for both the candidate and prose grader. Breakdown: decision-table 19/19, interaction-progress 4/4, pagination-dedup 2/2, pre-filter 21/21, terminal-links 5/5.
- [x] Behavioural regression fixtures included: 11 new cases. All 51 expected JSON files were retained without weakening assertions or removing cases.
- [x] Existing deterministic terminal renderer tests: **9/9 passed**.
- [x] Old/new comparison with GPT-OSS 120B: case-11 fails before the change (`continue`) and passes afterward (`skip/F5a`). Stronger models can pass the old prompt too, so this regression's discrimination is model-dependent.
- [x] Scope review and `git diff --check`: F5c/F6 rules, decision-table actions, query bounds and stale-sweep behavior are unchanged. No temporary reports, editor files or unrelated generated files are included.
## RFC-AI-0004 compliance
- [ ] **HITL** — not applicable; no new mutation or approval flow.
- [ ] **Sandbox** — not applicable; no new host access or network destination.
- [x] **Vendor neutrality** — production skill prose remains project-agnostic; fixture identities are synthetic test input.
- [ ] **Conversational + correctable** — no new adopter setting; the existing cooldown duration is preserved.
- [x] **Write-access discipline** — existing outbound-action gates are preserved; this change suppresses premature proposals.
- [ ] **Privacy LLM** — no new model route or data-egress path; evals use synthetic fixtures.
## Linked issues
Fixes #78
## Notes for reviewers
Case-17 covers feedback delivered only through a top-level review body, with no general comments or inline threads. Case-12 covers recent inline feedback. Cases 13–20 check exclusions and ordering, and case-21 preserves the use of submitted reviews as F5b reply evidence.
The existing bounded `latestReviews` connection returns the latest review per author; this PR does not expand historical review retrieval. The query now includes review bodies, but adds no query or classification step. Stale-sweep activity accounting is outside this change.
The F5a-specific spec is included in this change.
The full-suite investigation also corrected two existing prompt contracts:
the terminal rule contradicted its own visible-form preservation requirement,
and the decision-table eval omitted action details required by its reference
answers. The final change includes those clarifications and the explicit
empty-`NO_COLOR` precedence rule. Classification actions and renderer code are
unchanged. Earlier runs exposed these failures; the 51/51 result applies to
the final corrected prompts, not the earlier revision.
## Generative-AI disclosure
Implementation assisted by WorkBuddy (deepseek-v4.1-flash). Follow-up corrections, verification and submission preparation assisted by Codex (GPT-5 and GPT-6).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@magpie.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
Re: [PR] fix(pr-management-triage): include top-level reviews in F5a cooldown [magpie]
Posted by "potiuk (via GitHub)" <gi...@apache.org>.
potiuk merged PR #1202:
URL: https://github.com/apache/magpie/pull/1202
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@magpie.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org