Apollo Community Review

Use this skill when running the recurring Apollo community review automation.

This skill orchestrates:

• discovery of new/newly active issue and PR threads
• structured review via $apollo-issue-review or $apollo-pr-review
• deterministic policy evaluation via $SKILL_ROOT/scripts/community_review.py
• optional auto-posting for high-confidence issue comments and PR top-level comments only

Discovery notes:

• Ignore bot-authored comments/reviews and automation-only update churn from accounts such as stale[bot] and mergify[bot].
• Candidate scanning should only treat non-bot external activity after the lookback window as actionable.

Path notes:

• The commands below assume SKILL_ROOT points to the apollo-community-review skill directory.
• Example: export SKILL_ROOT=/path/to/apollo-community-review

Defaults

• repo: apolloconfig/apollo
• actor: resolve in this order: --actor → APOLLO_COMMUNITY_REVIEW_ACTOR → current gh auth login
• maintainers: prefer --maintainers / repoMaintainers; otherwise fall back to the resolved actor
• mirror dir: ~/.codex/tmp/apollo-review-mirror
• state file: if not passed explicitly, default to ~/.codex/tmp/apollo-community-review/<actor>/state.json
• default branch: master
• shared org policy: $SKILL_ROOT/references/repo-policy.json
• optional shared maintainer map: --maintainers-file /path/to/repo-maintainers.json (see $SKILL_ROOT/references/repo-maintainers.example.json)
• optional local operator override: --operator-file /path/to/operator-config.json (see $SKILL_ROOT/references/operator-config.example.json)

Workflow

1. Sync the isolated mirror:

   python3 "$SKILL_ROOT/scripts/community_review.py" \
     sync-mirror \
     --repo apolloconfig/apollo \
     --mirror-dir ~/.codex/tmp/apollo-review-mirror \
     --default-branch master
   

2. Discover candidates:
   • Single-repo scan (legacy / focused mode):

     python3 "$SKILL_ROOT/scripts/community_review.py" \
     scan \
     --repo apolloconfig/apollo \
     --actor <github-login> \
     --maintainers <comma-separated-maintainers> \
     --state-file ~/.codex/tmp/apollo-community-review/<github-login>/state.json \
     --initial-lookback-hours 4
     

     If --actor is omitted, the helper resolves it from APOLLO_COMMUNITY_REVIEW_ACTOR or the current gh auth login. If --maintainers is omitted, it falls back to the resolved actor.

   • Org-level scan (preferred for scheduled Apollo maintenance):

     python3 "$SKILL_ROOT/scripts/scan_org.py" \
     --policy-file "$SKILL_ROOT/references/repo-policy.json" \
     --maintainers-file /path/to/repo-maintainers.json \
     --operator-file /path/to/operator-config.json \
     --initial-lookback-hours 4
     

     This enumerates accessible apolloconfig repositories, skips archived / disabled repos, keeps the configured priority repo order first, and returns a flat candidate list plus per-repo scan results/errors. Use repo-maintainers.json for the shared repo→maintainer map and operator-config.json only for local actor overrides so the shared skill does not hardcode a specific maintainer identity.

3. For each candidate:
   • Fetch thread context with fetch-thread.
   • If it is a PR, also check out the PR head in the isolated mirror before reviewing:

     python3 "$SKILL_ROOT/scripts/community_review.py" \
     checkout-pr-head \
     --repo apolloconfig/apollo \
     --mirror-dir ~/.codex/tmp/apollo-review-mirror \
     --number <pr_number>
     

   • Run the appropriate review skill in output_mode=pipeline.
   • After the review output, append exactly one COMMUNITY_REVIEW_DECISION JSON block with:
     • language
     • decision_confidence
     • recommended_action
     • validation_completed
     • missing_info_fields
     • blocking
     • blocking_reasons
     • draft_body
   • Save the combined output to a temp file, then evaluate policy with:

     python3 "$SKILL_ROOT/scripts/community_review.py" \
     decide \
     --candidate-file <candidate_json> \
     --review-file <review_output_txt>
     

4. Auto-post only when auto_send_eligible=true.
   • Allowed auto-post actions:
     • issue top-level comment
     • PR top-level comment
   • Never auto-post:
     • approve
     • request changes
     • merge-ready
     • security/admin threads
     • draft PRs
     • threads with a newer maintainer reply after the latest external activity
   • Post via:

     python3 "$SKILL_ROOT/scripts/community_review.py" \
     post-comment \
     --decision-file <decision_json>
     

5. After each handled thread, persist the processed activity signature:

   python3 "$SKILL_ROOT/scripts/community_review.py" \
     mark-processed \
     --state-file ~/.codex/tmp/apollo-community-review/<github-login>/state.json \
     --candidate-file <candidate_json> \
     --decision-file <decision_json>
   

   If --state-file is omitted, mark-processed reuses the actor-scoped default derived from the candidate metadata.

6. At the end of the run, summarize all results:

   python3 "$SKILL_ROOT/scripts/community_review.py" \
     summarize \
     --results-file <results_json>
   

Output Rules

• Always produce one inbox item with these sections:
  • Auto-sent
  • Needs review
  • Skipped/Error
• If there is no actionable work, include ::archive-thread{} after the inbox item.
• Keep the full draft body in the Needs review section.
• Auto-sent comments must include the localized AI disclaimer; manual drafts must not add it automatically.

Hard Safety Rules

• Never use the current workspace remote configuration to decide which repository to query.
• Never post to GitHub outside the helper CLI flow above.
• Never auto-send a formal review state.
• If discovery, parsing, or GitHub API calls fail for a thread, route it to Skipped/Error.