[d-morrison]

Extends the "When the session can't reach the home repo" bullet in shared/workflow/report-mistakes-proactively.md.
A repo outside the session's GitHub scope is not yet unreachable when the session has an add_repo tool.
The bullet now says to attach it first, without asking, per default-to-action-grants, and to fall back to filing in the current repo only when add_repo refuses.

Why

On 2026-10-27 a cloud session working in Morrison-Lab/mlg found three ai-config tooling defects.
It reported them as unfiled because ai-config was outside the session's scope, then asked the user what to do once they attached it.
The user answered daytb.
The existing fallback bullet read "not in the session's GitHub scope" as unreachable, which let that happen.
The defects are now tracked as #4785 and comments on #4712 and #3544.

Review

A separate adversarial-reviewer subagent rejected the first draft (a new bullet): it overstated what add_repo can reach, contradicted the fallback above it, and duplicated default-to-action-grants.
The second commit folds the change into the existing bullet; the re-review on f1a32f2 was clean, with one optional nit (posting still passes the membership gate), which the preceding bullets' limit to repos we administrate already covers.

Checks

check-links.py, check-ascii-punctuation.py, check-mistake-patterns.py and check-bare-fragment-mentions.py pass.

---

Generated by Claude Code

[github-actions]

Claude finished review — View run

---

No new findings. The diff content, verification evidence, and conclusion all match the prior round's review: prior round review, which found the diff a clean, self-contained 9-line documentation addition with all three cited artifacts (ai-config#4785, the comment on ai-config#4712, and the comment on ai-config#3544) resolving and matching their claimed content, and the linked fragment default-to-action-grants.md resolving correctly. I independently re-confirmed ai-config#4785 is still open with matching title and that shared/workflow/default-to-action-grants.md exists. The two commits added since that round (62d5e63, 2486730) are a main-merge pulling in an unrelated shared/writing/citations.md change from main, not new content on this PR's own branch — git diff f1a32f2d..HEAD on the file this PR actually touches shows no change. That prior verdict stands.

Verdict

Ready for merge

Structured Review Data (JSON)

{
  "schema_version": "1.1",
  "reviewer": "claude",
  "commit_sha": "433219d",
  "verdict": "CLEAN",
  "findings": [],
  "detailed_assessment": "The PR's actual diff in shared/workflow/report-mistakes-proactively.md is unchanged from the prior clean round; the two new commits only merge in an unrelated shared/writing/citations.md change from main.",
  "holistic_assessment": "No new requirement, integration, regression, or validation concern arose since the prior round; all three cited forge artifacts and the linked fragment still resolve and match their claimed content."
}

Reviewed commit: 62d5e63

[github-actions]

💰 Cost: $0.3599 (review) — run

[d-morrison]

Merging under ai-config's standing merge grant. The cross-model adversarial review was waived: no qualifying reviewer could run from this cloud session.

• OpenCode Go (https://opencode.ai/zen/go/v1/chat/completions): the egress proxy refused the connection on 2026-10-27 (CONNECT tunnel failed, response 403, connect_rejected for opencode.ai:443).
• opencode, codex and agy CLIs: not installed, and no key for any of them is set.

A separate Claude adversarial-reviewer subagent reviewed f1a32f2 clean after one round of fixes. On head 62d5e63 the CI Claude review is clean, check-pr-fully-clean.py exits 0, and the branch is 0 commits behind main.

Posted by Claude Code (AI agent) --- not written by a human.

---

Generated by Claude Code