babysit-pr watcher ignores Copilot PR reviews from copilot-pull-request-reviewer

What issue are you seeing?

The babysit-pr watcher can miss actionable PR feedback from GitHub Copilot because it filters review authors too aggressively.

In the current main version of .codex/skills/babysit-pr/scripts/gh_pr_watch.py, review items are only surfaced when either:

• the author login ends with [bot] and contains an allowed keyword from REVIEW_BOT_LOGIN_KEYWORDS (currently just "codex"), or
• the author is treated as a trusted human with authorAssociation in OWNER, MEMBER, or COLLABORATOR

That means a review authored by copilot-pull-request-reviewer with authorAssociation: NONE is dropped entirely, even if it contains actionable feedback.

What steps can reproduce the bug?

1. Start monitoring an open PR with the babysit-pr skill / gh_pr_watch.py --watch or --once.
2. Have GitHub Copilot leave a top-level PR review as copilot-pull-request-reviewer.
3. Observe that the review is present in GitHub, but the watcher reports no new_review_items for it.
4. Inspect the current filter logic in gh_pr_watch.py around is_bot_login, is_actionable_review_bot_login, is_trusted_human_review_author, and fetch_new_review_items.

What is the expected behavior?

The watcher should surface actionable Copilot review feedback instead of silently ignoring it.

At minimum, copilot-pull-request-reviewer reviews should be treated as eligible review items in the same spirit as the existing Codex review-bot allowance.

Additional information

I verified this against the current openai/codex main branch on April 23, 2026. The local copy of the skill matched upstream for gh_pr_watch.py, so this does not appear to be a stale local fork.

Relevant current logic:

• REVIEW_BOT_LOGIN_KEYWORDS = {"codex"}
• is_bot_login(login) requires login.endswith("[bot]")
• is_trusted_human_review_author(...) only accepts OWNER, MEMBER, or COLLABORATOR

One concrete observed case was a review from copilot-pull-request-reviewer with state: COMMENTED and authorAssociation: NONE; it was visible via gh pr view --json reviews but filtered out by the watcher.

A plausible fix would be to expand the actionable-review-author detection so known automated reviewers like Copilot are not excluded just because they do not use a [bot] suffix or a trusted human association.