watch-review-requests.py, in this skill's folder, waits on the user's review queue at no token cost: open PRs that are not drafts, not approved, and request a review from the user's account directly. It exits when a PR enters the queue, and each exit wakes you. Each PR it reports gets its own top-level T3 Code thread running pr-brief, so this thread stays a lightweight dispatcher and every brief keeps its questions and follow-ups in its own conversation. A wake is done when every PR it reported has a launched thread and the watcher is running again.

Arguments: $ARGUMENTS are extra GitHub search qualifiers. Pass them through as --filter on every launch.

1. Launch the watcher

Run it with the Bash tool and run_in_background: true:

python3 <this-skill-dir>/watch-review-requests.py --filter "$ARGUMENTS"

Tell the user in one line that their review queue is being watched, then end your turn. The harness wakes you with the watcher's output when it exits, so while it stays silent, nothing is waiting. The first launch reports every PR already in the queue, so the first wake can come within seconds. --help defines the queue and explains how a re-request counts as new.

The watcher dies with the session. Running /watch-review-requests in a new session resumes it, and requests that were already reported stay quiet.

2. On each wake

The last stdout line is one JSON object. Act on its status: