Pr Watch

pr-watch

Use this skill when you need local, agent-friendly visibility into GitHub pull request conversation, review, and CI activity.

Core rules

• Do not ask the user which GitHub webhook events to subscribe to. pr-watch manages the required event set internally.
• Treat watched orgs and repos as configuration. Add or remove explicit targets instead of assuming defaults.
• Prefer machine-friendly commands: wait, events, status, list, and list-targets.
• Keep the last seen seq cursor and reuse it with --since-cursor so you only fetch deltas.
• Use pr-watch if it is installed on PATH. Inside the source repo, use cargo run -- as the fallback.

Preflight

Start with:

pr-watch doctor

Use doctor when setup is uncertain. It reports:

• whether gh is installed
• whether the cli/gh-webhook extension is available
• whether GitHub auth is present
• whether org forwarding may require admin:org_hook

Do not install extensions or change auth automatically unless the user explicitly asks.

Standard workflow

1. Start the daemon in a persistent shell session:

pr-watch daemon

2. Add the required targets:

pr-watch add-target org ORG
pr-watch add-target repo OWNER/REPO

3. Confirm daemon and forwarder state:

pr-watch status
pr-watch list
pr-watch list-targets

4. Wait for PR activity:

pr-watch wait OWNER/REPO PR_NUMBER --since-cursor LAST_SEQ --timeout 300

5. Fetch the full delta after wake-up:

pr-watch events OWNER/REPO PR_NUMBER --since-cursor LAST_SEQ

6. Persist the highest returned seq and reuse it on the next wait or events call.

Single-shot pattern

When the task is "wait for PR activity, then inspect the delta", use this procedure directly from the skill:

1. Check daemon health with pr-watch status.
2. If the daemon is unavailable, run pr-watch doctor and report the blocker clearly.
3. Inspect pr-watch list-targets.
4. If neither a matching repo target nor a covering org target exists, add pr-watch add-target repo OWNER/REPO.
5. Run:

pr-watch wait OWNER/REPO PR_NUMBER --timeout 300 [--since-cursor CURSOR]

6. If wait succeeds and returns an event with seq = N: Use CURSOR if one was provided. Otherwise fetch the delta with pr-watch events OWNER/REPO PR_NUMBER --since-cursor N-1.
7. Return a concise machine-friendly result that includes:
   • status
   • repo
   • pr_number
   • since_cursor
   • next_cursor
   • wait_event
   • events

Fallback in the source repo:

cargo run -- doctor
cargo run -- status
cargo run -- list-targets
cargo run -- add-target repo OWNER/REPO
cargo run -- wait OWNER/REPO PR_NUMBER --timeout 300 [--since-cursor CURSOR]
cargo run -- events OWNER/REPO PR_NUMBER [--since-cursor CURSOR]

Operating notes

• Run daemon in its own long-lived terminal session. Use separate short-lived commands for wait, events, and target management.
• If status or list shows the daemon is unavailable, start it instead of retrying control commands blindly.
• If add-target fails with GitHub 404, assume missing webhook or admin access on that org or repo until proven otherwise.
• wait exits 0 when it receives a new matching event.
• wait exits 2 on timeout. Treat that as "no new event yet," not a hard failure.
• events returns normalized event history for the PR and ignores unrelated non-PR events.

Limitations

• Forwarding is only org-scoped or repo-scoped because gh webhook forward works that way.
• Org forwarding may require admin:org_hook.
• check_run and workflow_run association is only as good as the PR linkage GitHub includes in the webhook payload.
• For detailed live CI observation on a single PR, gh run watch or gh pr checks --watch can still be useful alongside pr-watch.

Cleanup

pr-watch stop-forwarder ID
pr-watch remove-target org ORG
pr-watch remove-target repo OWNER/REPO
pr-watch stop