pr-watchdog

What it does: Once a day, scans the open pull requests on your repo(s) using the GitHub CLI, flags PRs that are stale or blocked, and writes a short digest you can paste into Slack. Schedule: 0 9 * * 1-5 — 09:00 weekdays. Risk: read-only — only reads from the GitHub API, writes a single markdown file. Tools used: shell_command (to run gh), write_file, web_search is not needed.

Why this is useful

PRs rot. A 14-day-old PR is usually a 14-day-old conflict + 14-day-old context. A daily nudge surfaces them before they become impossible to merge. This recipe is intentionally CLI-only — no GitHub Actions, no webhooks — so it works on private mirrors and self-hosted setups too, as long as gh is logged in.

The workflow file

---
name: pr-watchdog
description: Daily scan of open PRs — flag stale ones, draft a digest.
schedule: "0 9 * * 1-5"
autoApprove: read-only
catchUpOnRestart: true
maxCatchUpAge: 86400
maxIterations: 40
---

# Pull Request Watchdog

Build a digest of open pull requests across the repos I care about. Output is a markdown file I can copy-paste into Slack.

## Repos to scan

Edit this list directly in the WORKFLOW.md:

- `lvndry/jazz`
- `<owner>/<repo>`
- `<owner>/<repo>`

## Step 1 — Verify `gh` is authenticated

```bash
gh auth status

If this fails, write a single-line failure note to the output file (see Step 4) explaining gh auth login is required, and stop.

Step 2 — Pull open PRs per repo

For each repo, run:

gh pr list \
  --repo <owner>/<repo> \
  --state open \
  --json number,title,author,createdAt,updatedAt,isDraft,labels,reviewDecision,mergeable,additions,deletions,url \
  --limit 100

Step 3 — Classify

Bucket each PR using these rules. Stop at the first match.

• Blocked — mergeable == "CONFLICTING" or reviewDecision == "CHANGES_REQUESTED".
• Stale — updatedAt is more than 7 days ago and not draft.
• Awaiting review — open ≥ 24h, no reviewDecision, not draft.
• Big — additions + deletions > 800.
• Draft — isDraft == true (separate section, not flagged).
• Healthy — everything else (don’t include in the digest).

A PR can land in only one bucket. Order matters: blocked > stale > awaiting review > big > draft.

Step 4 — Write the digest

Save to $HOME/.jazz/pr-watchdog/$(date +%Y-%m-%d).md. Use this layout:

# PR Watchdog — <date>

## Blocked (<N>)
- **<repo>#<num>** <title> — by @<author> · <age>
  Why: <conflict | changes requested>
  <url>

## Stale (<N>)
- **<repo>#<num>** <title> — by @<author> · last updated <age>
  <url>

## Awaiting review (<N>)
- **<repo>#<num>** <title> — by @<author> · open <age>
  <url>

## Large PRs (<N>)
- **<repo>#<num>** <title> — +<add>/-<del>
  <url>

## Drafts (<N>)
- <repo>#<num> <title> — by @<author>

Show counts of zero as (0) and skip the section body. If every section is empty, write:

All clear. No PRs needed attention on .

Rules

• Read-only. Never comment, label, merge, close, or approve.
• If a repo errors (404, rate limit), note it under the digest in a ## Errors section and continue with the rest.
• Do not invent PRs that aren’t in the gh pr list output.

## How to install

```bash
# 1. Install and log in to gh
brew install gh                  # or your platform's package manager
gh auth login

# 2. Drop in the workflow
mkdir -p ~/.jazz/workflows/pr-watchdog
$EDITOR ~/.jazz/workflows/pr-watchdog/WORKFLOW.md  # paste the file above
# Edit the "Repos to scan" list

# 3. Dry run
jazz workflow run pr-watchdog

# 4. Schedule
jazz workflow schedule pr-watchdog

The output file lands at ~/.jazz/pr-watchdog/YYYY-MM-DD.md. Pipe it into Slack with anything that reads stdin:

# example: post to a webhook
curl -X POST -H 'Content-Type: application/json' \
  --data "$(jq -Rs '{text: .}' < ~/.jazz/pr-watchdog/$(date +%Y-%m-%d).md)" \
  $SLACK_WEBHOOK_URL

How to customize

• Filter by author — add --author "@me" to the gh pr list command if you only care about your own PRs.
• Different staleness threshold — change “more than 7 days” in Step 3 to whatever makes sense for your team.
• Slack-native output — replace the markdown template with Slack mrkdwn (no # headings; bullets work fine).
• GitHub Enterprise — gh auth login --hostname github.example.com once; the workflow does not change.

What you’ll see

A short, scannable file every weekday at 09:00, with at most ~25 PRs. The blocked/stale/awaiting-review buckets give you a one-glance triage list. After a week of running it, “Stale (0)” becomes the new normal.

Limits

• Needs gh installed and authenticated on the machine where the workflow runs.
• The agent is read-only against the GitHub API by design — it never auto-pings reviewers or auto-merges. Wire that in yourself if you want it.