pr-triage

What this is

run-pr-triage.sh discovers every mazze93 repo with at least one open PR, classifies each PR by why it's stuck (not just how old it is), and prints a report. The point is actor-oriented triage: every stuck PR is tagged with who needs to act next (author, reviewer, ci, or nobody/unknown), so the output is a to-do list, not just a status board.

Usage

skills/pr-triage/run-pr-triage.sh                     # terminal report (default)
skills/pr-triage/run-pr-triage.sh --render slack       # Slack Block Kit JSON — pipe to a webhook
skills/pr-triage/run-pr-triage.sh --render github       # GitHub-flavored markdown — post as a PR comment
skills/pr-triage/run-pr-triage.sh --output report.md    # write to a file instead of stdout

Requires gh (authenticated) and python3 on PATH — both are checked up front and the script fails fast with an install/auth hint if either is missing.

run-pr-triage.sh auto-discovers repos: it lists every repo under mazze93, keeps only the ones with open PRs, and builds a temp config on the fly — no setup needed for the common case. To scope a run to specific repos instead, invoke the Python entry point directly with your own config:

PYTHONPATH=skills/pr-triage python3 skills/pr-triage/pr_triage.py --config path/to/config.json --render terminal

Config JSON shape (repos is the only required field):

{
  "repos": ["mazze93/some-repo", "mazze93/another-repo"],
  "reviewer_response_hours": 48,
  "abandon_days": 5,
  "staleness_warning_days": 3,
  "render": "terminal"
}

Classification taxonomy

Every PR gets exactly one StuckType (see classify.py for the precedence order — conflict and CI blockers are checked before review state):

Severity (HIGH/MEDIUM) escalates once a PR has been stuck past staleness_warning_days (default 3). The terminal report's "Needs Action" section surfaces HIGH-severity PRs first.

Architecture

• fetch.py — talks to GitHub via gh pr list (parallel across repos, ThreadPoolExecutor), normalizes into PRData. No classification logic here.
• classify.py — pure functions, no I/O: PRData + TriageConfig in, PRClassification out. This is where the taxonomy above is actually implemented; extend it here if you need a new stuck type.
• render.py — three renderers (render_terminal, render_slack, render_github) over the same PRClassification list; presentation only, never re-derives classification.
• pr_triage.py — CLI entry point: config loading, preflight (gh/python3/auth checks), orchestration, output delivery.
• run-pr-triage.sh — the actual entry point for day-to-day use; wraps pr_triage.py with automatic repo discovery so there's no config file to maintain.

Partial failures are non-fatal: if gh pr list fails for one repo, fetch_all logs a warning and continues with the rest.