GitHub Skill

SkillCommunication

Lets your agent triage GitHub notifications and classify issue and PR queues using a github skill.

Available today. Use it from your connected AI after setup.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use the GitHub Skill skill

About this skill

GitHub: notification triage, profile rule extraction.

What this skill tells your AI

The instructions your AI receives, as published by notque/vexjoy-agent in skills/integrations/github/SKILL.md and read by ahel’s review.

GitHub notification triage and issue/PR queue classification.

Mode Selection

RequestMode
Notifications, inbox, cleanupNotification Triage
Issue or PR queue, review itemsQueue Triage

Notification Triage

Fetch, classify, and report GitHub notifications via the triage script.

Commands

# Report-only (default)
python3 scripts/github-notification-triage.py

# Mark informational notifications as read
python3 scripts/github-notification-triage.py --mark-read

# Save report to ~/.claude/reports/notifications/
python3 scripts/github-notification-triage.py --save

# Cron/scheduled: auto-clear noise and save
python3 scripts/github-notification-triage.py --mark-read --save

Steps

  1. Run report-only: python3 scripts/github-notification-triage.py.
  2. Display the output. The report classifies notifications as:
    • Action required -- PRs awaiting review, mentions, assigned issues.
    • Informational -- CI results, bot comments, automated updates.
  3. If the user says "clean them up", "mark read", "clear the noise", or "yes" to clearing informational items, re-run with --mark-read. Confirm how many were marked read.

Scheduled (no interactive user): run with --mark-read --save.

Exit codes: 0 = success, 1 = error (auth, API, script not found).


Queue Triage

When triaging a queue of issues or PRs (not just notifications), produce one card per item.

Card Format

Open each card with the item's full GitHub URL on its own line, then these fields in order:

FieldContent
WhatOne sentence: what the issue/PR asks or changes.
WhyMotivation or problem it addresses.
Author trustMaintainer, known contributor, first-timer, or bot. Cite prior merged PRs.
FitMatch to project scope and conventions.
RiskBlast radius: files touched, API/behavior changes, security or data paths.
Proof stateCI status, tests added, repro steps, screenshots. "Unverified" when none.
BlockerSingle thing stopping progress, or "None".
Next actionOne concrete step with named actor. Label Autonomous or Needs-owner.

Example:

https://github.com/owner/repo/pull/123
What: Adds retry with backoff to the webhook sender.
Why: Webhooks drop on transient 5xx from receivers.
Author trust: Known contributor, 4 merged PRs.
Fit: Matches existing client retry pattern in net/client.go.
Risk: Low — one module, behavior gated behind config flag.
Proof state: CI green; unit tests cover backoff schedule.
Blocker: None.
Next action: Autonomous — review and merge.

Autonomous vs Needs-owner

  • Autonomous -- the agent can complete the action alone: reply with a question, label, close an obvious duplicate, merge a green PR within granted authority, rebase.
  • Needs-owner -- requires owner judgment: scope decisions, breaking changes, security-sensitive merges, releases, money, or access.

When in doubt, label Needs-owner and state the decision the owner must make.

Clean-Checkout Gate

Before any local work on an item (checkout, tests, repro):

  1. Run git status --short.
  2. Proceed only when output is empty.
  3. If dirty, mark the card Needs-owner with blocker "dirty working tree in <repo>" and continue triaging from API data only.

Signals

GitHub stars
425
Forks
46
Last commit
Sep 2026
Hacker News mentions
20

ahel review

  • S4info
    community integration, published by notque, not github

Automated review, not a security audit. Ruleset v1+k2.

Advanced
Catalog kind
skill
Key
github-notque
Source
github.com/notque/vexjoy-agent