What does this skill do, and when should you use it?
This skill is purpose-built for triaging issues and PRs on JoshuaC215/agent-service-toolkit. Before drafting, it researches the full item — body, comments, code diff, and the whole cross-reference cluster of linked issues, PRs, and sibling PRs — and stays consistent with the maintainer's prior statements. Every output is a draft for human review plus a short rationale; it never posts, merges, closes, or labels without an explicit go-ahead. It also encodes the maintainer's voice, accept/decline heuristics (deliberately conservative as of 2026), and a local gh CLI posting flow that preserves clean attribution.
- Reads the full issue/PR body, every comment, and the actual code/diff (for fork PRs it can't open locally, it phrases the fix as 'from the description')
- Builds a cluster map: Fixes/Closes links, references, and sibling PRs touching the same feature, and reads Joshua's prior comments across the whole cluster
- Classifies each cluster (usage question, bug report, feature request, good PR, needs design discussion, spam, security report) with an accept/decline/defer disposition
- Drafts replies in Joshua's voice with rationale, hyperlinked issue/PR numbers, and fenced draft blocks
- Defaults to draft-only; posts via the gh CLI with --body-file only after explicit per-item approval, with no AI attribution appended
- The maintainer himself batch-triaging a backlog of issues and PRs with consistent, non-contradictory reply drafts
- An AI maintenance agent drafting responses on the repo's roughly biweekly triage cycle, pending per-item human approval before posting
- Handling a feature request where you want the built-in heuristics applied first (config-driven? low maintenance? duplicates an existing integration?) before accepting or declining
- Untangling an issue with multiple linked PRs or competing sibling PRs, so replies share one coherent position
- Reviewing suspected bulk AI-generated PRs for real reproducibility, tests, and CI status before responding
- Posting comments as the maintainer himself with no Claude app badge or generation footer
- Users who want fully automated GitHub replies, closes, or merges — it never mutates GitHub state by default and requires explicit per-item approval
- Maintainers of other repos — the voice library, decision heuristics, stock phrases, and CI commands (ruff/pyrefly/pytest) are tailored to agent-service-toolkit and would need substantial rewriting
- Environments without gh CLI access authenticated as JoshuaC215 — the posting rules require verifying `gh api user` returns JoshuaC215 and stopping otherwise
How do you install this skill?
- Shell / CLI
- Network access
- Local filesystem
GitHub CLI (gh) authenticated as JoshuaC215
The skill lives at .claude/skills/maintainer-response/SKILL.md in the repository and ships as part of the whole agent-service-toolkit repo (the README documents no separate skill-install command). Place it under your project's .claude/skills/ directory:
git clone https://github.com/JoshuaC215/agent-service-toolkit.git
cp -r agent-service-toolkit/.claude/skills/maintainer-response <your-project>/.claude/skills/Note: this is 1 of 4 skills bundled in the same repository; this profile covers only maintainer-response.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Draft a reply to issue 290 on agent-service-toolkit in Joshua's voice — don't post it
- Review the open PRs on this repo and give me a cluster-based triage report with drafts
- Write a short reply to this bug report asking for repro steps
- Someone filed a feature request — apply the accept/decline heuristics and draft a response
Trigger it with a triage request naming repo items (e.g. "respond to issue 290", "review the open PRs"). The skill follows a fixed workflow: identify items and pull full context → build the cluster map and read Joshua's prior comments across it → classify and assign a disposition per cluster → draft replies in Joshua's voice → produce a report (per-item summary, research assessment, decision needed with recommendation, and a paste-ready quoted draft, with hyperlinked numbers and balanced fencing) → wait for the human decision. Posting only happens on explicit per-item clearance: use gh issue comment / gh pr comment with --body-file, always passing -R JoshuaC215/agent-service-toolkit, verifying identity via gh api user --jq .login first; issue and PR comment commands are not interchangeable.
What are this skill's strengths and limitations?
- Mandatory pre-draft research across the whole cluster plus adherence to the maintainer's stated positions, preventing contradictory or redundant replies
- Explicit anti-hallucination rules: technical claims must be verified in the repo with path:line citations; unverifiable claims get a repro request instead
- A ready-made voice library, stock phrases, CI/lint command block, and per-category response patterns make drafts immediately usable
- Attribution handling is unusually thorough: posting via the maintainer's own gh token yields no app badge or footer, with exact commands provided
- Tightly bound to one repo and one maintainer's persona; reusing it elsewhere means a near-total rewrite
- The built-in 2026 conservative stance (default toward decline/defer/close) is hardcoded and suits maintainers who want to actively accept contributions
- The multi-step workflow (research, clustering, report, wait for approval) is heavy for replying to a single quick comment
- No independent tests or validation evidence for the skill itself beyond the README and the skill document
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Maintainer Response (agent-service-toolkit) this page | 69 · Recommended | ★ 4.5k | 21d ago | MIT |
| CODEOWNERS PR Splitter ✓ NVIDIA · Official | 57 · Use with care | ★ 3.6k | 11d ago | Apache-2.0 |
| RenderCV Issue Solver | 51 · Use with care | ★ 18k | 7mo ago | MIT |
| Poll GitHub Work | 52 · Use with care | ★ 30k | 11d ago | MIT |
| babysit — PR Babysitting Skill | 54 · Use with care | ★ 10k | 11d ago | Apache-2.0 |
How did FollowSkills review this skill?
Explicitly requires drafts for human review only, never auto-posting to GitHub, and requires verifying claims against the code (citing path:line) rather than fabricating them when unverifiable — a well-designed safety constraint for the high-risk scenario of speaking as the maintainer.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
This review was completed under FSRS 1.0. Dimension scores are not converted; it is queued for re-review under FSRS 2.0.
See the full review method →