agentleFS
SamuelBostic29SamuelBostic29/claude-skills/skills/hum-comments/SKILL.md

hum-comments

Skill1 starsChanged 4 months agoMITView on GitHub

Pull the HUMAN reviewer comments from a pull request and triage them into a worked list: categorize each as SHOULD FIX, NITPICK, or QUESTION with a confidence rating, ask about anything under 85% confidence, and save the list to the human-review state file for /ai-next to work through. Use when the user says "pull the human comments", "triage the reviewer feedback", "/hum-comments", or pastes a PR link asking to work through what human reviewers said.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After new human comments land on a PR already triaged (re-run to merge them in)

    line 47
  2. Very specific:

    Resolve the PR: If a PR URL or number was passed, use it.

    line 57
  3. Distinctive:

    Understand each comment: Read the code it points at — file, surrounding function, covering tests.

    line 72

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Hum Comments: triage human reviewer feedback on a PR into a worked list

A PR has review comments from human colleagues and the user wants them triaged into the same checkpoint pipeline the AI comments go through (/ai-next → /ai-draft → /ai-post), but kept in their own long-lived file. Your job is to fetch every open human comment, actually understand the code each one touches, categorize it with an honest confidence rating, and write the result to the human-review state file. Two things are different from AI triage: humans ask real questions that deserve answers rather than fixes, and these files are a permanent archive the user keeps for one-on-ones — you append history, you never prune it. As with AI triage, the classic failure mode is categorizing from the comment text alone — read the code first, always. You triage only; working the items starts with /ai-next.

Configuration (set per adopter)

KeyDefaultUsed for
HUM_STATE_DIRdocs/plans/hum-review/Where state files live. Point at a global folder (outside any repo) to keep them long-term across repos.

State file path: {HUM_STATE_DIR}/{repo}-pr-{n}.md.

When to use this skill

  • "Pull / triage the human reviewer comments" on a PR
  • A PR link or number is pasted with intent to work through colleagues' feedback
  • After new human comments land on a PR already triaged (re-run to merge them in)

When NOT to use this skill

  • AI reviewer comments (Copilot, Claude bots) — use /ai-comments; never mix the queues.
  • Working through an already-triaged list — /ai-next; drafting/posting — /ai-draft / /ai-post.
  • Reviewing the PR yourself — this skill only ingests reviewers' comments.

Steps

  1. Resolve the PR. If a PR URL or number was passed, use it. Otherwise gh pr view for the current branch. If neither resolves to a PR, ask via AskUserQuestion and stop.
  2. Fetch the human comments. Three sources, keeping only comments by humans — not the PR author, not AI reviewers (Copilot / Claude bots), not CI or scanner bots:
    • Inline: gh api repos/{owner}/{repo}/pulls/{n}/comments — record each comment's id, path, and line.
    • Review summaries: gh api repos/{owner}/{repo}/pulls/{n}/reviews — reviews with substantive body text (skip empty approvals).
    • Issue-level: gh pr view {n} --json comments.
  3. Drop what's already handled. Skip inline threads where the PR author has already replied. If the state file exists, merge: add only items whose comment id isn't already listed, and never change existing items' statuses or history.
  4. Understand each comment. Read the code it points at — file, surrounding function, covering tests. For comments about shared code or behavior claims, get the full picture: invoke a call-trace skill if installed, otherwise trace manually with Grep/Read. A question ("why not use the existing endpoint?") needs the same grounding as a defect claim — the eventual answer must be code-accurate.
  5. Categorize with confidence. Each comment gets SHOULD FIX, NITPICK, or QUESTION plus a confidence percentage that the categorization is right; under 85% → UNSURE. SHOULD FIX carries a Why and a concrete Proposed solution; QUESTION carries What they're asking and, when the code answers it, a Grounded answer draft for the user to confirm at /ai-next.
  6. Resolve UNSURE items by asking targeted questions via AskUserQuestion (batched). Re-categorize with the answers; what the user can't settle stays UNSURE.
  7. Write the state file to {HUM_STATE_DIR}/{repo}-pr-{n}.md (create the folder if missing), same format as the AI queue (see /ai-comments) plus a [QUESTION] item shape, items ordered SHOULD FIX → QUESTION → UNSURE → NITPICK. Include the PR title, repo, reviewers, and today's date in the header — the file outlives the PR.
  8. Stop. Print the summary table and point to /ai-next. Do not begin fixing or answering anything.

Output format

Chat summary:

Triaged PR #12 — 5 human findings from 2 reviewers (2 should-fix, 2 questions, 1 nitpick) → <HUM_STATE_DIR>/repo-pr-12.md
| # | Cat        | Reviewer | Conf | Finding                    |
|---|------------|----------|------|----------------------------|
| 1 | SHOULD FIX | jsmith   | 90%  | list should be virtualized |
...
Next: /ai-next

Item shapes (full format spec lives in /ai-comments; QUESTION adds):

### 3. [QUESTION] why a separate endpoint for selections?
- **Source:** jsmith (inline comment <id> on `<path>:<line>`)
- **Confidence:** 88%
- **Comment:** <the question, condensed but faithful>
- **What they're asking:** <the actual information sought>
- **Grounded answer draft:** <code-accurate answer, if the code settles it>
- **Status:** OPEN

Rules

What to do

  • Read the code before categorizing — every item, questions included.
  • One comment = one item, each with its reviewer named — the archive is also a record of who raised what.
  • Honest confidence. Under 85% → UNSURE, no rounding up.
  • Merge, don't clobber. These files are permanent — append new items, touch nothing historical.

What NOT to do

  • NEVER ingest AI or bot comments — that queue belongs to /ai-comments.
  • NEVER delete, prune, or rewrite existing file content. The archive is the point.
  • NEVER start fixing or post an answer. Triage ends at the state file.
  • Don't editorialize about reviewers. Items record what was said and where it stands — no commentary on comment quality.

Format discipline

  • Summary table first, detail in the state file, stop after pointing to /ai-next.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.