agentleFS
Davidengstrom01Davidengstrom01/claude-skills/skills/take-feedback/SKILL.md

take-feedback

Skill0 starsChanged 41 days agoMITView on GitHub

Fetches GitHub pull request review comments via the gh CLI, judges each one purely on whether it identifies a real correctness/functionality issue, and reacts with a πŸ‘ (agree) or πŸ‘Ž (disagree) emoji reaction. πŸ‘Ž verdicts also get a short drafted reply, shown to the user for approval before anything is posted. Use this whenever the user says \"take-feedback,\" asks to \"review PR comments,\" wants to \"react to PR feedback,\" or otherwise wants Claude to triage and respond to reviewer feedback on a pull request β€” even if they just paste a PR link and ask what to do with the comments on it.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • β‰ˆ0

    Common Β· estimate

    Many files like this have a rule of this kind.

  • β‰ˆ4

    Distinctive Β· estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No preamble, no retelling of the diff, no closer.

    line 17
  2. Very specific:

    If the authenticated user already left +1 or -1, skip it.

    line 42
  3. Very specific:

    Read body and diff_hunk; for replies, also read the parent comment.

    line 46
  4. Distinctive:

    If the user overrides a πŸ‘ to πŸ‘Ž, draft a reply for it and show it before posting.

    line 67
  5. Distinctive:

    If the user asks to "preview," "dry run," "test," or "just show me," stop after presenting the table and drafts in step 4.

    line 88
  6. Distinctive:

    Never go from investigation straight to posting β€” step 4 confirmation is required every time, except dry-run where nothing is posted.

    line 98

β‰ˆ β€œCommon” 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. β€œVery 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

Take Feedback

Reviews the comments left on a GitHub pull request, investigates each one, and proposes a verdict β€” then waits for the user to confirm or override before reacting with πŸ‘ or πŸ‘Ž. Every πŸ‘Ž also gets a drafted reply explaining why, which the user sees and approves before it is posted. Judgment is based purely on whether a comment is functionally correct β€” not style, not phrasing, not preference.

Dialogue language & attention rules

Apply to every message to the user, and to drafted replies:

  • Minimize text: compress, drop fillers and unnecessary words. Stay as concise as possible.
  • Options over prose. Use AskUserQuestion for anything with a finite answer set; free text only for genuinely open questions.
  • A round earns its place: ask only when the answer changes a verdict or a reply.
  • No preamble, no retelling of the diff, no closer.

The user's attention is the scarce resource here; every extra word or round costs them more than it gives.

Prerequisites

  1. Confirm gh is authenticated: gh auth status. If not, tell the user to run gh auth login and stop.
  2. Identify the target PR. If owner/repo + PR number (or a PR URL) is missing, ask before anything else.

Workflow

1. Fetch review comments

gh api repos/{owner}/{repo}/pulls/{pr_number}/comments --paginate

Inline review comments. Each has id, path, diff_hunk, body, user.login, and in_reply_to_id if it's a thread reply.

2. Skip comments already handled

gh api repos/{owner}/{repo}/pulls/comments/{comment_id}/reactions

If the authenticated user already left +1 or -1, skip it.

3. Investigate every remaining comment

Read body and diff_hunk; for replies, also read the parent comment.

For each, work out:

  • Investigation β€” what the comment claims and whether it holds against the code.
  • Proposed verdict β€” πŸ‘ agree (real functional issue: bug, edge case, logic error, incorrect behavior) or πŸ‘Ž disagree (mistaken, misreads the code, no meaningful behavior change, or style/naming/preference with no functional claim).
  • Reasoning β€” one line.
  • Draft reply (πŸ‘Ž only) β€” a short reply to the reviewer stating why the comment doesn't hold, grounded in the code. Follow the language rules above: one or two sentences, factual, polite, no filler, no "Thanks for the comment". Point at the concrete reason (e.g. "x is null-checked on line 42, so this path can't throw.").

Post nothing yet.

4. Present everything for confirmation

One message, all comments, numbered table: #, author, short quote, verdict, one-line reasoning. Below the table, list every πŸ‘Ž draft reply verbatim under its comment number, so the user sees the exact text that would be posted.

Then confirm with AskUserQuestion, e.g.:

  • Post all as proposed
  • Edit replies / override verdicts (user specifies which)
  • Cancel

If the user overrides a πŸ‘ to πŸ‘Ž, draft a reply for it and show it before posting. If they edit a reply, use their text exactly. Do not proceed until confirmed.

5. Post the confirmed verdicts

Reactions:

gh api -X POST repos/{owner}/{repo}/pulls/comments/{comment_id}/reactions -f content='+1'
gh api -X POST repos/{owner}/{repo}/pulls/comments/{comment_id}/reactions -f content='-1'

Approved replies for πŸ‘Ž (as a reply in the comment's thread):

gh api -X POST repos/{owner}/{repo}/pulls/{pr_number}/comments/{comment_id}/replies -f body='<approved reply>'

πŸ‘ gets a reaction only, no reply. Never post a reply the user hasn't seen in its final form.

6. Dry-run mode

If the user asks to "preview," "dry run," "test," or "just show me," stop after presenting the table and drafts in step 4. No confirmation question, nothing posted.

7. Report back

One short line: reviewed / skipped / πŸ‘ / πŸ‘Ž / replies posted counts. No recap of the table.

Notes

  • Covers review comments (on the diff) only. If the user wants general PR comments too, also pull gh api repos/{owner}/{repo}/issues/{pr_number}/comments.
  • Posting needs write access; read-only tokens fail at step 5.
  • Never go from investigation straight to posting β€” step 4 confirmation is required every time, except dry-run where nothing is posted.

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.