Process pr comments lite

Skill NaderNabil216/skillsmith/skills/process-pr-comments-lite

Install a versioned catalog of AI-agent skills into any CLI agent (Claude, Codex, Gemini, Kimi, GLM, opencode).

Install

npx -y skills add NaderNabil216/skillsmith --skill process-pr-comments-lite

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Fast, autonomous version of process-pr-comments. Given a PR/MR number, it implements the review comments that are logically valid against the code, saves recurring feedback as reusable rules, and stops — no prompts, no approval gates, no commit, no commit-message step. Use when asked to quickly or automatically apply PR/MR review comments without user input, e.g. "process PR 123 comments lite", "auto-apply the review comments on MR X", "just fix the valid PR comments".

SKILL.md

3.4 KB, as published. Nobody here has run it

Process PR / MR Review Comments — Lite

A stripped-down, autonomous take on process-pr-comments: given a PR/MR number, apply the review comments that are logically sound against the real code, record recurring feedback as rules, and stop. Nothing else.

For the full interactive workflow — batched per-comment approval, an implementation-plan gate, and a commit-suggest step — use process-pr-comments instead.

Rules

No user input. Take the PR/MR number from the request. Never ask questions, never wait for approval, never gate on a plan.
Never post to the forge and never commit. Edit files in the working tree only — leave staging and committing to the user.
Only apply what's logically sound. Validate every comment against the actual code; silently skip anything that's already handled, outdated, or wrong.

Steps

1. Resolve & fetch

Detect the forge from git remote get-url origin (github / gitlab / bitbucket) and use its CLI (gh / glab / bb), always passing the explicit PR number.
Check out the PR's source branch if not already on it (git fetch first if the branch is missing).
Fetch all unresolved comment threads, filtered to the needed fields (CLI --json/--jq). Walk each thread fully — the real request is often in a reply — and skip bot / system comments. If there are none, report "No unresolved comments on <PR|MR> #{n}." and stop.

2. Validate against the code

For each comment, Read the referenced code (inline: around the anchor line, ±~10 lines; general: locate the area) and decide whether the requested change is logically correct — a genuine fix or improvement that isn't already handled, outdated, or based on a misread. If a conventions file exists (AGENTS.md, CLAUDE.md, .github/copilot-instructions.md, .cursorrules), grep it for keywords relevant to the comment and read only the matched sections. Keep a one-line verdict per comment.

3. Implement the valid ones

Apply every comment that passed step 2 directly in the working tree, grouping edits by file. Skip the rest. Do not stage or commit.

4. Save the rules

Extract durable, checkable rules from what was applied — the underlying principle, not the one-off fix (e.g. "Composables accept modifier: Modifier = Modifier as the first optional parameter"). Append them to a cumulative docs/pr-review-rules.md (create it if missing), de-duped, under a PR-tagged heading. Add a one-line pointer to that file from the conventions file (priority AGENTS.md → CLAUDE.md → .github/copilot-instructions.md → cursor rules) if one isn't there already.

5. Report & stop

Print a short summary and end — no commit, no further steps:

Applied — one line per implemented comment: path:line — <gist>.
Skipped — one line per skipped comment: path:line — <one-line reason>.
Rules — the rules written to docs/pr-review-rules.md.