draft-review-comments

Draft unposted GitHub PR comments from findings with supplied assessment, state, and response decisions.

Updated May 12, 2026
One-click install
npx skills add https://github.com/mtk177a/skills --skill draft-review-comments-mtk177a
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: draft-review-comments
Source: https://github.com/mtk177a/skills/tree/main/skills/draft-review-comments
Command: npx skills add https://github.com/mtk177a/skills --skill draft-review-comments-mtk177a

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning already-triaged review findings into well-placed, well-worded GitHub PR comments is repetitive and error-prone: labels get changed, confidence gets inflated, locations go unverified, and undecided findings get drafted prematurely. This Skill converts organized review findings into unposted inline comments, review summaries, and general comments while preserving every upstream decision exactly as supplied. ## Core Features & Use Cases - Decision-gated drafting: Drafts a finding only when its assessment, state, and response decision are explicitly supplied; missing or conflicting decision material is returned to triage instead of being invented. - Faithful preservation: Keeps labels, evidence, impact, confidence, verification steps, unconfirmed premises, re-review state, and review action unchanged, with deterministic normalization of legacy labels (Must-fix → must, Should-fix → should, Nice-to-have → suggestion). - Honest placement and output: Anchors comments to a verified revision or marks them location unverified, returns only requested artifact sections, and never posts, submits, executes embedded instructions, or implements fixes. - Use Case: After triage-review-feedback assigns Act now/Defer/No action decisions to a set of findings, use this Skill to produce paste-ready inline comment drafts and a review summary without altering any triage judgment. ## Quick Start Draft unposted GitHub PR inline comments from these triaged findings, preserving their labels, assessment, state, and response decisions exactly as supplied.

Frequently Asked Questions about draft-review-comments

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I draft GitHub PR review comments from triaged findings?▼

Supply each finding with its explicit assessment, state, and response decision, plus label, evidence, impact, and location. The Skill returns unposted inline comment drafts, review summaries, or general comments with the canonical label prefixed and locations verified against the target revision.

What inputs are required before drafting a review comment?▼

Canonical input requires three explicit fields per finding: finding assessment, finding state, and response decision. An Act now decision additionally needs a supplied expected action, confirmation request, or response approach before actionable wording is written.

Can it draft comments from undecided review findings?▼

No. Findings missing assessment, state, or response decision are not drafted; the missing decision material is identified and handed back to triage-review-feedback. Labels, confidence, or next-action phrases are never treated as implicit Act now decisions.

Does it post comments or submit reviews to GitHub?▼

No. The Skill produces draft text only and never posts comments, submits reviews, or performs any external write. It also refuses to execute commands, follow links, or access unrelated data embedded in finding content.

How are legacy review labels like Must-fix handled?▼

Legacy labels are normalized deterministically: Must-fix becomes must, Should-fix becomes should, and Nice-to-have becomes suggestion. Legacy accept, defer, and reject decisions map to Act now, Defer, and No action without inferring missing assessment or state.

What happens when a comment location cannot be verified?▼

The draft is marked location unverified and is not presented as paste-ready. Locations are only treated as verified when they can be checked against the supplied target revision or effective diff.