skills/review-steps/SKILL.md
Structured review for polishing documents. Fixes language, improves clarity, checks structural consistency, and compares against best practice. Use when a draft has structure but needs a thorough review pass.
npx skillsauth add nakane1chome/claude-skills review-stepsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
This skill is for polishing existing documents at $ARGUMENTS — not generating structure (flesh-out) or critiquing substance (strong-edit).
Stop after each stage and have changes reviewed with user.
Feedback loop: When the user gives feedback on a stage, revise that stage's output by synthesizing the feedback with your review findings. Don't just acknowledge the feedback or apply it verbatim — incorporate it into the review work you've already done and present updated suggestions. Only move to the next stage when the user approves.
Note: Review improves what exists within its structure. The agent handles mechanical checks and research; the developer holds final authority on judgment calls. When a suggestion changes meaning, ask — don't assume.
See
responsibilities.mdfor the full agent/developer ownership matrix.
Read and understand the document (developer confirms)
Review for language and consistency (agent leads, developer approves)
Review for conceptual clarity (agent leads, developer approves)
Review vs relevant structure (agent leads, developer approves)
Review vs industry best practice (agent assists, developer leads)
Tidy up (agent leads, developer approves)
Verify links and claims (agent leads, developer approves)
Why this stage exists: Agent-sourced references can be fabricated. A hallucinated URL with a plausible domain name will survive every other stage because reviewers (human and agent) evaluate structure, argument, and voice, not link targets. Verification must be an explicit step, not assumed.
This skill sits in the middle of the composition pipeline: flesh-out -> review-steps -> strong-edit -> agent-optimize. Use it after a document has structure, before it needs critical evaluation.
| Document State | Use | |----------------|-----| | Raw notes, bullets, stream of consciousness | flesh-out | | Draft with structure, needs polish | review-steps | | Complete draft needing critical evaluation | strong-edit | | Finalized document needs agent-friendly restructuring | agent-optimize |
tools
Helps with things
data-ai
Critical editorial analysis of articles. Examines structure, argument strength, relevance, and readability. Use for substantive editing - challenging what's said and how, not just polish.
testing
Review a document against its SDLC lifecycle context — assess completeness and check consistency with parent documents in the document hierarchy.
development
Reviews a SKILL.md file or skill directory for quality, correctness, and alignment with Claude Code skill conventions. Use when you've written a new skill and want to check it before committing, or when evaluating an existing skill for improvements.