andriyze/review-skills

Review Skills

Use this skill to review the entire ShakerScan skill and command surface like a code review, not as a lightweight summary.

Scope

Review all of these unless the user explicitly narrows scope:

• skills/*.md
• skills/**/SKILL.md
• skills/**/references/*
• skills/**/agents/openai.yaml
• .claude/commands/*.md
• .claude/agents/*.md

Review Goals

Look for:

• broken or stale file references
• unsupported or risky Claude Code frontmatter
• contradictory instructions across skills, commands, and agents
• prompt wording that increases false positives or weakens evidence standards
• missing machine-usable outputs
• missing fallback behavior
• missing checklists or hard gates for long tasks
• poor model/tool selection for the intended workload

Mandatory Checklist

Maintain this checklist in markdown while you work. Do not move on to synthesis or a final answer until every item is [x] or [n/a] with a short reason.

• [ ] Enumerate all in-scope skill, command, and agent files
• [ ] Read every in-scope skill file
• [ ] Read every in-scope command file
• [ ] Read every in-scope agent file
• [ ] Verify references between files resolve correctly
• [ ] Verify frontmatter fields are supported and coherent
• [ ] Verify each skill has a clear workflow, outputs, fallbacks, and guardrails
• [ ] Verify long-task checklist and hard-gate behavior where needed
• [ ] Produce findings ordered by severity with file references

Output

Return:

1. Findings first, ordered by severity, with file references.
2. Open questions or assumptions.
3. A short change summary or health summary.

If there are no issues, say so clearly and still note any residual risk or test gaps.

Rules

• Review behavior, not branding.
• Prefer concrete failures over stylistic nits.
• If a prompt is likely to create fabricated findings, call that out explicitly.
• Do not move on until the checklist is complete.