Agent skill

qual

Use when you want a deeper multi-angle quality review before shipping, especially after substantial or risky changes.

Stars 163
Forks 31

Install this agent skill to your Project

npx add-skill https://github.com/majiayu000/claude-skill-registry/tree/main/skills/other/other/qual

SKILL.md

target_path = $ARGUMENTS

If target_path provided, analyze that path. Otherwise, files changed since the default branch. Full-codebase analysis requires explicit user request.

Architecture

You coordinate specialist teammates who analyze in independent context windows. Separation matters — your implementation bias doesn't leak into their findings, and their diverse perspectives catch what a single pass misses.

Two waves: detect then simplify. Wave 2 exists because LLMs reliably over-engineer fixes — even your own wave 1 fixes are suspect and need an independent simplification pass.

Wave 1: Detect

Spawn specialist teammates from ${CLAUDE_SKILL_DIR}/agents/. Each teammate gets the target file list and the project's detected stack.

Teammate Agent file Focus
Skeptic skeptic.md Bugs, security, performance, correctness
Silent-failure hunter silent-failure.md Hidden errors, swallowed exceptions
Pattern harmonizer patterns.md Divergent implementations, inconsistent patterns
Comment auditor comments.md Misleading/stale/useless comments

Wave 1 teammates are read-only — analysis tools only, no editing. Mixing analysis and editing in one pass leads to fixes that miss the bigger picture.

The Simplifier runs in wave 2, not here.

Actionability Gate

Every finding must pass all five criteria. This filter prevents the report from being dominated by noise — LLM analysis without it produces 60%+ false positives:

  1. Scoped — within the target path or recent changes
  2. Specific — points to a concrete code path, not a general observation
  3. Certain — "this WILL cause X", not "this could theoretically..."
  4. Non-intentional — not a deliberate pattern justified by comments, tests, or conventions
  5. Discrete — has a single, focused fix (not "rewrite this module")

Deduplicate across lenses: when multiple teammates flag the same underlying issue, merge into one finding citing all relevant lenses. Keep the highest severity.

Triage

Severity Boundary
CRITICAL Security vulnerabilities, data loss risks, silent failures hiding bugs
HIGH Logic errors, missing error handling on critical paths
MEDIUM Pattern inconsistencies, genuine maintainability issues
LOW Minor improvements
IGNORE False positives, justified patterns, style-only concerns

Conflict resolution: Safety > style. Cohesion > fragmentation. Direct code > abstraction. Pattern consistency > local optimization. Type annotations > JSDoc.

Approval Gate

Present findings grouped by severity. Each finding: severity, confidence, location, issue, recommended fix, source lens(es).

Stop. Wait for user to select which fixes to apply.

After approval: implement selected fixes, verify with quality gates. If a fix breaks something, revert rather than cascade.

Wave 2: Simplify

Spawn one Simplifier teammate (simplifier.md) on the post-fix codebase. Same actionability gate, triage, and approval gate flow as wave 1.

Expand your agent's capabilities with these related and highly-rated skills.

Didn't find tool you were looking for?

Be as detailed as possible for better results