Agent Skill
2/7/2026

skills-review

Analyze Claude Code skills for context efficiency. Checks description verbosity, file sizes, duplication, and extraction opportunities. Use when auditing skills or optimizing context usage.

T
timkhoury
1GitHub Stars
1Views
npx skills add timkhoury/.claude

SKILL.md

Nameskills-review
DescriptionAnalyze Claude Code skills for context efficiency. Checks description verbosity, file sizes, duplication, and extraction opportunities. Use when auditing skills or optimizing context usage.

name: skills-review description: > Analyze Claude Code skills for context efficiency. Checks description verbosity, file sizes, duplication, and extraction opportunities. Use when auditing skills or optimizing context usage.

Skills Review

Audit .claude/skills/ to optimize context consumption while maintaining effectiveness.

Quick Start

# Run full analysis (size + description quality)
.claude/skills/skills-review/analyze-skills.sh --all

# Run individual analyses
.claude/skills/skills-review/analyze-skills.sh --report        # Size only
.claude/skills/skills-review/analyze-descriptions.sh --report  # Description quality only

# Other output formats
.claude/skills/skills-review/analyze-skills.sh --json   # For scripting
.claude/skills/skills-review/analyze-skills.sh --csv    # For spreadsheets

# Analyze different locations
.claude/skills/skills-review/analyze-skills.sh --path ~/.claude/template/skills

Thresholds

MetricTargetWarningCritical
File size<4k>4k>10k
Description<200 chars>200>300
Table rows<20>20-

Review Process

Step 1: Run Size Analysis

.claude/skills/skills-review/analyze-skills.sh --report

The script identifies:

  • Critical skills (>10k chars) needing immediate reduction
  • Warning skills (>4k chars) worth optimizing
  • Verbose descriptions (>200 chars)
  • Large tables (>20 rows) that should be extracted

Step 2: Run Description Quality Analysis

.claude/skills/skills-review/analyze-descriptions.sh --report

The script checks:

  • Action verb: Does description start with third-person verb?
  • Trigger phrase: Contains "Use when..." or similar?
  • Length: 15-100 words (not too vague, not context waste)
  • Red flags: First-person language, generic terms

Step 3: Analyze Flagged Skills

For each flagged skill, check:

  1. Reference tables - Extract to PATTERNS.md or REFERENCE.md
  2. Code examples - Remove redundant examples
  3. Duplicate content - Compare with related skills

Step 4: Apply Optimizations

Extract tables:

<!-- Before: inline in SKILL.md -->
| Pattern | Example |
|---------|---------|
... 30 rows ...

<!-- After: reference file -->
**See `PATTERNS.md` for complete reference.**

Trim descriptions:

# Before (verbose)
description: >
  This skill helps you detect changes in the project .claude/ directory
  and propagate any template-worthy improvements back to ~/.claude/template/.
  Use this when saying "update template"...

# After (lean)
description: >
  Propagate project .claude/ improvements to ~/.claude/template/.
  Auto-invokes after editing skills/rules in both locations.

Use consolidated pattern for related skills:

## Workflow
/spec-review              # All analyses
/spec-review coverage     # Implementation coverage only
/spec-review tests        # Test coverage only

Skill Locations

LocationPurpose
.claude/skills/Project-specific
~/.claude/template/skills/Template (synced to projects)
~/.claude/skills/Global (user-wide)

Always fix global skills too. When issues are found in ~/.claude/skills/, suggest and apply fixes there as well - these affect all projects.

Common Violations

IssueImpactFix
Description >300 charsHarder trigger matchingTrim to essentials
SKILL.md >10k charsContext waste on activationExtract reference files
Inline tables >20 rowsContext bloatExtract to PATTERNS.md
"Use this tool to..." prefixWasted charsStart with action verb

Description Quality Checks

Good descriptions follow the WHAT + WHEN pattern for discoverability.

CheckRequirementWhy
Length20-100 tokensToo short = vague; too long = noise
Trigger scenariosContains "Use when..."Claude needs activation cues
Action verbStarts with third-person verbPOV consistency in system prompt
KeywordsIncludes user-facing termsNatural language discovery

Red flags (review manually after running analyze-skills.sh):

  • Description < 20 words (too vague for reliable activation)
  • No "Use when" or trigger phrases (Claude can't select correctly)
  • First-person language ("I will help you...")
  • Generic terms without domain specifics ("Helper for stuff")

After Completion

Record this review:

.claude/scripts/systems-tracker.sh record skills-review
Skills Info
Original Name:skills-reviewAuthor:timkhoury