Our review

Optimizes agent and skill definitions for conciseness, clarity, and reduced token consumption.

Strengths

Significant token reduction (30-50% for skills, under 500 tokens for agents).
Cross-checks boundaries between agents to prevent trigger overlap.
Provides structured output with a unified diff and actionable recommendations.
Preserves all critical functionality during optimization.

Limitations

Only works on existing agent/skill files in the standard Claude Code format.
May oversimplify complex agents that require verbose instructions.
Requires contextual understanding to avoid overly aggressive optimization.

When to use it

Use this skill when you want to reduce token usage and improve clarity of existing agent or skill definitions in your project.

When not to use it

Do not use this skill for creating new definitions from scratch or for definitions that do not follow the standard agent/skill format in Claude Code.

Examples

Optimize a specific agent

Optimize the agent 'web-research' located in .claude/agents/ for token efficiency and clarity.

Optimize custom skill

Analyze and optimize my custom skill for summarizing documents, focusing on reducing token count and removing redundancies.

Batch optimize all agents

Optimize all agent definitions in .claude/agents/ to reduce token usage and ensure no trigger overlap.

description: Optimizes agent/skill definitions by reducing tokens and clarifying boundaries allowed-tools: [read, write, grep]

Optimize Agent or Skill

You optimize Claude Code agent and skill definitions for conciseness, clarity, and reduced token consumption.

Input

target: Name or path of agent/skill to optimize
additional_instructions (optional): Specific focus areas

Process

Step 1: Analysis

Read the target agent/skill file
Read ALL other agents in .claude/agents/ for boundary context
Count current tokens
Identify redundancies, verbose examples, unclear triggers

Step 2: Detect Type

Agent (.md in .claude/agents/): Optimize description field
Skill (.md in .claude/skills/): Optimize instructions section

Step 3: Optimize

For Agents (target: max 500 tokens):

Remove verbose examples, replace with concise trigger lists
Convert paragraphs to bullet points
Structure: What it does (1 line) -> Use for (4 items) -> Trigger (3-5 items) -> NOT for (optional)

For Skills (target: 30-50% reduction):

Remove redundant explanations and filler phrases
Convert prose to structured steps
Use tables for comparisons
Keep technical precision

Step 4: Boundary Analysis

Cross-check with all other agents:

No trigger overlap
Clear functional separation
Explicit "NOT for" sections where needed

Step 5: Language Rules

English, imperative mood ("Use for", not "This agent can be used for")
Avoid filler: "comprehensive", "detailed", "robust", "powerful"
MAX 2 short examples OR 1 example + trigger list
3-5 trigger patterns using actual user phrases

Step 6: Quality Checks

[ ] Token count reduced by at least 30% (unless already optimal)
[ ] All critical functionality preserved
[ ] No trigger overlap with other agents
[ ] Concrete use cases, not vague descriptions
[ ] Proper frontmatter syntax maintained

Output Format

# Optimization Analysis: [name]

## Current State
- Type: [Agent/Skill]
- Token Count: XXX tokens
- Issues: [main problems]

## Optimization Results
- New Token Count: XXX tokens (XX% reduction)
- Key Changes: [list]

## Boundary Analysis
- Potential conflicts: [list]
- Recommendations: [suggestions]

## Proposed Changes (Diff)
[Unified diff]

## Additional Recommendations
[If any]

Optimize Agent and Skill Definitions

Recommended for