AgentSkillsCN

claude-setup-optimizer

分析Claude Code的变更日志，审查您当前的智能体/技能配置，并根据新增功能提出改进建议。当您被要求“优化Claude配置”、“检查Claude更新”、“优化我的智能体”、“同步Claude变更日志”，或输入“/optimize-setup”时，可使用此功能。

SKILL.md
--- frontmatter
name: claude-setup-optimizer
description: Analyzes Claude Code changelog, reviews your current agents/skills setup, and recommends improvements based on new features. Use when asked to "optimize claude setup", "check for claude updates", "improve my agents", "sync with claude changelog", or "/optimize-setup".
user-invocable: true
context: fork
allowed-tools:
  - WebFetch
  - WebSearch
  - Read
  - Write
  - Edit
  - Glob
  - Grep
  - Bash
  - Task
  - AskUserQuestion
  - TaskCreate
  - TaskUpdate
  - TaskList
hooks:
  Stop:
    - hooks:
        - type: command
          command: "cd \"$HOME/Library/Mobile Documents/com~apple~CloudDocs/claude-setup\" && { git diff --quiet && git diff --cached --quiet && [ -z \"$(git ls-files --others --exclude-standard)\" ] && echo 'No changes to commit'; } || { git add -A && git commit -m 'feat: apply claude-setup-optimizer recommendations' && git push origin master && echo 'Committed and pushed to GitHub'; }"

Claude Setup Optimizer

Automatically analyzes the Claude Code changelog and recommends improvements to your agents, skills, and configuration based on new features.

What This Skill Does

  1. •Fetches Latest Changelog - Gets the most recent Claude Code changelog from official sources
  2. •Analyzes Your Setup - Reviews all your agents, skills, commands, and configuration
  3. •Identifies Opportunities - Finds ways your setup could benefit from new Claude features
  4. •Recommends Improvements - Provides specific, actionable recommendations
  5. •Implements Changes - Optionally applies improvements automatically

Paths

Portable iCloud Path Variable

IMPORTANT: Always use $HOME in bash commands and ~/ in documentation to ensure portability across different machines/users.

Define this variable at the start of any bash operations:

bash
ICLOUD_SETUP="$HOME/Library/Mobile Documents/com~apple~CloudDocs/claude-setup"

Source of Truth Structure:

code
~/Library/Mobile Documents/com~apple~CloudDocs/claude-setup/
├── skills/       ← All skills
├── agents/       ← All agents
├── commands/     ← All commands
├── memory/       ← Memory files (core-memory.json, etc.)
└── settings.json ← Settings

IMPORTANT: Always read from AND write to the iCloud path directly. Do NOT use symlink paths like ~/.claude/ - use the iCloud path to ensure:

  • •Consistent behavior regardless of symlink state
  • •Changes persist correctly to iCloud
  • •Syncs across all devices

Background (FYI only)

Claude Code searches for skills/agents in multiple locations, but symlinks point everything to iCloud:

  • •~/.claude/skills → iCloud
  • •~/.claude/agents → iCloud
  • •~/.claude/commands → iCloud

Workflow

Step 1: Fetch Claude Code Changelog

Fetch the latest Claude Code changelog using strategies that handle large files:

Primary Source - GitHub CLI (Recommended):

Use the gh CLI to fetch recent releases which avoids large file issues:

bash
# Get the 5 most recent releases with full details
gh release list --repo anthropics/claude-code --limit 5

# Get detailed notes for a specific release
gh release view <tag> --repo anthropics/claude-code

Secondary Source - GitHub Releases Page:

Use WebFetch on the releases page (summarized, not full changelog):

code
https://github.com/anthropics/claude-code/releases

Prompt: "Extract the 3 most recent releases with their version numbers, dates, and key new features. Focus on: new tools, skill/agent updates, MCP changes, configuration options."

Tertiary Source - Web Search:

Search for "Claude Code changelog 2026" or "Claude Code new features" for recent announcements.

IMPORTANT: Avoid fetching the full CHANGELOG.md file directly - it exceeds token limits (~32K tokens). If you must read it, use the Read tool with limit parameter to read only the first 500 lines:

code
Read file with limit=500 to get only recent entries

Extract key information:

  • •New features and capabilities
  • •New tools or MCP changes
  • •Skill/agent system updates
  • •Configuration options
  • •Breaking changes
  • •Best practice updates

Step 2: Analyze Current Setup (Parallel)

IMPORTANT: Always use the iCloud path as the source of truth: ~/Library/Mobile Documents/com~apple~CloudDocs/claude-setup

Use parallel Task agents to inventory everything concurrently. Launch these in a single message:

  1. •

    Skills inventory agent - Read all SKILL.md frontmatter from every skill directory under skills/. Extract name, description, allowed-tools, context, model, and any other frontmatter fields. Return the complete frontmatter for each skill.

  2. •

    Agents inventory agent - Read all agent .md files from agents/ (including subdirectories like review/). Extract name, description, allowed-tools, disallowedTools, model, color, skills, and any other frontmatter fields. Return the complete frontmatter for each agent.

  3. •

    Config inventory agent - Read settings.json from the iCloud path and ~/.claude.json. Also list all commands in commands/. Return the full configuration content.

All three agents should use subagent_type="Explore" and run_in_background=true.

Wait for all agents to complete, then proceed to Step 3 with the combined inventory data.

Step 3: Compare and Identify Opportunities

Cross-reference changelog features against your current setup:

Feature Categories to Check:

  1. •New Tools: Are there new tools you're not using that could benefit your workflows?
  2. •Skill System Updates: Has the skill format changed? Are there new frontmatter options?
  3. •Agent Improvements: New agent capabilities or patterns?
  4. •MCP Updates: New MCP servers or protocol changes?
  5. •Performance Optimizations: Parallelization, caching, or efficiency improvements?
  6. •Security Features: New permission patterns or security best practices?
  7. •Configuration Options: New settings that could improve your experience?

Analysis Checklist:

For each skill/agent, check:

  • • Uses latest skill format (SKILL.md with frontmatter)
  • • Takes advantage of parallel tool calls where applicable
  • • Uses appropriate model selection (haiku for fast tasks, etc.)
  • • Has proper error handling patterns
  • • Follows current best practices for triggers/descriptions
  • • Uses new tools/features that weren't available before

Step 4: Generate Recommendations

Create a structured report with:

Priority Levels:

  • •HIGH: Security improvements, breaking change migrations, significant efficiency gains
  • •MEDIUM: New feature adoption, best practice updates
  • •LOW: Nice-to-have improvements, style updates

Recommendation Format:

markdown
## [Priority] Recommendation Title

**Affected Items:** skill-name, agent-name, etc.

**Current State:**
Description of how it works now

**Recommended Change:**
What should be changed and why

**New Feature Reference:**
Link to changelog or documentation

**Implementation:**
Specific steps or code changes needed

**Effort:** Low/Medium/High

Step 5: Present and Confirm

Present findings to the user:

code
# Claude Setup Optimization Report

## Changelog Summary
- Version X.Y.Z released on DATE
- Key new features: [list]

## Your Setup Analysis
- X skills analyzed
- Y agents analyzed
- Z commands analyzed

## Recommendations Found: N

### High Priority (M items)
1. [Brief description]
2. ...

### Medium Priority (N items)
1. ...

### Low Priority (P items)
1. ...

Would you like me to:
1. Show detailed recommendations for all items
2. Show only HIGH priority recommendations
3. Auto-implement all HIGH priority changes
4. Auto-implement specific recommendations
5. Skip implementation (just review)

Step 6: Implement Improvements with Parallel Agents

IMPORTANT: iCloud is the source of truth. All changes MUST be made directly in the iCloud path to ensure they sync across devices.

CRITICAL: Use parallel Task agents to implement changes concurrently. Group approved changes into independent work streams and launch them as parallel agents. This dramatically speeds up implementation.

6a. Back up affected files first

bash
ICLOUD_SETUP="$HOME/Library/Mobile Documents/com~apple~CloudDocs/claude-setup"
mkdir -p "$ICLOUD_SETUP/backups/$(date +%Y%m%d-%H%M%S)"
cp <file> "$ICLOUD_SETUP/backups/$(date +%Y%m%d-%H%M%S)/"

6b. Group changes into independent work streams

Categorize approved changes into groups that can run in parallel without conflicts (no two agents should edit the same file):

  • •Skills agent: All SKILL.md frontmatter and body changes
  • •Agents agent: All agent .md file changes
  • •Config agent: settings.json, hooks, and other config changes
  • •Filesystem agent: Directory moves, file deletions, archive operations

If a group touches too many files, split further (e.g., separate M&A skills from dev skills).

6c. Launch parallel agents

Use the Task tool to launch multiple agents in a single message with multiple tool calls. Each agent gets a detailed prompt listing exactly which files to change and what to do.

Example pattern:

code
# In a SINGLE message, launch all these Task calls simultaneously:

Task(subagent_type="general-purpose", description="Fix skill frontmatter", prompt="Read and edit these SKILL.md files: [list]. For each: [specific changes]...", run_in_background=true)

Task(subagent_type="general-purpose", description="Update agent files", prompt="Read and edit these agent .md files: [list]. For each: [specific changes]...", run_in_background=true)

Task(subagent_type="general-purpose", description="Update config files", prompt="Read and edit settings.json: [specific changes]...", run_in_background=true)

Rules for parallel agents:

  • •Each agent prompt must be self-contained with ALL context needed (file paths, exact changes)
  • •No agent should edit a file that another agent is also editing
  • •Use run_in_background=true to launch them concurrently
  • •Wait for all agents to complete, then verify results
  • •If a change depends on another change, put them in the same agent or run sequentially

6d. Verify all changes

After all agents complete, spot-check key files to confirm changes were applied correctly. Read the frontmatter of a few modified files to verify.

6e. Git commit and push (automatic)

A Stop hook in this skill's frontmatter automatically commits and pushes all changes to the GitHub backup repo (github.com/escotilha/claude) when the skill finishes. No manual git commands needed.

Example Recommendations

Example 1: New Tool Usage

markdown
## [MEDIUM] Add WebSearch to financial-data-extractor skill

**Affected Items:** financial-data-extractor

**Current State:**
Skill only uses WebFetch for external data

**Recommended Change:**
Add WebSearch to allowed-tools to enable broader research capabilities when extracting financial data from unfamiliar sources.

**New Feature Reference:**
Claude Code v2.x added native WebSearch tool

**Implementation:**
Add `WebSearch` to allowed-tools in frontmatter:

```yaml
allowed-tools:
  - Read
  - WebFetch
  - WebSearch # NEW
```

Effort: Low

code

### Example 2: Parallel Execution
```markdown
## [HIGH] Enable parallel agent execution in project-orchestrator

**Affected Items:** project-orchestrator.md

**Current State:**
Agent launches subagents sequentially

**Recommended Change:**
Update orchestrator to launch independent agents (frontend, backend, database) in parallel using multiple Task tool calls in single message.

**New Feature Reference:**
Task tool now supports parallel agent launches

**Implementation:**
Update the workflow section to specify parallel launch pattern for independent tasks.

**Effort:** Medium

Example 3: Skill Format Migration

markdown
## [HIGH] Migrate legacy .skill files to SKILL.md format

**Affected Items:** financial-data-extractor.skill

**Current State:**
Using legacy .skill format

**Recommended Change:**
Migrate to new SKILL.md format with YAML frontmatter for better Claude recognition and metadata support.

**New Feature Reference:**
SKILL.md is now the recommended format

**Implementation:**

1. Rename file to SKILL.md
2. Add YAML frontmatter with name, description, allowed-tools
3. Update content to match new format

**Effort:** Low

Triggers

This skill activates on:

  • •"optimize my claude setup"
  • •"check claude code updates"
  • •"improve my agents"
  • •"sync with claude changelog"
  • •"what's new in claude code"
  • •"update my skills for new features"
  • •"/optimize-setup"

Error Handling

Large File Errors

If you encounter "File content exceeds maximum allowed tokens" when fetching changelog:

  1. •Use gh CLI instead - Fetch releases via gh release list and gh release view
  2. •Read with limits - Use Read tool with limit=500 to get only recent entries
  3. •Use WebSearch - Search for "Claude Code new features 2026" for recent announcements
  4. •Fetch releases page - WebFetch the releases page with a focused prompt

Never attempt to fetch the full CHANGELOG.md via WebFetch - it's too large.

Notes

  • •Always preserve existing functionality when making changes
  • •Create backups before modifying files
  • •Test changes before committing
  • •Some recommendations may require manual review
  • •Keep track of which changelog versions have been reviewed to avoid duplicate work
  • •All paths use $HOME variable for cross-machine compatibility