agent-skills-setup（Agent 技能设置）

Agent Skills Setup Guide

This skill provides standardized instructions on how to install, structure, and configure custom skills for various AI agents. It covers directory paths, file requirements, and triggering mechanisms for global and project scopes.

🚀 Quick Start: IDE Migration Workflow

When user mentions IDE migration, follow this workflow:

Step 1: Ask for Migration Details

Ask the user:

1. Source IDE: Which IDE are you migrating FROM?
2. Target IDE: Which IDE are you migrating TO?
3. Content Types: What to migrate? (skills, rules, prompts, mcp, config, project) - default: all available

Supported IDEs:

• antigravity, claude, codex, copilot, cursor, windsurf
• jetbrains, openclaw, trae, trae-cn
• vscode, zed, neovim, emacs
• continue, aider, roo-code, cline, amazon-q

If the IDE is not in the list, follow the "Handling Unknown IDEs" section below.

Step 2: Preview Migration (Dry Run)

Always run dry-run first:

bash <skill-path>/scripts/smart-ide-migration.sh \
    --source <source-ide> \
    --target <target-ide> \
    --dry-run

Skill path varies by current IDE:

• Trae CN: ~/.trae-cn/skills/agent-skills-setup
• Trae: ~/.trae/skills/agent-skills-setup
• Claude Code: ~/.claude/skills/agent-skills-setup
• Antigravity: ~/.gemini/antigravity/skills/agent-skills-setup
• OpenClaw: ~/.openclaw/skills/agent-skills-setup

Step 3: Execute Migration

After user confirms the preview:

bash <skill-path>/scripts/smart-ide-migration.sh \
    --source <source-ide> \
    --target <target-ide> \
    --report ~/migration-report.txt

Step 4: Post-Migration Steps

Inform user about manual steps based on target IDE:

• VS Code Copilot: Update settings.json to reference migrated skills
• Cursor/Windsurf: Review .cursorrules/.windsurfrules files
• OpenClaw: Update openclaw.json and run openclaw doctor
• Trae: Restart IDE and verify in Skills Center
• Codex: Add agents/openai.yaml for UI visibility

Supported IDEs

Supported CLI Tools

🔍 Handling Unknown IDEs

When user mentions an IDE not in the supported list, AI should:

1. Search for information about the IDE's AI assistant configuration:

- Official documentation - GitHub repository - Community forums

1. Identify key configuration paths:

- Global config directory (usually in ~/.<ide-name>/ or ~/.config/<ide-name>/) - Project-level config files (usually .<ide-name>/ in project root) - Rules/instructions file format - MCP server configuration (if supported)

1. Ask the user to confirm findings:

   "I found that [IDE name] uses the following paths for AI configuration:
   - Global: [path]
   - Project: [path]
   - Rules file: [filename]
   
   Is this correct? Would you like me to proceed with migration using these paths?"

1. If unable to find information, ask user:

   "I couldn't find detailed configuration information for [IDE name].
   Could you help me by providing:
   1. Where does this IDE store its AI assistant configuration?
   2. Does it support custom rules/instructions files?
   3. Does it support MCP servers?

1. After gathering information, proceed with migration:

- Use the discovered paths - Document the new IDE for future reference

Content Types to Migrate

Migration Example

User: "Help me migrate from Trae CN to Claude Code"

AI Response:
1. Ask: "I'll help you migrate from Trae CN to Claude Code. What content would you like to migrate?"
   - Options: skills, rules, prompts, mcp, config, project (default: all)
   
2. Run dry-run:
   bash ~/.trae-cn/skills/agent-skills-setup/scripts/smart-ide-migration.sh \
       --source trae-cn --target claude --dry-run
       
3. Show preview and ask for confirmation

4. Execute migration:
   bash ~/.trae-cn/skills/agent-skills-setup/scripts/smart-ide-migration.sh \
       --source trae-cn --target claude --report ~/migration-report.txt
       
5. Show report and remind user to restart Claude Code session

0. Source of Truth Rule

Unless the user explicitly requests otherwise, treat Antigravity as the canonical source:

• Antigravity global skills: ~/.gemini/antigravity/skills/
• Mirror targets: Claude Code, OpenAI Codex, VS Code Copilot, Trae, Trae CN, and OpenClaw
• For directory-based agents, sync whole skill folders and remove extras not present in Antigravity
• For VS Code Copilot, flatten each SKILL.md into ~/.copilot-skills/<skill-name>.md
• For Codex, preserve internal directories such as .system/
• For OpenClaw, sync whole skill folders into ~/.openclaw/skills/ for shared skills and into <agent-workspace>/skills/ for per-agent overrides
• After changes, verify inventory parity and content parity rather than assuming the copy succeeded

1. Quick Reference: Skills Paths

2. Universal Skill Structure

Regardless of the agent, every skill should follow this anatomy:

<skill-name>/
├── SKILL.md (Required)
│   ├── YAML frontmatter (name, description)
│   └── Markdown instructions
├── scripts/ (Optional) - Executable automation
├── references/ (Optional) - Detailed docs and schemas
└── assets/ (Optional) - Templates and resources

3. Agent-Specific Deep Dives

For detailed configuration instructions, structure nuances, and UI requirements per agent, refer to:

• Antigravity: See antigravity.md
• Claude Code: See claude-code.md
• OpenAI Codex: See codex.md (requires agents/openai.yaml for UI features)
• OpenClaw: See openclaw.md (supports bundled/managed/workspace skills, per-agent workspaces, ClawHub, and installer metadata)
• Trae / Trae CN: See trae.md (supports skills CLI and UI import)
• VS Code Copilot: See vscode-copilot.md (supports multiple configuration levels)
• Public Distribution: See publishing.md for GitHub, skills.sh, and Awesome Copilot release paths

4. Setup Workflow

When installing a new skill:

1. Determine Scope: Should this be Global (all projects) or Project-level (shared in repo)?
2. Create Directory: Navigate to the appropriate path above and create the <skill-name> folder.
3. Draft SKILL.md: Ensure the description is comprehensive, as it is the primary trigger for ALL agents.
4. Agent-Specific Polish:

- For OpenClaw, decide whether the skill belongs in shared ~/.openclaw/skills/ or an individual agent workspace skills/ folder, and configure skills.entries / agents.list as needed - For Codex, add the agents/openai.yaml for UI visibility - For VS Code Copilot, add file reference to settings.json - For Trae, can use skills CLI or UI import

5. Recommended Maintenance Workflow

When updating a shared skill used across agents:

1. Edit the Antigravity copy first.
2. If needed, add or update helper scripts under scripts/.
3. Sync all target IDEs from Antigravity.
4. Verify:

- directory inventories match for OpenClaw managed skills after sync - directory inventories match for Claude, Trae, and Trae CN - directory inventories match for Codex after excluding .system/ - Copilot markdown files match Antigravity SKILL.md files - changed skills have matching content after sync

1. Only then report completion.

6. Sync Script

Use the bundled script for repeatable global sync operations:

~/.gemini/antigravity/skills/agent-skills-setup/scripts/sync-global-skills.sh

Examples:

# Sync all supported IDEs from Antigravity
~/.gemini/antigravity/skills/agent-skills-setup/scripts/sync-global-skills.sh

# Preview changes without modifying files
~/.gemini/antigravity/skills/agent-skills-setup/scripts/sync-global-skills.sh --dry-run

# Sync a subset of targets
~/.gemini/antigravity/skills/agent-skills-setup/scripts/sync-global-skills.sh --targets claude,codex,copilot,openclaw,trae,trae-cn

Behavior:

• Creates missing target directories
• Removes extra skills from mirror targets so they exactly match Antigravity
• Preserves Codex internal .system/
• Mirrors OpenClaw managed skills into ~/.openclaw/skills/
• Rebuilds Copilot markdown files from Antigravity SKILL.md
• Prints a concise verification summary

6.1 OpenClaw Automation Helpers

Use the bundled OpenClaw helpers when you need first-class OpenClaw setup rather than a plain file copy:

# Install OpenClaw if needed, sync skills, install declared dependencies,
# and patch ~/.openclaw/openclaw.json for shared + per-agent skills
bash ~/.gemini/antigravity/skills/agent-skills-setup/scripts/auto-configure-openclaw-skills.sh \
    --scope both \
    --agent work:~/.openclaw/workspace-work \
    --default-agent work

# Update the OpenClaw runtime, registry-managed skills, and mirrored local skills
bash ~/.gemini/antigravity/skills/agent-skills-setup/scripts/update-openclaw-skills.sh

The auto-configure helper:

• installs OpenClaw via the official installer when missing
• installs ClawHub when missing
• syncs shared skills into ~/.openclaw/skills/
• syncs per-agent overrides into each configured workspace skills/ directory
• installs skill dependencies declared in metadata.openclaw.install
• writes skills.load, skills.install, skills.entries, and agents.list settings into ~/.openclaw/openclaw.json
• supports --skip-doctor when you need a non-intrusive apply on a machine with an existing gateway service

The update helper:

• runs openclaw update for the runtime
• runs clawhub update --all for registry-managed workspace skills
• detects and applies local source-of-truth skill changes to shared/per-agent OpenClaw skill directories
• runs openclaw doctor after non-dry-run updates
• supports --skip-doctor for machine-safe update runs when zero gateway interaction is required

Use the ClawHub release helper when you want an exact publish command and metadata validation for a public OpenClaw release:

bash ~/.gemini/antigravity/skills/agent-skills-setup/scripts/prepare-clawhub-release.sh \
    --skill-dir ~/code/agent-skills-setup-public/agent-skills-setup \
    --slug agent-skills-setup \
    --name "Agent Skills Setup" \
    --version 1.0.0 \
    --tags latest,setup,openclaw

7. Smart IDE Migration

The Smart IDE Migration system provides intelligent migration between any supported IDE environments. You specify the source and target IDEs, and the tool handles the migration with real-time progress feedback and comprehensive reporting.

7.1 Supported IDEs

7.2 Basic Usage

bash ~/.gemini/antigravity/skills/agent-skills-setup/scripts/smart-ide-migration.sh \
    --source <source-ide> \
    --target <target-ide>

Required Parameters:

• --source <ide>: The IDE to migrate FROM
• --target <ide>: The IDE to migrate TO

7.3 Migration Options

7.4 Migration Content Types

The following content types can be migrated:

7.5 Migration Strategies

7.6 Example Workflows

Migrate All Content from Trae CN to Claude Code

bash ~/.gemini/antigravity/skills/agent-skills-setup/scripts/smart-ide-migration.sh \
    --source trae-cn \
    --target claude

Preview Migration (Dry Run)

bash ~/.gemini/antigravity/skills/agent-skills-setup/scripts/smart-ide-migration.sh \
    --source trae-cn \
    --target claude \
    --dry-run

Migrate Only Skills and Rules

bash ~/.gemini/antigravity/skills/agent-skills-setup/scripts/smart-ide-migration.sh \
    --source cursor \
    --target windsurf \
    --objects skills,rules

Migrate with Custom Workspace

bash ~/.gemini/antigravity/skills/agent-skills-setup/scripts/smart-ide-migration.sh \
    --source trae-cn \
    --target copilot \
    --workspace /path/to/project \
    --report ~/migration-report.txt

7.7 Post-Migration Steps

After migration, the report will indicate any manual steps required:

• VS Code Copilot: Update settings.json to reference migrated skills
• Cursor/Windsurf: Review .cursorrules or .windsurfrules files
• OpenClaw: Update openclaw.json and run openclaw doctor
• Trae: Restart IDE and verify in Skills Center
• Codex: Add agents/openai.yaml for UI visibility

7.8 Content Type Mapping by IDE

Different IDEs use different names and locations for similar content:

8. Migrating Skills Between Agents (Legacy)

From Antigravity to All Other Agents

# === To Trae (International) ===
for dir in ~/.gemini/antigravity/skills/*/; do
    skill_name=$(basename "$dir")
    mkdir -p ~/.trae/skills/$skill_name
    cp -r "${dir}"* ~/.trae/skills/$skill_name/
done

# === To Trae CN (China) ===
for dir in ~/.gemini/antigravity/skills/*/; do
    skill_name=$(basename "$dir")
    mkdir -p ~/.trae-cn/skills/$skill_name
    cp -r "${dir}"* ~/.trae-cn/skills/$skill_name/
done

# === To VS Code Copilot ===
mkdir -p ~/.copilot-skills
for dir in ~/.gemini/antigravity/skills/*/; do
    skill_name=$(basename "$dir")
    if [ -f "${dir}SKILL.md" ]; then
        cp "${dir}SKILL.md" ~/.copilot-skills/${skill_name}.md
    fi
done
# Then add to settings.json (see vscode-copilot.md)

# === To Claude Code ===
for dir in ~/.gemini/antigravity/skills/*/; do
    skill_name=$(basename "$dir")
    mkdir -p ~/.claude/skills/$skill_name
    cp -r "${dir}"* ~/.claude/skills/$skill_name/
done

# === To OpenAI Codex ===
for dir in ~/.gemini/antigravity/skills/*/; do
    skill_name=$(basename "$dir")
    mkdir -p ~/.codex/skills/$skill_name
    cp -r "${dir}"* ~/.codex/skills/$skill_name/
done

# === To OpenClaw (shared / managed) ===
for dir in ~/.gemini/antigravity/skills/*/; do
    skill_name=$(basename "$dir")
    mkdir -p ~/.openclaw/skills/$skill_name
    cp -r "${dir}"* ~/.openclaw/skills/$skill_name/
done

From Trae CN to Antigravity

for dir in ~/.trae-cn/skills/*/; do
    skill_name=$(basename "$dir")
    mkdir -p ~/.gemini/antigravity/skills/$skill_name
    cp -r "${dir}"* ~/.gemini/antigravity/skills/$skill_name/
done

9. Configuration Priority (All Agents)

10. Quick Migration Commands

One-command migration to all agents

# Create all target directories
mkdir -p ~/.trae/skills ~/.trae-cn/skills ~/.copilot-skills ~/.claude/skills ~/.codex/skills ~/.openclaw/skills

# Copy to all agents
for dir in ~/.gemini/antigravity/skills/*/; do
    skill_name=$(basename "$dir")
    
    # Trae
    mkdir -p ~/.trae/skills/$skill_name && cp -r "${dir}"* ~/.trae/skills/$skill_name/
    
    # Trae CN
    mkdir -p ~/.trae-cn/skills/$skill_name && cp -r "${dir}"* ~/.trae-cn/skills/$skill_name/
    
    # VS Code Copilot (flat structure)
    [ -f "${dir}SKILL.md" ] && cp "${dir}SKILL.md" ~/.copilot-skills/${skill_name}.md
    
    # Claude Code
    mkdir -p ~/.claude/skills/$skill_name && cp -r "${dir}"* ~/.claude/skills/$skill_name/
    
    # OpenAI Codex
    mkdir -p ~/.codex/skills/$skill_name && cp -r "${dir}"* ~/.codex/skills/$skill_name/

    # OpenClaw managed skills
    mkdir -p ~/.openclaw/skills/$skill_name && cp -r "${dir}"* ~/.openclaw/skills/$skill_name/
done

echo "Migration complete! Don't forget to update VS Code settings.json"

11. Operational Notes

• Prefer rsync -a --delete over ad hoc copy loops when exact mirror behavior is required.
• Avoid destructive cleanup commands when a mirror sync can express the same intent more safely.
• If Trae international is not installed yet, creating ~/.trae/skills/ is sufficient for pre-seeding the directory.
• If a target contains system-managed content, add an explicit preserve rule before syncing.
• For OpenClaw multi-agent setups, remember that shared skills live in ~/.openclaw/skills/ while per-agent overrides live under each agent workspace skills/ directory.
• OpenClaw skill env injection from skills.entries.*.env only applies to host runs; sandboxed agents need matching agents.defaults.sandbox.docker.env or agents.list[].sandbox.docker.env when a skill must execute inside the sandbox.

12. Public Release Workflow

If the goal is to publish a skill so more people can find and install it:

1. Keep the Antigravity copy as the authoring source.
2. Export the target skill into a public GitHub repository skeleton.
3. Add a repository README with install commands and compatibility notes.
4. Publish the repository publicly on GitHub.
5. Optionally increase discovery by:

- listing or sharing it via skills.sh - contributing it to github/awesome-copilot - posting examples and screenshots in the repo README and release notes

Use the bundled export helper:

bash ~/.gemini/antigravity/skills/agent-skills-setup/scripts/export-public-skill.sh \
    --skill agent-skills-setup \
    --output ~/tmp/agent-skills-setup-public \
    --repo your-github-name/agent-skills-setup

The export helper copies the selected skill into a publishable repository layout and generates a starter README.md.

13. Publishability Criteria

Before publishing a skill publicly, verify that it:

• solves a specific problem rather than repeating generic model behavior
• does not depend on private local paths without documenting replacements
• includes clear install instructions for at least one agent ecosystem
• explains what the skill does, when to use it, and any safety boundaries
• avoids bundling sensitive credentials, internal URLs, or proprietary assets