Implement Agent Skills
Purpose
This skill helps implement new Agent Skills by first identifying the target AI coding tool environment to ensure the created skill is compatible with the user's workflow.
1. Identify AI Coding Tool
Check the project for indicators of AI coding tools:
- •Cursor: Presence of
.cursor/directory,.cursorrules, or.cursorignore. - •Claude Code: Presence of
.claude/directory orclaude.json. - •Gemini CLI: Presence of
gemini.jsonor specific environment variables.
Logic
- •Scan the workspace root for these indicators.
- •Analyze the findings:
- •If one tool is detected -> Proceed with that tool as the target.
- •If multiple tools are detected -> Ask the user which tool to target.
- •If no tools are detected -> Ask the user to specify the tool.
2. Interaction Strategy
If you need to ask the user, use the AskQuestion tool if available, or ask conversationally.
Example Question: "I detected multiple AI coding tools (Cursor and Claude Code). Which one should I target for this Agent Skill?"
- • Cursor
- • Claude Code
- • Gemini CLI
3. Implementation Context
Once the tool is identified, tailor the Agent Skill implementation accordingly:
- •
Cursor:
- •Target:
.cursor/skills/(project) or~/.cursor/skills/(global). - •Format: Standard
SKILL.mdwith YAML frontmatter.
- •Target:
- •
Claude Code:
- •Target:
.claude/skills/(project) or~/.claude/skills/(global). - •Format: Follow Claude Code skill conventions.
- •Target:
- •
Gemini CLI:
- •Follow Gemini CLI specific skill structure.
4. Standards & Templates
When implementing a new skill, always follow the established best practices and select the most appropriate template as a starting point.
Specification Compliance
Ensure the skill follows the official Agent Skills specification:
- •Naming: Directory and
SKILL.mdname must be kebab-case and match. - •Frontmatter: Include all required fields (
name,description). Use optional fields (license,compatibility,metadata,allowed-tools) if applicable. - •Structure: Use
scripts/,references/, andassets/subdirectories to offload details for progressive disclosure.
Validation
Before completing the task:
- •Check Syntax: Verify YAML frontmatter is valid.
- •Validate: If the
skills-reftool is available, runskills-ref validate ./<skill-directory>. - •Manual Review: Ensure the
descriptionis keyword-rich for discovery.
Best Practices
See references/best-practices.md for detailed naming, description, and progressive disclosure standards.
Available Templates
Select the template that best matches the skill's purpose:
| Template | Purpose |
|---|---|
| generic.md | Standard boilerplate for any skill. |
| cli-tool-wrapper.md | Wrapping CLI tools with help checks and safety. |
| sop-workflow.md | Checklists and SOPs for hybrid/manual processes. |
| library-upgrade.md | Safe, multi-step dependency upgrades. |
| code-review.md | Reviewing code against project standards. |
| security-audit.md | Systematic vulnerability scanning and reporting. |
| fixer-loop.md | Autonomous "test-analyze-fix" loops (TDD/Lint). |
| docs-sync.md | Keeping documentation in sync with source code. |