arvindand/agent-skills
Overview
This skill helps create, fix, and validate reusable agent skills for Claude, Copilot, and other assistants. It focuses on making skills discoverable, reliable under pressure, and easy to maintain. Use it to convert session workflows into repeatable skills, audit trigger coverage, and run automated checks for structure and token budgets.
How this skill works
The skill runs lightweight validation scripts to catch mechanical issues (compliance, token counts, missing triggers) and performs a qualitative review of the skill files and references. It evaluates trigger wording, allowed tools, workflow clarity, and recovery paths, then returns prioritized, actionable recommendations. It also guides creation from a live session by assessing repeatability and generalizability before generating a reusable skill.
When to use it
- Create a skill from a recent session or workflow
- Build a new skill to automate a repeating multi-step task
- Fix a skill that fails to trigger or misbehaves
- Audit skills for discovery, token budgets, and tooling scope
- Check character/token budget across all skills
Best practices
- Start the description with trigger phrases users actually say and avoid summarizing workflow
- Keep the main skill document concise; move long resources to references
- Use progressive disclosure: metadata first, body next, bundled resources last
- Pick one default approach and provide an escape hatch rather than many equal options
- Run automated checks (compliance, token count, trigger detection) before publishing
Example use cases
- Turn a repeated DB setup or deployment sequence from a chat into a reusable skill
- Diagnose why a skill didn't activate and update trigger wording to match user language
- Audit a skills collection for a 15K character/token limit and split references where needed
- Create a discovery-based GitHub CLI skill that finds command usage via the CLI help output
- Pressure-test a workflow to ensure the skill enforces desired discipline under time constraints
FAQ
Assess repeatability, non-trivial coordination, domain expertise captured, and generalizability. If you can answer who will reuse it and what repeats, it’s likely worth extracting.
What if the skill keeps failing to trigger?
Check trigger wording for exact user phrases and synonyms, ensure the description includes those terms, and re-run a trigger-detection script to confirm coverage.
