Audit Agent Skills with skillkit core

Where this fits

You've installed a dozen skill packs, your agent's context window is bloating, and you suspect some skills are dead weight. Every session burns tokens on schemas and prompts you never use. You need a way to see what's actually loaded, what's conflicting, and what's costing you—without sending telemetry to a cloud service. skillkit core is a local-first CLI that gives you that visibility.

Why agents benefit

• Measure context cost before and after changes – Use skillkit context --save-baseline and --compare to see exactly how many tokens a new pack adds to always-loaded context.
• Detect overlapping triggers – skillkit conflicts finds skills that fire on the same keywords, preventing ambiguous agent behavior.
• Track real usage with private receipts – skillkit receipts shows which skills were actually invoked, so you can prune unused ones.
• Audit uninstalled skills or repos – Run skillkit audit <path> to evaluate a skill pack before you commit to installing it.
• Measure dead weight per skill – skillkit coverage <skill-path> reveals what portion of a skill's prompts and tools are ever used.

Practical scenarios

Before installing a new skill pack

You're evaluating a community skill pack. Instead of reading its README and guessing, run skillkit audit ./candidate-skill to see its structure, estimated context cost, and potential conflicts with your existing skills.

Debugging token burn after a recent update

Your agent's usage spiked after you added a new MCP server. Run skillkit context --mcp to see how much schema is always loaded, then compare against a baseline you saved earlier. Identify the culprit and adjust.

Cleaning up an inherited agent setup

You took over a project with 30 installed skills. Run skillkit stats to see which ones are actually used, then skillkit coverage on the top suspects. Remove the dead ones and cut your context bill by half.

Add it to your agent workflow

Install the skill with:

npx skills add crafter-station/skill-kit --skill core

Then start with a health check:

skillkit health

This verifies your skill database and installed-skill integrity before you dig into audits.

Compare before adopting

Before you rely on skillkit core, compare it against other skill-audit tools. Look for:

• Quality signals – Does it parse real session data or just static files?
• Maintenance freshness – Check the GitHub repo (currently 77 stars) for recent commits and issue responses.
• Alternatives – Some cloud-based observability tools offer similar metrics but require sending data externally.
• Workflow fit – Does it integrate with your existing CLI and session storage? skillkit reads local directories and agent session stores, so it fits local-first setups.

Why it is worth tracking

skillkit core is a practical, privacy-preserving answer to a growing problem: agent context bloat and skill overlap. With a modest but active community (77 stars) and a clear roadmap, it's worth evaluating when you hit the pain point of token costs or ambiguous agent behavior. The tool is designed to be read before auditing any skill pack, making it a foundational piece of your agent observability stack. For more details and community context, check the skill page on OpenAgentSkill.