It exposes a hidden reality of agent skill management—most skills on disk are dead copies or unfindable—and provides measurable before/after metrics, with no API keys or model calls required.
This tells you which ones — and fixes them.
Works with Claude Code, OpenClaw, Hermes, or anything that loads SKILL.md files.
Imagine your agent has a toolbox with 200 tools. Every tool has a label on it.
When you ask the agent to do something, it doesn't look inside the tools. It reads the labels and picks one.
Two things go wrong, and you can't see either of them:
🧰 Half the tools aren't actually in the box. They're in a drawer, a spare box in the garage, or they're copies of the same hammer. You think you have 200. You might have 80.
🏷️ Half the labels are unreadable. A tool labelled "Helper mode" will never be picked for "fix my leaky tap" — because nothing in the label matches what you asked for.
This tool opens the box, counts what's really there, and tells you which labels nobody can read. Then it writes you a to-do list for fixing them.
No setup. No install. No API key. It ships with four example skills so you can see exactly what it does before pointing it at anything of yours.
git clone https://github.com/rushindrasinha/skill-audit-router.git
cd skill-audit-router
python3 skill_audit.py --roots examples/fixture-skills --include-workspace
That's the screenshot above. Four example skills, each teaching one lesson:
| Example skill | Score | What it shows you |
|---|---|---|
good-example | 86 🟢 | Says what it makes, lists the exact words a user would type, and says what it's not for |
confusable-a | 60 🟡 | Nearly identical to its twin — the tool catches the collision |
confusable-b | 60 🟡 | …the other half of that pair, flagged at margin 0.038 |
weak-example | 21 🔴 | Labelled "Helper mode." — unfindable, and the report says exactly why |
Why
--include-workspace? The examples live in this repo, not in your agent's real skills folder. By default the audit ignores anything your agent can't actually load — which is the whole point. This flag says "look at them anyway."
Run it on a
Find out which of your agent's skills can actually be found, then route a query to the right one. Claude Code / OpenClaw / Hermes. Pure stdlib.
Similar projects matched by category, topics, and programming language.
Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.
Kimi K3 is an open-weight, 2.8T-parameter native multimodal agentic model with a 1M-token context window, designed for frontier coding, knowledge work, and reasoning tasks.
A browser-based first-person shooter built entirely with procedural generation and orchestrated AI agents, featuring 55k lines of Three.js/WebGL2 code and no art assets.