skill-audit
sharbelxyz/skill-audit
Skill that audits every installed agent skill for trigger wording, token cost and shipped code
skill-audit is a Python-backed skill that scans the SKILL.md files installed on an agent and reports which ones can never fire, what they cost in tokens and which ship runnable code. It installs into ~/.hermes/skills and also works with other agents that read SKILL.md.
What skill-audit does
Skills load only their name and description at startup, so a description that says what the skill is but never when to use it gives the agent nothing to match. skill-audit scans a skills directory and flags that case as NEVER SAYS WHEN. It also flags bodies over 5,000 tokens or 500 lines, names and descriptions longer than the 64 and 1,024 character limits in the Agent Skills specification, and folders that ship executable files.
The script prints a summary with the number of skills scanned, how many have flagged trigger wording, the reported startup context in tokens, and the total if every skill loaded. Token counts are measured with the tiktoken o200k_base encoding when tiktoken is installed, and the header states whether the figure was measured or estimated. The README cautions that NEVER SAYS WHEN is a text match rather than a judgement, and that the startup figure is an approximation. Run it by asking your agent, or directly with python3 scripts/audit.py and an optional --json flag.
Key features
- Flags skills whose descriptions never say when to use them
- Flags skill bodies over 5,000 tokens or 500 lines
- Flags names and descriptions over the Agent Skills specification limits
- Flags skill folders that ship executable code
- Measured token counts with tiktoken, with the mode stated in the output header
- JSON output option
When to use it
- Find installed Hermes skills whose descriptions never tell the agent when to run them
- Estimate how much context your installed skills cost at startup
- Review which skills contain scripts before an agent runs them again
Who it is for: Hermes Agent users with a growing skills folder who want to find broken, oversized or code-shipping skills.
How it fits with Hermes Agent
Installs into the Hermes skills directory by default and also works with other agents that read SKILL.md, by pointing it at ~/.claude/skills or ~/.agents/skills.
How to install skill-audit
These commands are copied from the project's README. Check the repository for the latest steps before you run them.
git clone https://github.com/sharbelxyz/skill-audit ~/.hermes/skills/skill-audit
pip install tiktokenRequirements: Python 3; tiktoken is optional and gives exact token counts
FAQ
What is skill-audit?
skill-audit is a skill that audits every skill installed on your agent. It finds skills that can never fire, measures their token cost and flags any that ship runnable code.
Does skill-audit work with Hermes Agent?
Yes. The README installs it into ~/.hermes/skills and runs the audit script against that folder. It also works with Claude and other agents that read SKILL.md.
How do I install skill-audit?
Clone the repository into ~/.hermes/skills/skill-audit, then ask your agent to audit your skills. Installing tiktoken with pip is optional and gives exact token counts.
Similar skills for Hermes Agent
All skillsHermes Agent skill for controlling an Android phone over ADB with an observe-first safety loop
gokulsvision Crewm8 Social Media Manager Skill Graph37 agent-agnostic SKILL.md files for social media strategy, creation, publishing and analytics
pingchesu GSD HermesHermes-first distribution of Get Shit Done with strict per-agent model routing
imphillip SoulTavernImport SillyTavern V2 character cards as SOUL.md personas for Hermes Agent and OpenClaw
aakashg Hermes PM ToolkitSelf-writing Hermes skills for product managers: competitive intel, signal log and decision log
AlekseiUL Agentic Business AnalysisHermes skill and methodology for mapping business processes and scoping safe AI-agent automation pilots
Related guides: What is Hermes Agent? · SOUL.md for Hermes Agent: what it is and how to write one