By name: harness-audit description: Audits ~/.claude/CLAUDE.md for pointer purity and hook integrity. Computes a 0-10 score, classifies every line as behavioral/pointer/fat, and tracks trend history. Does not edit the harness. version: 1.0.0 author: Hermes Agent license: MIT metadata: hermes: tags: [claude-code, wrapper, slash-command] related_skills: [claude-code-skills]
Harness Audit Claude Code Skill Wrapper
Purpose
This is a Hermes-native slash-command wrapper for Kevin's Claude Code skill harness-audit.
Claude remains the source of truth. This wrapper exists so Kevin can invoke the workflow in Hermes as:
/harness-audit <task or target>
Required behavior
When this skill is invoked:
- Immediately read the source Claude skill:
/Users/kevin/.hermes/external-repos/claude-code-skills/harness-audit/SKILL.md
- Treat that Claude
SKILL.mdas the authoritative workflow for this task. - Follow the Claude skill exactly where compatible with Hermes tools and safety boundaries.
- Preserve any user instruction supplied after
/harness-auditas the task target. - If the Claude skill references supporting files under
references/,templates/,scripts/, orassets/, resolve them relative to:/Users/kevin/.hermes/external-repos/claude-code-skills/harness-audit
- Do not write to
/Users/kevin/.claudeunless Kevin explicitly asks. If the Claude skill includes memory/session-writing behavior, reinterpret it into Hermes-native memory or Hermes-owned files unless Kevin explicitly asks to update Claude state.
Source of truth
- Claude skill directory:
/Users/kevin/.hermes/external-repos/claude-code-skills/harness-audit - This wrapper is generated by:
/Users/kevin/.hermes/scripts/generate_claude_skill_wrappers.py
Verification
Before acting on the task, confirm the source skill was read successfully. If it cannot be read, stop and report the exact missing path.