| name | skill-auditor |
| description | Audits new or modified skills for Optimus compliance before deployment. Use this skill whenever reviewing a skill that has been created or changed — check file paths, MCP tool references, structural quality, and error handling. Also use when doing a quality sweep of all skills in the project. |
Skill Auditor
Validates that skills meet Optimus standards before they ship. Think of this as a
linter for SKILL.md files — it catches problems that would cause runtime failures
or confuse agents.
When to audit
- After creating a new skill (via
skill-creator or manually)
- After modifying an existing skill
- Before a release to verify all skills are consistent
- When an agent reports a skill-related failure
What to check
1. File structure
The skill must live at .optimus/skills/<skill-name>/SKILL.md. The directory name
must be lowercase, hyphenated, and match the name field in frontmatter.
✅ .optimus/skills/data-migration/SKILL.md
❌ .optimus/skills/DataMigration/skill.md
❌ .optimus/skills/data-migration.md (no subdirectory)
2. YAML frontmatter
Required fields:
name: Must match the directory name
description: Should be specific about WHEN to trigger, not just WHAT it does.
A good description is slightly "pushy" — it tells the model to use this skill
even in borderline cases.
description: Helps with database tasks.
description: Manages database migration workflows including schema changes,
seed data, and rollbacks. Use whenever the user mentions migrations, schema
updates, database versioning, or Flyway/Liquibase.
3. MCP tool accuracy
Every tool name referenced in the skill must actually exist in the MCP server.
Current valid tools:
| Tool | Purpose |
|---|
vcs_create_work_item | Create Issue / Work Item |
vcs_create_pr | Create Pull Request |
vcs_merge_pr | Merge Pull Request |
vcs_add_comment | Comment on Issue or PR |
vcs_update_work_item | Update Issue state |
vcs_list_work_items | List work items with filters |
roster_check | List available agents/roles/skills |
delegate_task / delegate_task_async | Dispatch work |
dispatch_council / dispatch_council_async | Multi-expert review |
check_task_status | Poll async task |
append_memory | Save learnings |
Red flags:
github_create_issue → replaced by vcs_create_work_item
github_create_pr → replaced by vcs_create_pr
- Any tool name not in the table above → likely hallucinated
4. Error handling
Skills that reference MCP tools should include recovery guidance. At minimum:
- What to do on auth/credential errors (401/403)
- What to do on invalid parameters
- When to halt vs. retry
5. Writing quality (per skill-creator guidelines)
- Explain the WHY, not just impose rules. Heavy-handed MUSTs and NEVERs are a
yellow flag — reframe as reasoning when possible.
- Include at least one concrete example showing correct usage.
- Keep the skill under 500 lines. If approaching the limit, split into
reference files in the skill's subdirectory.
- Anti-patterns section should describe real mistakes, not hypothetical ones.
Audit output
After reviewing, report:
- Pass/Fail for each check
- Specific fixes needed (with line references if possible)
- Severity: blocking (must fix before use) vs. advisory (improve later)
Syncing audited skills
If working on the optimus-code repository, remember to sync changes to both:
.optimus/skills/<name>/SKILL.md (host project)
optimus-plugin/skills/<name>/SKILL.md (ships to users)