External Skill Fixture Builder
Use this skill when a third-party ecosystem pattern should be tested by skill-harness but should not become a live dependency.
Process
- Choose the fixture lane by install posture, not by brand:
- pure
SKILL.mdpack - plugin manifest
- Cursor
.mdcrules - Codex/Claude/Cursor subagents
- MCP/tool config
- generated record/replay skill
- Beads or task-memory repo
- multi-agent workspace layout
- pure
- Use minimal synthetic fixtures when possible.
- Use pinned external snapshots only when real structure matters and license/provenance allows it.
- Keep fixtures out of global user directories and live install caches.
- Make the fixture assert one behavior: parse, classify, trust recommendation, drift detection, render target, or installer opt-in.
- Add a command or check that proves the fixture is exercised.
Output
Fixture Lane
Source Or Synthetic Basis
Files Included
Expected Classification
Risk Flags Expected
Check Command
Follow-Up Adoption Question
Avoid
- making fixtures large mirrors of public repos
- using fixtures as a back door for third-party adoption
- mixing fixture coverage with default installer behavior
- storing credentials, real traces, browser recordings, or private task state in fixtures