| name | yogieat-harness-governor |
| description | Use when repeated Yogieat prompts, review comments, or misunderstandings should be translated into harness candidates or stable rules. Classifies new observations, applies the candidate-to-promotion policy, and keeps harness updates intentional rather than noisy. |
Yogieat Harness Governor
Use this skill when the conversation reveals a repeated instruction, a recurring review finding, or a stable preference that should live in the harness.
Workflow
- Read
./AGENTS.md, .codex/rules/governance.md, and .codex/governance/rule-candidates.md.
- Classify the new observation as:
- already covered by a stable rule
- new candidate rule
- candidate ready for promotion
- task-local preference that should stay out of the harness
- Prefer candidate registration before promotion unless the miss is high severity.
- During unrelated feature work, propose harness updates instead of silently editing them.
- During dedicated harness work, update the relevant rule, skill, agent, or governance file directly.
Good Candidates
- Repeated service or facade boundary clarifications
- Repeated validation scope corrections
- Repeated transaction or test harness pitfalls
- Repeated long prompt boilerplate that a skill or rule can replace
Bad Candidates
- One-off naming tastes
- Isolated wording preferences
- Non-recurring refactor choices
Notes
- Promotion threshold: two repeats, or one high-risk miss.
- Stable rules should stay short and operationally meaningful.
