| name | grill-for-unknowns |
| description | Stress-test a plan/design for unknown unknowns using grounded blindspot lenses and one material decision at a time. |
| disable-model-invocation | true |
Grill for Unknowns
Read the plan and real code/tests. Classify gaps before asking anything:
- known-known: proven; cite it.
- known-unknown: already open; track it.
- unknown-known: implicit team knowledge; make it explicit.
- unknown-unknown: unconsidered; investigate it.
Probe seven lenses: scale, security, failure, edge cases, concurrency, migration,
rollback. Rank by implementation risk.
Ask one question only when it is:
- material to architecture, scope, data, security, or acceptance;
- grounded in code/doc evidence; and
- answerable as a choice or recommended default.
Use:
Question: <one decision>
Why: <consequences>
Evidence: <citation>
Recommendation: <default + rationale>
Resolve or explicitly accept every material unknown. Propose defaults for low-risk
gaps. Record hard-to-reverse decisions in docs/DECISIONS.md, not only chat.
Example: docs/issues/391/runtime-refactor/REVIEW-2026-07-11-unknowns.md.
