Modes
| Trigger | Mode | Result |
|---|---|---|
grill me, grill code, entrevistame |
plain | grilling interview |
grill with docs, grill docs open |
docs | interview plus approved domain notes |
grill me sdd, entrevistame sdd |
plan | one approved neutral plan |
Rules
- Ask each open-ended question directly in normal chat, one at a time, then stop. Add
Recommendation: ...only when useful. - Use the
questiontool only for closed confirmations, modes, ratings, or enumerated choices. - Explore discoverable answers instead of asking. Preserve the user's language; artifacts default to English.
- Plan mode is planning-only: no code edits, builds, installs, tests, commits, or push. The approved plan is the only write.
Flow
- Plain: load
grilling; interview until the decision is clear. - Docs: also load
domain-modeling; write notes only with approval. - Plan: confirm a kebab-case slug and collision handling. Interview Outcome, Scope, Behavior, Approach, Work groups, Dependencies, Verify, Risks, and Execution guidance. Load
execution-planandimplementation-skill-routing; never load implementation skill bodies. Record the result for every work group. Ask once to write one neutral plan. Use.ai/deep-planner/plans/<slug>.mdwhen the active primary owns that path; otherwise return the complete draft and direct the user todeep-planner. On refusal, keep the draft in chat.
Return the current question or conclusion, write status, path, and unresolved decisions. Missing skills are reported, never silently skipped.
Attribution
Inspired by Matt Pocock's grilling skills at https://github.com/mattpocock/skills.