refactor(skills): make grill a strict workflow

This commit is contained in:
2026-09-08 14:41:33 -04:00
parent e5d53700c5
commit 046b5e7dc7

View File

@@ -5,7 +5,7 @@ description: Stress-test a plan, decision, or design through a one-question-at-a
# grill # grill
Interview the user until the session has no open questions. Interview the user until the session has no open or blocked questions.
Open questions define the ending condition for the grill session. Open questions define the ending condition for the grill session.
Blocked questions track known decisions with unsettled prerequisites. Blocked questions track known decisions with unsettled prerequisites.
@@ -14,106 +14,76 @@ A completed grill summary is not an instruction to build.
## Run the interview ## Run the interview
1. Investigate first. 1. Prepare the grill.
Read the conversation, files, and docs that can settle facts or reveal constraints. Read the conversation, files, and docs that can settle facts or reveal constraints.
Resolve factual questions yourself when tools can answer them. Resolve factual questions yourself when tools can answer them.
Do not ask the user for facts you can find. Do not ask the user for facts you can find.
Build the required-question set from the known plan or idea.
Put ready questions under `Open questions:`.
Put known questions with unsettled prerequisites under `Blocked questions:`.
Finish when known facts, constraints, and unresolved decisions are visible. Finish when both sections contain every known required question.
2. Build the open-question list. 2. Start the interview.
Map the decisions needed to finish the grill session. Show `Open questions:` and `Blocked questions:` in short form.
Include known dependent questions as blocked questions.
Finish when the list contains every known required question.
3. Start the session.
Present the Open questions and Blocked questions sections in short form.
Include titles only, not full question bodies. Include titles only, not full question bodies.
Format each item with the markers from [Open-question accounting](#open-question-accounting). Use Markdown list items in both sections.
Then ask the first numbered question in the same turn. Mark each initial item with `- [ ]`.
Ask exactly one numbered question in the same turn.
The question title must match an item under `Open questions:`.
Finish when the interview has started with exactly one numbered question. Finish when the interview has started with exactly one numbered question.
4. Run the question loop. 3. Process the answer.
Show the Open questions and Blocked questions sections on every active grill turn. Account for the user's answer before asking another question.
Ask one numbered question at a time. Resolve, update, remove, unblock, or add questions as the answer requires.
Apply [Question format](#question-format) to each question. Use `- [ ]` for unchanged unresolved questions.
Wait for the user's answer. Use `- [+]` for questions added by the latest answer.
Apply [Open-question accounting](#open-question-accounting) before asking again. Use `- [u]` for questions updated by the latest answer.
Resolve newly answerable facts yourself before asking again. Use `- [x]` for questions resolved by the latest answer.
Use `- [-]` for questions removed by the latest answer.
Include a cause note for each `- [+]`, `- [u]`, and `- [-]` change.
Indent cause notes by two spaces under the changed question.
Show `- [+]`, `- [u]`, `- [x]`, and `- [-]` for one turn only.
After that turn, keep an updated question visible as `- [ ]`.
After that turn, remove resolved and removed questions from the visible list.
Move unblocked questions from `Blocked questions:` to `Open questions:` with `- [u]`.
Do not use `- [+]` for unblocked questions because they are not new.
Add questions only when the latest answer directly causes them.
When the user extends scope instead of answering, account for the new question before pursuing it.
Do not add optional follow-up questions to either section.
Put optional follow-up questions in the final summary instead.
Resolve newly answerable facts yourself.
Finish when the answer is accounted for and the turn ends with a numbered question, summary, or pause statement. Finish when every list change has an explicit marker and required cause note.
5. Close the interview. 4. Ask the next question.
Stop when the open-question list is empty. Show `Open questions:` and `Blocked questions:` on every active grill turn.
Output a concise summary of every decision reached. Keep `- [ ]` items visible every turn.
Include optional follow-up questions that were not added to the open-question list. Preserve agent judgment when the user reframes or invalidates a question.
Do not act on the summary unless the user gives a new implementation instruction. Show the resulting list change instead of following a rigid reframe algorithm.
If the user catches incorrect accounting, briefly explain the error.
Finish when the summary is delivered and grill waits. Then show the corrected question sections.
Ask exactly one numbered question when `Open questions:` contains an unresolved item.
## Question format The question title must match an item currently shown under `Open questions:`.
End each active grill turn with exactly one numbered question.
Give concrete options with trade-offs. Give concrete options with trade-offs.
Label options with `A.`, `B.`, `C.`, and later letters. Label options with `A.`, `B.`, `C.`, and later letters.
Use the same labels in the recommendation. Use the same labels in the recommendation.
Give a recommendation and rationale. Give a recommendation and rationale.
Probe vague words, overloaded terms, and hidden assumptions. Probe vague words, overloaded terms, and hidden assumptions.
Use concrete scenarios to test concept boundaries and edge cases. Use concrete scenarios to test concept boundaries and edge cases.
When code contradicts the user's claim, quote both the claim and the code behavior. When code contradicts the user's claim, quote both the claim and the code behavior.
## Open-question accounting Finish when the turn ends with one numbered question, summary, or pause statement.
Show ready questions under `Open questions:`. 5. Close or pause.
Show questions with unsettled prerequisites under `Blocked questions:`. Close when `Open questions:` and `Blocked questions:` are empty.
Use Markdown list items in both sections. Output a concise summary of every decision reached.
Indent cause notes by two spaces under the changed question. Include optional follow-up questions that were not added to either section.
Do not nest questions under other questions. Do not act on the summary unless the user gives a new implementation instruction.
If the user changes tasks, state that grill is paused.
Use the bracket marker to show item changes on the current turn: Finish when the summary is delivered or grill is explicitly paused.
- `- [ ]` for unchanged unresolved questions
- `- [+]` for questions added by the latest answer
- `- [u]` for questions updated by the latest answer
- `- [x]` for questions resolved by the latest answer
- `- [-]` for questions removed by the latest answer
Keep `- [ ]` items visible every turn.
Show `- [+]`, `- [u]`, `- [x]`, and `- [-]` for one turn only.
After that turn, keep an updated question visible as `- [ ]`.
After that turn, remove resolved and removed questions from the visible list.
When a blocked question becomes ready, move it to `Open questions:` with `- [u]` for one turn.
Do not use `- [+]` for unblocked questions because they are not new.
Add questions only when the latest answer directly causes them.
When the user extends scope instead of answering, account for the new question before pursuing it.
Do not add optional follow-up questions to the open-question list.
Put optional follow-up questions in the final summary instead.
Include a cause note for each `- [+]`, `- [u]`, and `- [-]` change.
Use this format:
```md
Open questions:
- [u] Copyleft preference
Cause: The `src/skills/` exception leaves the rest of the monorepo under the root license.
Blocked questions:
- [+] Patent explicitness
Cause: Your answer made patent terms relevant to the license choice.
```
Preserve agent judgment when the user reframes or invalidates a question.
Show the resulting list change instead of following a rigid reframe algorithm.
When the user catches incorrect open-question handling, briefly explain the error.
Then show the corrected open-question list.
## Turn endings ## Turn endings