grilling
Grill the user relentlessly about a plan, decision, or idea, one question at a time, keeping a running notepad of every decision. A reusable primitive; end with /make-it-so to promote or /checkpoint to pause. Use when the user wants to stress-test thinking or uses any 'grill' trigger phrase.
- License 许可
- MIT
- 最近更新
- 2026-09-05
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- grilling
- description
- Grill the user relentlessly about a plan, decision, or idea, one question at a time, keeping a running notepad of every decision. A reusable primitive; end with /make-it-so to promote or /checkpoint to pause. Use when the user wants to stress-test thinking or uses any 'grill' trigger phrase.
03 SKILL.md 正文
This is the core interview primitive. It runs the questioning and keeps a running
notepad of the result. It never writes canonical files itself — the /make-it-so
skill promotes the notepad, so grilling can also run on its own.
Grill the user relentlessly about every aspect of this until we reach a shared understanding. Walk down each branch of the decision tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.
Ask the questions one at a time, waiting for feedback on each question before continuing. Asking multiple questions at once is bewildering.
If a fact can be found by exploring the environment (filesystem, tools, etc.), look it up rather than asking me. The decisions, though, are mine — put each one to me and wait for my answer.
The notepad (start here, before asking anything)
Keep a running notepad of the whole session so nothing is lost to a context compaction or a hand-off to another assistant. The notepad is the working record; it is not canonical until it is promoted.
- Explore enough to build the real decision tree.
- Create the notepad at
.agents/grilling diary/<topic-slug>-<YYYY-MM-DD>.md(use the current date; create the folder if missing). Header it exactly:STATUS: PROVISIONAL — not canonical until /make-it-so - Write the FULL planned question list up front, so I see the terrain before answering. It is a best-effort map; it will flex.
Format — stable IDs that are never renumbered, one status tag per line:
# Grilling — <topic>
STATUS: PROVISIONAL — not canonical until /make-it-so | Started <date>
## Planned questions
1. [open] <question>
2. [open] <question>
2A. [open] <sub-question / dependency of 2>
3. [open] <question>
## Spawned branches
- A sub-question or dependency nests as
2A,2Bunder its parent. - A NEW line of questioning I open appends to the bottom under
## Spawned branchesas the next top-level number; never renumber the items above it. - Tags:
[open](unasked/undecided),[tentative](leaning, revisit),[locked](decided).
During the interview
- Ask the next
[open]question with your recommendation. - After I confirm an answer, immediately flip that line to
[locked](or[tentative]) and append the one-line result in the notepad, before asking the next question. This tiny per-answer write is the entire cost. - Do not act on it until I confirm we have reached a shared understanding, and do not touch canonical files during the interview.
Resuming after a compaction or model switch
The notepad is the durable record, not the chat. If the conversation is
compressed, truncated, or handed to another assistant, re-read the notepad and
continue at the first [open] line. Never re-ask a [locked] question.
Exits
Keep grilling until I explicitly run one of these skills — do not stop on a lookalike phrase said in passing:
/make-it-so— I am done; confirm the approvals, promote the notepad's locked decisions to canon, implement them, and push the results to the remote./checkpoint— save and stop for now; commit the notepad to resume later.
Only those invoked skills end the interview. Continue only within the standing project authority and safety boundaries.
04 同仓库更多 Skills
writing-great-skills
Reference for writing and editing skills well — the vocabulary and principles that make a skill predictable.
update-harness
Update a project to the current canonical LLM Workbench harness without losing project-specific controls, work state, or proof.
tracer-bullet
Cut a capability into vertical tracer-bullet slices that each pierce every layer of a project's stack, then order, scope, and assign them.
to-tickets
Break an assigned Workbench spec into dependency-aware tracer-bullet tickets.
to-spec
Turn an already-settled conversation into one stable Workbench capability spec without restarting the interview.
to-docs
Route settled conversation truth into existing Workbench documentation owners without restarting discovery or creating another store.