grill
Interview the user about a plan, a design or a decision until every branch is settled, with a recommended answer for each question. Use when the user asks to be grilled, wants to stress-test an idea, or a request has several readings that change what gets built.
- License 许可
- ISC
- 最近更新
- 2026-11-15
- Stars
- ★ 3
- Forks
- 0
01 前言 Front Matter
- name
- grill
- description
- Interview the user about a plan, a design or a decision until every branch is settled, with a recommended answer for each question. Use when the user asks to be grilled, wants to stress-test an idea, or a request has several readings that change what gets built.
02 资源文件 1
- 参考文档reference/questionnaire.md2.0 KB
03 SKILL.md 正文
grill — settle every decision before the work starts
In a repository with .scratchpad/dossier/, use /dossier:plan: it runs this interview and writes the answers into the plan.
Map the request as a design tree: each decision branches into the decisions that depend on it. The frontier is every decision whose prerequisites are settled — the questions you can ask now without a guess at an answer you have not heard.
Steps
- Find the facts yourself. A question the repository, the tools or the web can answer is a fact, not a decision. Look it up, or send a subagent for a long search, and ask the other questions while it runs. Only the questions that depend on that fact wait. Done when each open node is a fact with its source or a decision for the user.
- Ask the frontier in one round. Ask it as one decision block, in the format below, then wait for the answers. Number the decisions across groups and rounds, so a reply reads "2b, 3c". Put the recommended option first, labelled
Recommended:. It names what backs it: thepath:line, the command or the URL you probed. A taste call says so. Every option states what it changes or costs. - Recompute the frontier. Each answer settles nodes and unblocks others. A decision that depends on an open one rides in the block with a "Skip if" note. When its answer adds or removes whole groups, it goes to a later round. When an answer changes a premise, re-derive each earlier recommendation that rested on it and ask again the ones that change. "Discuss more" on a decision splits it into the decisions it hides; ask them as new numbered decisions.
- Audit the set. Before the last round closes, read the decisions together against the constraints the user stated. Two decisions that conflict, or a decision that breaks a constraint, go back to the user as a new question.
- Stop on an empty frontier and a confirmation. Every branch is visited and nothing is silently assumed. Summarise the decisions, one line each, and act only after the user confirms the summary.
<the fact that is dangerous now, with its source; omit when none>
Below are the decisions, in plain English. Each has my recommendation first.
A. <group topic>
<facts the group's decisions share, one bullet each, with source; omit when none>
1. <question, one sentence>
- (a) Recommended: <option>. <what it changes or costs>, backed by <source>
- (b) <option>. <what it changes or costs>
2. <question> Skip if 1(b).
- ...
Answer decisions 1–2. You can reply "all recommended" and list only the ones you want differently.
- Write each fact once, above the group that uses it.
- End the block on the answer line.
A decision the user cannot answer
When a decision needs someone outside the conversation, write a questionnaire for that person with questionnaire.md. Mark the decision as waiting on it, and go on with the branches that do not depend on it.
Honesty labels
| claim | enforced by |
|---|---|
| a fact is looked up, not asked | model-judgment — no script tells a fact from a decision |
| a recommendation names what backs it | model-judgment — ds grill-add exits 64 on a missing probe=; this skill has no such check |
| the frontier is empty | model-judgment — no fixed tree exists to check against |
| the user confirmed | model-judgment — the confirmation is the user's own text |
04 同仓库更多 Skills
ship
Review the dossier branch once, fold the fixup commits into the planned commits, hand the push and PR to the operator, then watch CI. Invoke when the user says "ds:ship", "ship the dossier", or when ds next exits 4.
plan
Open a dossier and write its plan with the operator, then get it checked by a fresh agent before any build. Invoke when the user says "ds:plan", "start a dossier", "plan this", "grill me" in a repo with `.scratchpad/dossier/`, or when work needs more than one commit.
suite-review
Run the monthly evidence review of this repo's plugin skills. Count each skill's opportunities and uses with the census, read what happened where a skill was not used, put each skill in one bucket, act, and write the actions as predictions with a check date. Use when the operator says "suite review" or "monthly review", or when the suite-review line on LATER.md is due.
research-report
Research a technical question with probes on the local system, a web fan-out through deep-research, a grounding pass on every load-bearing claim, and a report rendered as one HTML page. Use when the user asks for a research report, a design study or an evidence-backed recommendation about their own system.
html-report
Render a markdown report as one self-contained HTML page for a human reader, in plain English, with the answer first, a table of contents and Mermaid diagrams drawn as images. Use when the user asks for an HTML report, or for a readable page made from a markdown document.
domain-modeling
Build and sharpen a project's domain model. Use when discussing codebase terminology, writing or editing a GLOSSARY.md, or recording or editing an ADR.