grilling
Use when the user wants a plan, decision, design or idea stress-tested before committing to it, or says "grill me", "grill this", "stress-test this", "poke holes in this", "what am I missing". Do not fire on your own.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-14
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- grilling
- description
- Use when the user wants a plan, decision, design or idea stress-tested before committing to it, or says "grill me", "grill this", "stress-test this", "poke holes in this", "what am I missing". Do not fire on your own.
02 资源文件 2
- 其他.skill-check260 B
- 其他evals.toml165 B
03 SKILL.md 正文
Grilling
Run only when the user asks to be grilled or to stress-test something; never start a grilling on your own. Interview the user until you both understand the decision the same way. Map it as a design tree: every decision branches into the decisions that hang off it.
Work the tree in rounds. The frontier is every decision whose prerequisites are already settled, which are the questions you can ask now without guessing answers you have not heard. Ask the frontier at most three questions per round. Number each question, give the options, and give your recommendation. Then wait for the answers before the next round.
Nothing stays in the user's head between turns, so every round carries its own state.
Round format
**Settled (N):** <one short clause per settled decision, most recent last>
**Round K. Frontier: <count>.**
**Q1 - <title>**: <question>
a) ... b) ... c) ...
Recommendation: <your answer, what you would reject and why, and context they may have forgotten>
**Deferred**: <titles of questions blocked on this round, or "none">
Never skip the settled line or the deferred line, even in a one-question round. A question whose answer depends on another open question belongs to a later round. A frontier wider than three spills into the next round, and its titles go in the deferred list.
Skip
The user may answer skip. Take your recommendation, mark it (assumed), and keep going. At the end, list every assumed answer once for a yes or no confirmation.
Facts are yours, decisions are theirs
When a question needs a fact from the environment (files, tools, docs, the web), find it yourself, with an explorer subagent when subagents are available. A web lookup from a sandbox without network needs an approved escalation; if it is refused, say the fact is unverified and ask the rest of the frontier. Never ask the user for something you can look up. Only questions that depend on a pending lookup wait for it; ask the rest of the frontier now. State the fact when you use it.
Tree file
Write grill-<slug>.md in the current workspace (never outside it, which the sandbox may refuse) and update it at the end of every round. In a read-only sandbox, keep the same structure in your reply instead:
# Grill: <topic>
## Settled
- Q1 <title>: <answer> (over <rejected option>) - <one-line why>
- Q2 <title>: <answer> (assumed)
## Open
- Frontier: <titles>
- Deferred: <titles>
Ending
The session ends when the frontier is empty: every branch visited, nothing silently assumed. Confirm the assumed answers, then summarise the settled tree and ask whether to write it up as a spec or plan. Do not act on the outcome until the user confirms you agree.
These thoughts mean you are quitting early. Recompute the frontier and ask the next round instead: "the rest is implementation detail", "we can figure that out later", "the remaining questions are obvious", "good enough to start".
04 同仓库更多 Skills
web-consensus
Use when a question has to be settled against the live web rather than from memory - researching a topic, proving or disproving a claim, checking whether something is still true, a sourced second opinion on a factual question - or when the user says "research", "prove it", "with sources", "check if this is true", "still accurate?".
skill-validate
Use when a SKILL.md is about to ship or was just edited, when a skill is reviewed or audited, when a skill seems ignored, skipped or misread by agents, or when the user says "validate this skill", "lint the skill", "check my skill", "is this skill any good".
regression-tests
Use when fixing a bug, changing behaviour or adding a feature in code that has tests, or when the user asks whether a change is covered by tests, wants regression tests, or wants coverage of the changed lines.
redacao
Use when writing, rewriting or editing a technical article or blog post in Portuguese, or when the user says "escreve um artigo", "reescreve o post", "redação", "pauta", "revisa o texto", "tira a cara de IA".
improve
Use when the user asks to improve the harness, its rules, Guardian policy, skills or knowledge from experience on this machine, or says "remember this for next time", "make this a rule", "teach the harness".
hld-trace
Use when the user asks whether the code follows the high-level design (HLD), which requirements have tests, which are untested, or wants tests linked to requirement or work item IDs.