Grill requirements
Skill K95M65/AI_ONBOARD/skills/workflow-foundations/grill-requirements
Runs an explicit, repository-informed interview that pressure-tests a request, plan, or design one consequential decision at a time before implementation. Use only when the user asks to be grilled, interviewed, challenged, or wants assumptions and requirements stress-tested; do not invoke implicitly for routine ambiguity or ordinary clarifying questions.From its SKILL.md
npx -y skills add K95M65/AI_ONBOARD --skill grill-requirementsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
3.9 KB, 705 tokens by cl100k_base, as published. Nobody here has run it
Grill requirements
Resolve consequential ambiguity before planning or implementation. Keep the interview demanding but constructive, and end with a shared decision brief rather than code.
Hold the boundary
- Do not implement, edit files, or enact the plan during the grill.
- Do not silently turn every unclear task into an interview. The user must explicitly request this mode.
- Use the harness's native structured-question interface when available; otherwise ask in normal conversation.
- Write a glossary, ADR, specification, or other durable artifact only when the user explicitly requests documentation.
- Let the user pause or end the interview at any time.
Prepare from evidence
- Read the supplied request, plans, documents, and relevant repository context.
- Resolve questions from code, configuration, existing decisions, or documentation before asking the user.
- Identify the unresolved decision branches and their dependencies.
- Start with the highest-leverage decision that constrains the most downstream choices.
Do not ask the user to recite discoverable repository facts. Surface conflicts between the request and existing evidence directly.
Run the interview loop
For each unresolved branch:
- State the decision and why it matters in one concise sentence.
- Ask one primary question at a time. A native question UI may present two or three mutually exclusive options for that single decision.
- Put the recommended answer first and explain its main consequence or tradeoff.
- Avoid neutral option dumps, false dichotomies, leading questions, and questions whose answers do not change the plan.
- Wait for the answer before moving to a dependent branch.
- Record the accepted decision, rejected alternatives that matter, and new downstream implications.
- Challenge contradictions, unsafe assumptions, hidden actors, missing states, and untestable success claims instead of merely collecting preferences.
When an answer opens a new material branch, add it to the decision map and resolve it in dependency order. Keep the map concise; do not overwhelm the user with the entire tree.
Cover the material surfaces
Adapt the interview to the work. Resolve only surfaces that could materially change the result:
- intended outcome, audience, and primary job;
- scope, non-goals, ownership, and authority;
- terminology and domain invariants;
- workflows, actors, states, and permissions;
- inputs, outputs, data boundaries, integrations, and migrations;
- failure, recovery, edge cases, compatibility, and rollout;
- quality thresholds, acceptance evidence, and operational constraints.
Depth should follow risk and reversibility. A small reversible choice needs less grilling than an architecture, security boundary, data migration, or costly product direction.
Know when to stop
End the grill when:
- no unresolved decision could materially change the objective, architecture, behavior, safety, or acceptance evidence;
- remaining unknowns are explicitly deferred with an owner or validation path; or
- the user asks to stop.
Do not continue asking questions to perform thoroughness.
Return the decision brief
Conclude with:
Shared objective:
Accepted decisions:
Rejected alternatives that matter:
Assumptions still in force:
Open questions or deferred validation:
Non-goals:
Completion evidence:
Recommended next mechanism:
Recommend goal-contract for substantial, resumable, or goal-backed work. Recommend direct planning for a
bounded task, or a one-off prompt when no persistent mechanism is warranted.
What ships with it: 1 file
274 B alongside SKILL.md
agents/
- openai.yaml274 B
Gives 1 of the 12 instructions most review quality skills give in 705 tokens
Counted across 1,048 of the 1,783 authors here whose files we hold, read 2026-08-07
- Ask questions one at a timehere, and in 81 of 1048, across 64 files
- Provide a recommended answer for each questionin 73 of 1048, across 50 files
- Explore the codebase instead of asking answerable questionsin 66 of 1048, across 42 files
- Resolve dependencies between decisions one-by-onein 42 of 1048, across 17 files
- Interview the user relentlessly about the planin 38 of 1048, across 13 files
- Order findings by severityin 31 of 1048
- Resolve each branch of the decision treein 27 of 1048, across 5 files
- Run a grilling sessionin 26 of 1048, across 5 files
- Update CONTEXT.md immediately when a term is resolvedin 26 of 1048, across 11 files
- Propose precise canonical terms for vague languagein 25 of 1048, across 7 files
- Create documentation files lazilyin 24 of 1048, across 5 files
- Assign severity to every findingin 24 of 1048
Said here and by no other author read
- resolve consequential ambiguity before planning
- read supplied request plans and repository context
- wait for the answer before moving to a dependent branch
- record accepted decisions and rejected alternatives
- challenge contradictions and unsafe assumptions
- adapt the interview to the material surfaces
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.