agentleFS
BagelHoleBagelHole/BagelSkills/skills/grill-me-until-its-clear/SKILL.md

grill-me-until-its-clear

Skill2 starsChanged 4 months agoMITView on GitHub

Use this skill when a user has a vague product, feature, architecture, or project idea that must be clarified before planning or implementation. Interrogate assumptions, expose contradictions, resolve consequential unknowns, and convert the answers into a decision-complete brief with testable acceptance criteria.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 7 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If a required input is retrievable from available tools or artifacts, retrieve it rather than asking the user.

    line 43
  2. Distinctive:

    Ask at most five high-leverage questions at a time.

    line 61
  3. Distinctive:

    Complete when: no high-impact item remains silently ambiguous

    line 75
  4. Very specific:

    Write the outcome, users, scope, non-goals, flows, rules, data, constraints, risks, decisions, open items, and Given/When/Then acceptance …

    line 79
  5. Distinctive:

    Ask one question when its answer determines all subsequent questions.

    line 85
  6. Distinctive:

    Do not ask questions recoverable from available files or tools.

    line 86
  7. Distinctive:

    Treat “I do not know” as useful data: offer a default and state its consequence.

    line 88
  8. Distinctive:

    Do not start implementation merely because the conversation is long.

    line 89

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Grill Me Until It’s Clear

Outcome

Turn an exciting but underspecified idea into a brief that another person or agent can execute without inventing product decisions.

This skill changes the agent's process: it requires evidence, explicit claim strength, and a checkable completion boundary. Load references/playbook.md only when deeper prompts or rubrics are needed. Use assets/output-template.md when producing the final artifact.

When to use

  • The request contains fuzzy words such as “simple,” “fast,” “better,” or “AI-powered.”
  • Different reasonable interpretations would produce materially different builds.
  • The user asks to be grilled, challenged, interviewed, or helped refining an idea.
  • Implementation would force the agent to guess users, constraints, data, failure behavior, or success criteria.

Do not use when

  • The change is narrow, reversible, and already has testable acceptance criteria.
  • The user explicitly wants a disposable exploration and accepts unresolved assumptions.

Required inputs

  • The raw idea and intended outcome
  • Known users, constraints, deadlines, and non-negotiables
  • Existing product or repository context when available

If a required input is retrievable from available tools or artifacts, retrieve it rather than asking the user. Ask only when a missing answer represents authority, preference, or risk that tools cannot resolve.

Workflow

1. Frame the decision

Restate the idea in one sentence, identify the decision owner, and separate the desired outcome from the proposed solution. Do not begin the interview until the user can correct this frame.

Complete when: the frame is explicit and corrected or accepted

2. Build the uncertainty map

List only unknowns that could change scope, architecture, UX, risk, or acceptance. Classify them as user, value, behavior, data, integration, operations, or constraint unknowns.

Complete when: consequential unknowns are named and ranked

3. Grill in ranked rounds

Ask at most five high-leverage questions at a time. Prefer forced tradeoffs and concrete scenarios over “tell me more.” Follow an answer when it reveals a contradiction; do not mechanically advance a questionnaire.

Complete when: the current question resolves the highest-impact fork

4. Run contradiction and edge passes

Compare new answers with earlier claims. Test unhappy paths, permissions, empty states, failure ownership, migration, and what the product must never do.

Complete when: contradictions and failure behavior have explicit dispositions

5. Close or label every unknown

Resolve each consequential unknown, assign a named owner/date, or record an explicit assumption plus the cost of being wrong. “TBD” without ownership is not closure.

Complete when: no high-impact item remains silently ambiguous

6. Produce the decision-complete brief

Write the outcome, users, scope, non-goals, flows, rules, data, constraints, risks, decisions, open items, and Given/When/Then acceptance criteria. End with a readiness verdict and its evidence.

Complete when: a new executor can act without inventing a consequential decision

Operating rules

  • Ask one question when its answer determines all subsequent questions.
  • Do not ask questions recoverable from available files or tools.
  • Challenge the proposed solution without dismissing the underlying goal.
  • Treat “I do not know” as useful data: offer a default and state its consequence.
  • Do not start implementation merely because the conversation is long.

Output contract

The final result must contain:

  • One-sentence outcome and readiness verdict
  • Decision ledger: decision, rationale, owner, confidence
  • Scope and explicit non-goals
  • Primary and failure flows
  • Assumptions and unresolved items with owners
  • Given/When/Then acceptance criteria

Distinguish observed, inferred, assumed, and unverified claims. Link to files, commits, screenshots, commands, logs, or supplied sources when those artifacts support a consequential conclusion. Keep the executive result concise; move detail behind appendices when necessary.

Common pitfalls

  • Questionnaire theater: asking many low-impact questions instead of resolving major forks.
  • Premature convergence: turning the first plausible answer into a requirement.
  • Solution capture: protecting the proposed implementation from scrutiny.
  • Infinite interview: continuing after remaining unknowns are cheap and reversible.

Verification checklist

  • The requested outcome—not merely the procedure—has been satisfied.
  • Every consequential claim is supported, qualified, or explicitly unverified.
  • The workflow's completion criteria are met in order.
  • Safety boundaries and forbidden side effects were respected.
  • The output follows assets/output-template.md or contains equivalent information.
  • Remaining unknowns have an owner, next action, or stated consequence.

Resources

  • references/playbook.md — deeper domain prompts and evaluation guidance.
  • assets/output-template.md — reusable final-output structure.
  • evals/evals.json — output-quality cases.
  • evals/trigger-evals.json — positive and negative description-trigger cases.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.