grill-me
Skill11 starsChanged 11 days agoMITView on GitHub
Relentless Socratic interview that turns a loose idea into decisions you can defend — any plan, design, program, strategy, or decision, software or not. Follows each thread until it holds or breaks. Use when you want to be grilled on anything before committing. Consult @ml-ai-engineer where eval design, judging rubrics, or benchmark interpretation are under review.
193 other projects use a version of this.
Adopted by
Most recently taken up by
- bamlab/flashlighttheir own version, first seen Nov 16, 2026
- philoserf/claude-code-configtheir own version, first seen Nov 15, 2026
- niki914/zafirotheir own version, first seen Nov 15, 2026
- quickchatai/quickreviewtheir own version, first seen Nov 15, 2026
- HiteSit/real-scientist-skillstheir own version, first seen Nov 15, 2026
- alexj11324/Mikkatheir own version, first seen Nov 15, 2026
Used with
Projects with it also use brief-to-tasks, writing-beats and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
julianoczkowskiSkillbrief-to-tasks
Break a design brief into an ordered checklist of independently buildable tasks using vertical slices. Saves as a markdown checklist. Use when user wants to break down work, create tasks from a brief, plan implementation order, or mentions "tasks" or "breakdown".
Also in 7 projects that use this
LesaniSkillwriting-beats
Shape an article as a journey of beats, choose-your-own-adventure style. The user picks a starting beat from the raw material, you write only that beat, then offer options for where to pivot next, beat by beat, until the article reaches a natural end. Use when the user has raw material and wants to assemble it as a narrative rather than an argument.
Also in 74 projects that use this
simo97Skillmigrate-to-shoehorn
Migrate test files from `as` type assertions to @total-typescript/shoehorn. Use when user mentions shoehorn, wants to replace `as` in tests, or needs partial test data.
Also in 78 projects that use this
simo97Skillrequest-refactor-plan
Use this skill when user wants to create a refactor plan.
Also in 45 projects that use this
LesaniSkillwriting-shape
Take a markdown file of raw material and shape it into an article through a conversational session — drafting candidate openings, growing the piece paragraph by paragraph, arguing about format (lists, tables, callouts, quotes) at each step. Use when the user has a pile of notes, fragments, or a rough draft and wants help turning it into something publishable.
Also in 73 projects that use this
RoentekSkillloop-me
Grill me about specs for the workflows I want to build, within this workspace.
Also in 64 projects that use this
- +1 more from simo97
- +1 more from Lesani
Other versions193 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 193
- mattpocock/skills★ 325ksame name
- The-Vibe-Company/quivr★ 40ksame name
- QuivrHQ/quivr★ 40ksame name
- langfuse/langfuse★ 37ksame name
- nrwl/nx★ 29ksame name
- windmill-labs/windmill★ 19ksame name
- GreyDGL/PentestGPT★ 16ksame name
- humanlayer/humanlayer★ 13ksame name
- jaredpalmer/kev★ 9.8ksame name
- openstatusHQ/openstatus★ 9.3ksame name
183 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository44 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- analyze-dataSkill
- brainstormingSkill
- craft-lessonSkill
- create-agentSkill
- create-skillSkill
- customize-agentSkill
- customize-skillSkill
- designSkill
- deslopSkill
- developSkill
- discovery-reportSkill
- elicitationSkill
- empathy-mapSkill
- explore-ideaSkill
- help-meSkill
- humanizeSkill
- incidentSkill
- iterateSkill
- journey-mapSkill
- jtbdSkill
- kanbanSkill
- launchSkill
- legal-safety-auditSkill
- maintenance-executeSkill
- maintenance-planSkill
- memorySkill
- motionSkill
- phaseSkill
- planSkill
- research-planSkill
- retroSkill
- reviewSkill
- root-causeSkill
- round-tableSkill
- shape-upSkill
- shut-upSkill
- statusSkill
- teach-meSkill
- testSkill
- unpack-problemSkill
- validate-ideaSkill
- validation-patternsSkill
The file
We split this skill into the rules it gives an agent
10 rules. 0 are common, 7 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 178 edited versions we read keep these.
7
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 23
Before /validate-idea exits the loop phase
- Distinctive: line 24
Before any ADR is written (forces decisions to be explicit)
- Very specific: line 25
After /design produces a PRD (sanity check before /develop)
- Distinctive: line 26
Before committing budget, headcount, or calendar to any non-software program
- Distinctive: line 33
When you can already specify the thing precisely — you don't need to grill it
- Very specific: line 34
When the ask is a critique of *text* rather than a judgement on an *idea* — use /elicitation
- Distinctive: line 54
Follow each thread until it either holds or breaks:
- Distinctive: line 84
This list is not complete and is not meant to be: If your subject has a risk that fits none of these — legal, ethical, a person who will be …
- Distinctive: line 132
Grilling without ground truth: No material, no interview — go collect it first.
- Distinctive: line 133
Grilling an ungrillable question: If it needs something to react to, stop and prototype.
“Common” is measured: kept by at least 6 in 10 of the 178 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Grill-Me — Socratic Stress-Test
What this skill does
Interviews you until you can commit to an idea. It's a conversation, not an interview — you own the scope, the skill asks the questions, and the order follows your answers rather than a script.
It is stateless. It writes no files and leaves no workspace behind; the only result is a clearer version of the idea, in your head. It needs no repo and does not assume the subject is software — people use it for business calls, for writing, and for what to do next.
When to use
- "Grill me on this …" — a plan, architecture, spec, training program, business idea, hiring decision, migration, launch, policy
- Before
/validate-ideaexits the loop phase - Before any ADR is written (forces decisions to be explicit)
- After
/designproduces a PRD (sanity check before/develop) - Before committing budget, headcount, or calendar to any non-software program
When NOT to use
- For open-ended brainstorming — use
/validate-ideafirst - For technical debugging — use
/incident - For retrospective analysis — use
/retro - When you can already specify the thing precisely — you don't need to grill it
- When the ask is a critique of text rather than a judgement on an idea — use
/elicitation
Prerequisites
- A concrete artifact to interrogate (plan, spec, architecture, hypothesis, design, curriculum, proposal)
- Its ground-truth material — for software the actual codebase; otherwise existing documents, data, metrics, prior attempts, audience information. If none exists, ask for the closest thing first. A grill without ground truth is opinions trading punches.
When invoked indirectly
Most of the triggers above are other skills. When one of those calls this skill, do not ask "What worries you most about this?" — the caller has a specific question, so answer it. Do not open a chooser. Hand the caller your findings and stop; the caller owns its own artifact.
The stance
Do not accept the first answer. That is the whole mechanism. A perfect question list in the hands of an agreeable interviewer produces shallow answers and a nod.
Follow each thread until it either holds or breaks:
- "You said X — why X and not Y?" then "what would have to be true for Y to be better?"
- "What's the strongest argument against that?" — and don't accept a strawman.
- "You're confident. What would change your mind?" — if nothing would, that is the finding.
Give your own recommended answer with reasoning each time, so there is something to push against. Depth is set by blast radius, not by remaining curiosity: a low-stakes question closes in one honest exchange, a load-bearing one gets the full drill.
Before you close, read it back. Scan what was said for contradictions — a later answer that undercuts an earlier one, a cost that rules out something the design assumed, a question you skipped that turned out to be load-bearing. Surface them and make them resolve. An interview that ends with two answers that cannot both be true has not finished; it has stopped.
The questions you will forget
An interviewer defaults to the happy path. These are the questions that get skipped — check them before you close, and instantiate each in the subject's own terms rather than asking it verbatim.
| Question | Ask it as |
| What must stay true? | "What invariant must hold after every operation — and what restores it if it doesn't?" |
| What else moves? | "What breaks elsewhere when this changes — has anyone checked the dependents?" |
| Who attacks it? | "Where does untrusted input enter, what could leak, and what's the check on the 'internal-only' path?" |
| How does it fail? | "Crash mid-operation: clean resume, manual repair, or corruption? Can a retry duplicate the effect?" |
| What does it cost at 100x? | "Marginal cost per unit at 100x volume — and what maintenance burden does this create?" |
| What can we cut? | "Smallest increment that still delivers value — and who actually asked for the rest?" |
| What does done mean? | "What metric moves, who declares it done, and what are we willing to sacrifice for it?" |
This list is not complete and is not meant to be. If your subject has a risk that fits none of these — legal, ethical, a person who will be upset — name it and drill it with the same rigor. The list exists to catch what an agreeable interviewer skips, not to bound what may be asked.
Worked instantiation — a training program (not software, to show the questions travel). Each label below stays on one line so the wrapping cannot hide one: What must stay true? Prerequisite chains — a learner who misses week 2 can still pass week 4, or the sequence is decorative. What else moves? Learners' teams absorb the workload; the instructor's other cohorts lose prep time. Who attacks it? Assessment gaming and content leakage — and what detects either. How does it fail? The instructor drops out mid-cohort: who finishes it, and does the cohort repeat or skip? What does it cost at 100x? Per-learner marginal cost at ten cohorts, plus the instructor time each cohort permanently consumes. What can we cut? A one-day workshop plus a reference doc, delivering 80% of the outcome. What does done mean? Retention at 90 days, not attendance — and who signs off that it worked.
Three boundaries
Some questions cannot be answered by talking
"One long form or three pages?" and "how should this interaction feel?" are ungrillable — you need something to react to. When you hit one, stop grilling: build the throwaway version, look at it, come back and answer in one line. Sessions grow too long when you try to talk through an ungrillable question, and the scope grows with the uncertainty.
Count rounds, not questions
"Forty-six questions across four rounds is an ordinary session." Two hundred questions usually means the scope was too large — break the work into smaller pieces and grill each. Very long sessions also drift into the dumb zone, where the context window fills and the questions get worse. Length is not thoroughness.
Passivity — you can be the sycophant
"Answering 'agreed, agreed, agreed' for forty questions and coming out with a plan the agent wrote and you nodded at. It feels productive because it was long. But you decided nothing."
There is no coverage count that proves you thought. So: push back on a question less detailed than you need, say when the scope is drifting, and answer "I don't know" and mean it. An unanswerable question is usually a signal to prototype, not to guess.
It's working if you disagree with something, later rounds build on earlier answers, you end somewhere you didn't expect, and you could defend each choice to someone who wasn't there.
Anti-patterns
- Reciting the questions in fixed order. They are a coverage check, not a script.
- Accepting the first answer. "That's what you'd expect — what if the opposite is true?" beats "yes, that works."
- Asking multiple questions at once. The interview loses depth if you bundle.
- Softening a probe to keep the mood warm. The DNA applies double here: the interviewer's job is the uncomfortable question.
- Grilling without ground truth. No material, no interview — go collect it first.
- Grilling an ungrillable question. If it needs something to react to, stop and prototype.
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
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.

