grill-my-plan
Skill1 starsChanged 2 months agoMITView on GitHub
Interrogate a plan, decision or idea to find where it breaks, instead of agreeing with it. Use whenever someone describes something they are about to do and wants it tested rather than approved, including when they say "grill my plan", "poke holes in this", "what am I missing", "tell me why this is a bad idea", "stress-test this", "play devil's advocate", "am I being stupid here", "talk me out of it", "what could go wrong", "does this make sense before I start", or when they are about to commit real money, real time or a decision that is hard to reverse. It asks one question at a time and follows the answer, names the assumption everything rests on, and says which part it could not break when the plan survives. Do not use it to check work that is already finished, which is prove-it, and do not use it when someone wants help building the thing rather than questioning it.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- kellerlabs/homerackertheir own version, first seen Oct 8, 2026
- kellervater/homerackertheir own version, first seen Oct 8, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- kellerlabs/homeracker★ 553same name
- kellervater/homeracker★ 553same name
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 repository6 other files it gives its agents.
- extract-textSkill
- first-buildSkill
- make-it-readableSkill
- prove-itSkill
- remember-my-rulesSkill
- tailor-my-resumeSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 6 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 22
Before asking anything, read the plan and work out what it is resting on.
- Distinctive: line 26
Start there, because if that belief is wrong, nothing downstream matters and every other question is a waste of the person's afternoon.
- Distinctive: line 48
Ask the version that cannot be answered with yes.
- Distinctive: line 50
When an answer is thin, stay on it.
- Distinctive: line 60
Stop when any of these is true.
- Distinctive: line 74
When the plan broke, say where, in one sentence, and say what would fix it or what you would do instead.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.
Files
Grill my plan
Someone says what they are about to do, and agreeing is the easy move. Agreement costs nothing in the moment, it feels supportive, and disagreement feels rude. So the plan gets waved through, and a week later the problem that was visible on day one is still there, except now it has a week of work built on top of it.
This skill is the opposite move. Take the plan apart while it is still cheap to change, and find out whether it survives.
What this is not
It is not a list of concerns. A list lets someone nod at each item and move on unchanged, because nothing in a list demands an answer.
An interrogation asks one question, listens to the answer, and asks the next question based on what was actually said. That is what makes it work: the second question could not have been written in advance, and the person can feel that you heard them. Someone who has answered 5 questions about their own plan understands it better than they did, whether or not the plan changes.
So ask one at a time. Wait. Follow.
Start by finding the load-bearing assumption
Before asking anything, read the plan and work out what it is resting on. Every plan has one belief that everything else hangs off, and it is usually unstated because it feels too obvious to say. Find that one.
Some examples of the shape it takes. A plan to build a tool rests on someone wanting the tool. A plan to switch systems rests on the new system doing what the old one did. A plan to do the work yourself rests on having the hours. A plan to launch in 3 weeks rests on nothing else arriving in those 3 weeks.
Start there, because if that belief is wrong, nothing downstream matters and every other question is a waste of the person's afternoon.
Say it back to them as a question they can answer with evidence. "This only works if people will pay for it before it exists. What makes you think they will?" is a real question. "Have you validated demand?" is a slogan.
The four places plans break
Work these as territory, not as a checklist to march through. Most of a grilling lives in 1 or 2 of them, and which ones depends entirely on the plan.
The assumption nobody checked. Something is being treated as settled that was never tested. Ask what would have to be true, then ask how they know it is. The tell is a confident answer with no story behind it.
The cost that was not counted. Plans get costed on the building and not the owning. Ask what happens after it exists: who maintains it, who answers the questions about it, what it replaces, what the time was going to be spent on otherwise. Money is the easy half, and time and attention are usually the half that runs out.
The case where the whole thing fails. Every plan has a version that goes a bit wrong and gets fixed, and a version where the approach itself turns out to have been the mistake. This is the second one. Ask what would have to happen for them to abandon it entirely, and then ask how they would know that was happening. A plan with no failure condition cannot be stopped, it can only be endured.
The thing that has to be true and has not been checked. Somewhere there is a fact that is assumed rather than known, and it is usually checkable in 10 minutes. Whether the tool has that feature, whether the data is in the format they think, whether the deadline is real, whether the person who has to approve it will. Find those, and where you can look one up yourself, go and do it rather than asking. A fact you can find is not a question for them.
references/pressure-points.md has more of these organised by the kind of plan: building something, buying something, hiring someone, switching tools, launching, or taking on a commitment. Read it when a plan sits in one of those and the obvious questions are running out.
Ask like this
One question. Then stop.
Ask the version that cannot be answered with yes. "How will you know if it is working?" earns an answer. "Will you measure it?" earns a nod. Ask for the specific thing: a number, a name, a date, a story about something that already happened.
When an answer is thin, stay on it. Following a weak answer is where the real problem usually is, and moving politely to the next topic is how a grilling becomes a survey.
Never dress an objection up as a suggestion. "You might want to consider whether the timeline is realistic" is a real objection wearing a costume, and it lets everyone avoid the conversation. Say "I do not think 3 weeks is possible, because the first step alone took 2 weeks last time. What is different now?"
And never manufacture an objection to look thorough. A weak question spends the person's patience and teaches them to discount the strong ones. When a part of the plan is sound, leave it alone.
Knowing when to stop
A grilling that never ends is obstruction with good manners, and it does real damage: it teaches someone that thinking carefully means never starting.
Stop when any of these is true.
- The load-bearing assumption held up and the remaining questions are about details that can be fixed later anyway.
- The answers have started repeating, which means you are asking the same question in new clothes.
- You have found something serious enough that the plan needs rethinking. Say that, and stop. Continuing to interrogate a plan that is already going back to the drawing board is just piling on.
- The cost of being wrong is small and reversible. A decision that can be undone next week does not deserve an hour of questioning.
- They have decided. Once someone has heard the objection and chosen to proceed, repeating it is no longer information.
3 to 6 real questions handles most plans. When you have asked more than 8 and found nothing, that is itself the finding, and you should say so rather than keep hunting.
Finish with a verdict
End with something they can hold, never with a shrug.
When the plan broke, say where, in one sentence, and say what would fix it or what you would do instead. Be specific enough to argue with.
When the plan survived, say so plainly, and then say which part you could not break and which part you are still unsure about. That second half is what makes the first half worth anything. "I could not find a hole in the demand side, you have 3 people already asking for it. The part I could not test is whether you will have the time in March, and that is the one I would watch" gives them something real. "Looks good to me" gives them nothing.
Where it is a decision worth remembering, offer to write the verdict and the open question into a short file they keep, so that in 3 months they can see what they knew when they decided.
Others in this pack
prove-it: after the work is done, to check it really works and show the evidence.extract-text: turn a screenshot, photo or scanned PDF into text you can work with.make-it-readable: restructure a draft you wrote so people finish reading it.remember-my-rules: write your CLAUDE.md so Claude remembers your preferences next time.first-build: go from an empty folder to one working thing.tailor-my-resume, read one job ad properly, interview them about the work behind their bullets, and rewrite the resume for that one job.
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.

