Grill plan
Skill stark-ai-de/agent-skills/incubator/skills/productivity/grill-plan
Public Agent Skills for Codex operations, Cursor operations, Claude operations, repo maintenance, skill maintenance, productivity, and engineering workflows.
npx -y skills add stark-ai-de/agent-skills --skill grill-planAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 4 stars4 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.
What its author says it does
Copied from the file, not written here
Relentlessly question a plan, design, migration, release, or repo maintenance proposal until assumptions, risks, constraints, and decision branches are clear. Use when the user wants to be grilled before implementation or wants a plan challenged.
The file declares its own license as Apache-2.0. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
3.2 KB, as published. Nobody here has run it
Grill Plan
Goal
Pressure-test a plan before execution so hidden assumptions, risky branches, missing approvals, and validation gaps become explicit decisions.
When to use
- The user asks to be grilled, challenged, or questioned before implementation.
- A migration, release, architecture change, or repo maintenance plan has unclear constraints.
- The user wants decision branches clarified before work starts.
When not to use
- Do not use when the user has already asked for implementation and the plan is clear.
- Do not use for passive code review; use the relevant review skill.
- Do not use to delay an urgent fix when the blocking decision is already known.
Inputs to inspect
- Inspect the plan, issue, PR, design doc, repo constraints, validation commands, and approval boundaries.
- Explore code or docs directly when a question can be answered from local evidence.
Inputs
- The proposed plan, design, checklist, issue, PR, or release notes.
- Repo constraints, workflow docs, validation commands, and known risks.
- User goals, deadlines, approval boundaries, and rollback expectations.
Process
- Restate the objective and the riskiest assumption.
- Ask focused questions that change the plan if answered differently.
- Group questions by decision branch, not by curiosity.
- Identify missing evidence, approvals, rollback plans, and validation.
- Convert answers into a tightened plan or explicit blockers.
- Stop when the next action is clear enough to execute or reject.
Workflow
Use the process above, asking one high-leverage question at a time. End with a tightened plan, explicit blockers, or a decision that implementation should not proceed.
Decision points
- If a question does not change the plan, omit it.
- If the plan can fail irreversibly, require an approval and rollback answer.
- If uncertainty is acceptable, mark it as a conscious risk instead of blocking.
Safety rules
- Do not perform implementation while grilling unless the user switches modes.
- Do not overwhelm the user with a long questionnaire; ask the highest-leverage questions first.
- Do not treat assumptions as decisions.
References
Read references/questioning-rubric.md when deciding which questions to ask first.
Scripts
No bundled scripts.
Output format
Return:
- Restated objective
- Highest-risk assumption
- Focused questions
- Decision branches
- Tightened plan or blockers
Completion criteria
- Key assumptions are explicit.
- The next execution step is clear.
- Remaining risks are accepted, mitigated, or blocked.
Failure modes
- If the user wants implementation instead, stop grilling and switch to execution.
- If a question can be answered from the repo, inspect the repo instead of asking.
- If the plan is too broad, recommend splitting it before continuing.