How do I install the Grilling Plan skill?

Run npx skillmds@latest add socketdev/grilling-plan in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Grilling Plan safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Grilling Plan?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Grilling Plan free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Grilling Plan?

socketdev (@socketdev) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Grilling Plan by socketdev — Agent Skill

Name: Grilling Plan
Author: socketdev

File contents

grilling-plan

The proactive complement to plan-review-nudge (the after-plan gate). Interview the user relentlessly about every branch of the design tree until you reach shared understanding — then the plan is ready to land.

1. Resolve the deterministic questions first

Before asking the user anything, answer every question the codebase already answers — this is the code-first-then-ai discipline applied to design. If the change proposes a NEW enforcement surface (hook, lint rule, check, skill, agent), run the inventory and compare:

node scripts/fleet/codify-scan/inventory.mts

It emits the authoritative set of existing hooks / lint rules / checks / docs. If the thing you'd build already exists, the answer is "extend it", not "add one" — don't ask the user a question grep or the inventory settles. Read the relevant files in place. Only what the codebase cannot answer goes to the user.

2. Interview one question at a time

Ask the residue — the design-intent questions no script resolves: trade-off framing, naming, cascade blast-radius, the reversal condition. Rules:

One question at a time. Wait for the answer before the next. Multiple questions at once is bewildering.
Every question carries your recommended answer and the reason, so the user can accept with one word.
Walk the tree in dependency order — resolve a decision its children depend on before descending.
If a later answer can be derived from an earlier one, derive it; don't re-ask.

3. Gate before landing

A fleet-shared change is settled only when the plan has the shape plan-review-nudge checks for:

Numbered steps, each naming the real files + rules it touches.
The final name / schema shape decided up front (a name that will cascade must not churn across commits).
A second-opinion invite when the plan edits a fleet-shared resource (CLAUDE.md, template/, .claude/hooks/, _shared/).

Completion criterion

Every design-tree branch is resolved (by the codebase or by the user), the plan names its files + rules in numbered steps, the cascaded names are settled, and — for a fleet-shared change — a second opinion has been invited. Then write the plan to .claude/plans/<name>.md and execute.

---name: grilling-plandescription: Stress-test a plan one question at a time before building, especially for shared fleet resources.---# grilling-planThe proactive complement to `plan-review-nudge` (the after-plan gate). Interviewthe user relentlessly about every branch of the design tree until you reachshared understanding — then the plan is ready to land.## 1. Resolve the deterministic questions firstBefore asking the user anything, answer every question the codebase alreadyanswers — this is the `code-first-then-ai` discipline applied to design. If thechange proposes a NEW enforcement surface (hook, lint rule, check, skill, agent),run the inventory and compare:```shnode scripts/fleet/codify-scan/inventory.mts```It emits the authoritative set of existing hooks / lint rules / checks / docs.If the thing you'd build already exists, the answer is "extend it", not "add one"— don't ask the user a question `grep` or the inventory settles. Read therelevant files in place. Only what the codebase *cannot* answer goes to the user.## 2. Interview one question at a timeAsk the residue — the design-intent questions no script resolves: trade-offframing, naming, cascade blast-radius, the reversal condition. Rules:- **One question at a time.** Wait for the answer before the next. Multiple questions at once is bewildering.- **Every question carries your recommended answer** and the reason, so the user can accept with one word.- **Walk the tree in dependency order** — resolve a decision its children depend on before descending.- If a later answer can be derived from an earlier one, derive it; don't re-ask.## 3. Gate before landingA fleet-shared change is settled only when the plan has the shape`plan-review-nudge` checks for:- Numbered steps, each naming the real files + rules it touches.- The final name / schema shape decided up front (a name that will cascade must not churn across commits).- A second-opinion invite when the plan edits a fleet-shared resource (`CLAUDE.md`, `template/`, `.claude/hooks/`, `_shared/`).## Completion criterionEvery design-tree branch is resolved (by the codebase or by the user), the plannames its files + rules in numbered steps, the cascaded names are settled, and —for a fleet-shared change — a second opinion has been invited. Then write the planto `.claude/plans/<name>.md` and execute.

Frequently asked questions

Grilling Plan