// Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
Use when creating, updating, refactoring, or reviewing an AI assistant or sub-agent system prompt - persona, role, capabilities, tools, output format, examples, and constraints. Covers Claude Code agents, OpenAI Codex / Responses-API agents, Pi assistants, and OpenCode agents. Use even if the user only says "agent prompt", "assistant", "subagent", "persona", or names the artefact by file path.
Use when creating, updating, or reviewing a slash command - shims that delegate to a skill or agent, standalone commands with an inline output format, and the `description` / `argument-hint` / `model` headers per provider. Covers Claude Code custom commands and the merged skill-as-command format, OpenCode commands, Pi prompt templates, and the legacy Codex `/prompts:` route. Use even if the user only says "slash command", "prompt template", "command shim", "create-command", or names the artefact by path.
Load when working with Nix, NixOS, Home Manager, nix-darwin, nixpkgs, flakes, derivations, overlays, modules, options, or registries; with .nix files such as configuration.nix, home.nix, default.nix, shell.nix, or flake.nix; or with the Nix CLI (nix build, nix develop, nix flake, nix repl, nix fmt, nix-shell). Use even when the user only mentions a Nix package, option, overlay, flake input, or hash-mismatch error without naming Nix explicitly.
Use when creating, updating, consolidating, or reviewing an AGENTS.md (or CLAUDE.md, CLAUDE.local.md, .claude/rules/*.md, .cursorrules, .cursor/rules/*.mdc, AGENTS.override.md, or .github/instructions/*.instructions.md) project instruction file. Covers the open agents.md spec, Codex precedence rules, Claude Code memory loading, and migration from legacy formats. Use even if the user only says "instructions", "rules", "project memory", or names a single legacy filename.
Use when creating, updating, or reviewing an Agent Skill - authoring or revising a `SKILL.md`, its frontmatter, layout, references, and progressive disclosure. Use when the user mentions writing, editing, splitting, renaming, or auditing a skill, even if they do not say "skill" explicitly. Covers cross-platform portability across Claude Code, Codex, OpenCode, and Pi.
Load when working with LÖVE 2D, the LÖVE engine, love2d, .love archives, Lua 5.1/LuaJIT 2.1 game development, or LÖVE callbacks, modules, conf.lua, and packaging.
| name | grill-me |
| description | Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me". |
| user-invocable | true |
Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.
Ask the questions one at a time.
If a question can be answered by exploring the codebase, explore the codebase instead. Answer obvious questions automatically - report what you decided and why but do not ask permission