What problem does it solve? Writing effective behavior instructions for AI agents (CLAUDE.md, AGENT.md, system prompts) is hard: most people produce vague persona descriptions instead of concrete operational rules. This Skill turns fuzzy preferences into a portable, agent-agnostic behavior file with explicit defaults, pushback rules, and failure-avoidance constraints. ## Core Features & Use Cases - Preset Mode: Instantly returns one of four built-in profiles (fast, balanced, strict, brutal) from the presets/ directory as a ready-to-save markdown block. - Interview Mode: Runs a short, adaptive Q&A that converts vague preferences ("don't be annoying", "be smart") into concrete operational rules covering disagreement, vagueness handling, teaching style, and uncertainty. - V2 Refinement Pass: A tightening mode that pressure-tests vague answers, forces concrete tradeoffs, and produces sharper, more enforceable rules. - Use Case: A developer setting up a new repo asks for a CLAUDE.md. The Skill interviews them about how much pushback they want and how the agent should handle ambiguity, then outputs a savable behavior file that also works for Codex, ChatGPT custom instructions, or local agents. ## Quick Start Ask the agent to interview you and generate a personalized CLAUDE.md behavior file, or request one of the fast, balanced, strict, or brutal presets directly.
