Autonomously optimize a Claude Code skill or agent system by running it repeatedly, scoring outputs against evals, mutating owned artifacts (prompt, references, scripts, agent definitions), and keeping improvements. Karpathy's autoresearch methodology. Trigger when the user asks to run autoresearch, or to optimize, benchmark, or run evals on a skill or agent. Also trigger on ์คํฌ ๊ฐ์ , ์คํฌ ์ต์ ํ, ์คํฌ ๋ฒค์น๋งํฌ, ์คํฌ ํ๊ฐ, ์์ด์ ํธ ๊ฐ์ , ์์ด์ ํธ ์ต์ ํ.
Agentic system design blueprint generator. Interviews the user to understand a task they want to automate, then produces a comprehensive integrated design document (.md file) that serves as a concrete implementation plan for Claude Code. The design document includes task context, workflow definition with LLM vs code boundaries, and implementation spec (folder structure, agent architecture, skills/scripts list, sub-agent design). Trigger on "/blueprint", "blueprint", "์์ด์ ํธ ์ค๊ณ", "์ค๊ณ์ ๋ง๋ค์ด", "agentic workflow design". Use this skill when the user mentions automation design, workflow planning, agent architecture, system blueprinting, or multi-step AI workflow planning.
Socratic interview skill to deepen a spec or refine an existing agent blueprint. Use when the user wants to clarify a task, stress-test design decisions, or fill gaps in a blueprint produced by /blueprint. Trigger on "/deep-dive", "deep dive", "interview me", "create a spec", "์๊ตฌ์ฌํญ ์ ๋ฆฌ", "๊ธฐํ์ ๋ง๋ค์ด", "์คํ ์์ฑ", "blueprint ์ฌํ", "์ค๊ณ ๋ณด๊ฐ".
Autonomously optimize any Codex skill or agent system by running it repeatedly, scoring outputs against evals (binary for rules + comparative for quality), mutating any owned artifact โ the skill's prompt, reference assets, and executable artifacts (scripts, agent/subagent definitions like `agents/openai.yaml`, MCP servers, hooks, harness code) โ and keeping improvements. Based on Karpathy's autoresearch methodology. Use this skill whenever the user mentions optimizing a skill, improving a skill or agent, running autoresearch, making a skill or agent better, self-improving a skill, benchmarking a skill, evaluating a skill, running evals on a skill, optimizing an agent system, or any request to iteratively test and refine a skill or agent โ even if they don't use the word "autoresearch" explicitly. Also trigger on ์คํฌ ๊ฐ์ , ์คํฌ ์ต์ ํ, ์คํฌ ๋ฒค์น๋งํฌ, ์คํฌ ํ๊ฐ, ์์ด์ ํธ ๊ฐ์ , ์์ด์ ํธ ์ต์ ํ. Outputs an improved target skill file (and any mutated executable artifacts), a results log, a changelog, and a research log of meaningful direction
Session wrap-up and fix-learning capture skill. Use when the user explicitly asks for "/reflect", "reflect", "session reflect", "end session", "์ธ์ ์ ๋ฆฌ", "์ค๋ ํ ๊ฑฐ ์ ๋ฆฌ", or "๋ง๋ฌด๋ฆฌ", or when the user clearly confirms that a concrete issue from the current coding session is fixed ("๊ณ ์ณ์ก๋ค", "ํด๊ฒฐ๋๋ค", "it's fixed", "working now", "that worked", "problem solved"). Explicit invocation runs Full Reflect; confident fix confirmation runs Quick Reflect.
Codex์ฉ ์๋ํ ์์ด์ ํธ ์์คํ ์ค๊ณ์ ์์ฑ ์คํฌ. ์ฌ์ฉ์ ์ธํฐ๋ทฐ๋ก ์๊ตฌ์ฌํญ์ ์ ๋ฆฌํ๊ณ Codex ๊ตฌํ ๊ธฐ์ค์ ํตํฉ ์ค๊ณ ๋ฌธ์(.md)๋ฅผ ์์ฑํ๋ค. Use when the user asks for "$blueprint", "blueprint", "์์ด์ ํธ ์ค๊ณ", "์ค๊ณ์ ๋ง๋ค์ด", "agentic workflow design", or any request to plan a Codex automation or agent workflow.
A Codex skill for Socratic interviews that deepen a spec or refine an existing agent blueprint. Conducts forcing-question interviews, challenges premises, generates alternatives, and validates the final document with a document-grounded reviewer pass. Use when the user asks for "$deep-dive", "deep dive", "interview me", "create a spec", "์๊ตฌ์ฌํญ ์ ๋ฆฌ", "๊ธฐํ์ ๋ง๋ค์ด", "์คํ ์์ฑ", "blueprint ์ฌํ", or "์ค๊ณ ๋ณด๊ฐ", or when they have a rough idea or an existing blueprint that needs structured stress-testing in Codex.
Session wrap-up and fix-learning capture skill for Codex. Use when the user explicitly asks for "$reflect", "reflect", "session reflect", "end session", "์ธ์ ์ ๋ฆฌ", "์ค๋ ํ ๊ฑฐ ์ ๋ฆฌ", or "๋ง๋ฌด๋ฆฌ", or when the user clearly confirms that a concrete issue from the current coding session is fixed ("๊ณ ์ณ์ก๋ค", "ํด๊ฒฐ๋๋ค", "it's fixed", "working now", "that worked", "problem solved"). Explicit invocation runs Full Reflect; confident fix confirmation runs Quick Reflect.