← ClaudeAtlas

meta-promptlisted

Turn a fuzzy ask into a sharp Claude Code prompt with context and acceptance criteria
ku5ic/dotfiles · ★ 0 · AI & Automation · score 72
Install: claude install-skill ku5ic/dotfiles
## When to use - You want to capture a reusable task as a new slash command - You are about to start a new Claude Code session and want to hand off context - You are handing a task to a teammate and want a shaped brief - A request came in vaguely and needs sharpening before you act ## Procedure 1. Get the scratch directory: `!`scratch-dir.sh``. 2. Read $ARGUMENTS. If it points to a file, read it. 3. Run `!`detect-stack.sh`` if the ask appears technical. 4. Identify what is missing: - Goal unclear or conflated with method - Scope undefined (no files, no surface area) - Success unstated (how do we know it worked) - Constraints missing (stack, style, deadline, performance) 5. Verify before writing. For any file path, symbol, or API the prompt is about to name, confirm it exists: read it directly when the input names a couple of files, or delegate to the scout agent when the input references code or architecture spanning more than a couple of files. Anything that cannot be verified goes into the output as "unverified, please confirm", not asserted. 6. Rewrite as a structured prompt. ## Output file Write to `$(scratch-dir.sh)/prompt-<slug>-<YYYYMMDD-HHMM>.md`. Print the path. Structure: ``` # Task: <clear one-line title> ## Goal <The outcome. What, not how.> ## Context <Relevant files, prior work, constraints the reader needs. Link or inline.> ## Inputs <What the agent or person has to work with. Paths, data, tickets.> ## Constraints <Stack, style, patte