Skip to content

/prompt-only

Skill that only produces a prompt (no execution)

Category: meta
Field: general
License: MIT
Updated: 2026-04
Stages: —

/prompt-only — Format Without Executing

v2.0 — Format informal requests into structured prompts. Output only — do not execute.

Format an informal request into a structured prompt. Output only — do not execute.

Reference Files

@~/.claude/commands/prompt-references/formatting-core.md

Input

$ARGUMENTS

Instructions

You are a prompt formatter. The user has given you an informal, conversational request (possibly dictated). Your job is to produce a clean, well-structured prompt they can use anywhere — Claude Code, Claude.ai, ChatGPT, or other tools.

  1. Parse the intent: Extract the core task, audience, and desired output from the informal input.

  2. Calibrate depth using the heuristic in formatting-core.md:

  3. Light (default): Format only. No depth injection.
  4. Standard: Format + append assumptions/rationale block.
  5. Deep: Format + append research/compare/verify block.

  6. User can override with depth:light, depth:standard, or depth:deep.

  7. Format into a structured prompt using the formatting elements in formatting-core.md. Apply elements as appropriate — match formatting complexity to task complexity.

  8. Inject depth directives if Standard or Deep (per the templates in formatting-core.md). For Light, skip this step entirely.

  9. Output the formatted prompt in a clean fenced code block.

  10. Tool-routing recommendation: After the code block, add **Best run in:** [tool] — [reason] if another tool would serve better (see formatting-core.md). If Claude Code is the best fit, omit this line.

  11. If the prompt looks reusable (template, workflow, recurring task):

  12. Add a version header: ## Prompt v1.0 — [short name]

  13. Suggest 3-5 eval test cases: brief input/expected-output pairs to verify quality

  14. If the prompt has agent/workflow context (system instructions vs. user turn):

  15. Separate into System Prompt and User Prompt sections within the code block

  16. Do NOT execute the prompt. Output only.

Important

  • Do NOT over-engineer simple requests. Match formatting complexity to task complexity.
  • Light depth is the default — most requests should pass through with formatting only.
  • For one-off prompts, skip the version header and eval cases.
  • Keep the prompt self-contained — someone with no context should be able to use it.