Skip to content

/prompt

Build a prompt from scratch with formatting-core

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

/prompt — Format and Execute

v2.1 — Opus 4.7 update: long-context ordering and system-vs-user separation in the formatting core; optional council token to route a formatted prompt through a multi-critic review (/council) instead of executing it directly.

Format an informal request into a structured prompt, then execute it.

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:

  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. Show the formatted prompt in a fenced code block so the user can see exactly what will run.

  10. Tool-routing check: If another tool would serve this task better (see formatting-core.md), add a brief note before executing. Don't block — just flag it.

  11. Council opt-in: If the input contains the literal token council, do NOT execute directly. Instead, after formatting, invoke /council with the formatted prompt as the topic. The council token is opt-in only — /prompt does NOT default-wrap in council. This prevents accidental council dispatches from casual /prompt uses.

  12. Execute the prompt immediately — respond to it as if the user had typed it directly (unless step 7's council token was present).

  13. Ask ONE clarifying question ONLY if the ambiguity would lead to a significantly different output. Otherwise, make reasonable assumptions and proceed.

Important

  • Do NOT over-engineer simple requests. A 1-sentence ask doesn't need a 20-line prompt.
  • Match complexity of formatting to complexity of task.
  • Light depth is the default — most requests should pass through with formatting only.
  • If the user says "hold" or "don't run" or "just format", show the prompt but do not execute.
  • council token handling: opt-in only. /prompt X depth:deep council → format, then dispatch via /council. /prompt X → format + execute directly (no council).
  • Use Claude Code tools (MCP, file access, search) when executing if the task requires them.