Community程式設計與開發github.com

dtiger1889-ops/breakdown

Convergent, action-first problem decomposition (the terse twin of /fanout). Use ONLY when the user types /breakdown or says "break this down" / "break it down" / "decompose this". Takes ONE tangled or ambiguous problem and returns the single smallest next ACTION plus a collapsed full plan. Terse, no hand-holding, no capability narration. Do NOT use for open-ended ideation with many valid answers (that is /fanout), and do NOT use when the task is already one clear action.

breakdown 是什麼?

breakdown is a Claude Code agent skill that convergent, action-first problem decomposition (the terse twin of /fanout). Use ONLY when the user types /breakdown or says "break this down" / "break it down" / "decompose this". Takes ONE tangled or ambiguous problem and returns the single smallest next ACTION plus a collapsed full plan. Terse, no hand-holding, no capability narration. Do NOT use for open-ended ideation with many valid answers (that is /fanout), and do NOT use when the task is already one clear action.

相容平台~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/dtiger1889-ops/claude-harness-toolbox/tree/main/skills/breakdown

Installed? Explore more 程式設計與開發 skills: steipete/bluebubbles, steipete/eightctl, steipete/blucli · View all 6 →

在你喜歡的 AI 中提問

開啟一個已預先載入此 Agent Skill 的新對話。

說明文件

breakdown -- convergent, action-first decomposition

The convergent complement to /fanout. /fanout opens a problem up (many valid answers); /breakdown closes one down (one tangle -> one next step). Single-shot, no sub-agents, ~1x cost. Grounded in the maintainer's session-transcript crawl: the loudest friction was "stop explaining, just do it," so this skill is ACTION-FIRST and minimal-prose by design. If your output reads like a gentle walk-through, it is wrong -- cut it.

Step 1 -- get the tangle

Accept the problem from an @-mention, a path, or pasted text. If it is genuinely vague, ask for one or two sentences -- once. Read any referenced file to EOF before decomposing. If the tangle is (or names) a project artifact, ALSO read that project's CHECKPOINT.md to EOF first (receipt/preview is not a read) -- the "real goal" and the "blocking decision" depend on live project state (deadlines, sequencing gates, what's already decided), and a decomposition blind to it points at the wrong next action. Skip only if genuinely project-free.

Step 2 -- output exactly these five blocks, tersely, in order

No preamble. No "I have access to...". Do not restate the question back at length.

  1. Real goal (one sentence). What "done" actually means. If the stated goal differs from the real one, say so in one line.
  2. Smallest next action (one concrete, physical step doable in one sitting -- e.g. "open X and write the three field names," not "design the schema"). Then OFFER to just do it now. Do not explain at length first.
  3. The one blocking decision -- the single fork that unlocks the most, or "none -- just start."
  4. Full plan (collapsed) -- the complete checklist under a one-line summary, explicitly de-emphasized: "ignore until step 2 is done."
  5. First trap -- the single most likely failure mode and its earliest signal.

Rules

  • One thread on the critical path. Do not list parallel workstreams.
  • Plain language, no invented codes (translate any filing tags to people-words).
  • Do NOT start the work until the user picks. /breakdown plans; it does not execute.
  • Persist nothing by default. If worth keeping, offer to write it into the project's CHECKPOINT "Next step" line.

Individual skills in this repo

This repo contains 3 individual skills — each has its own dedicated page.

相關技能