yodakeisuke/structuring-skill/structuring
Claude Code and agent skill by yodakeisuke/structuring-skill
- community
Checking that the SKILL.md file is still in the repository.
Decompose a vague whole into a MECE tree, exhaustive for its purpose, down to actionable leaves. Use for logic trees, issue trees, MECE breakdowns, logical thinking, or any request to structure or break down a subject.
Install
npx vskill@latest install yodakeisuke/structuring-skill/structuringSKILL.md
The same SKILL.md was published earlier as yodakeisuke/structuring-skill/structuring.
Structure a subject recursively into a tree. Keep the structure visible, check the board as you go, correct course, and converge step by step.
In/Out
input:
- a defined scope: what whole to split, for what purpose, and over what time horizon
output:
- a tree broken down to actionable granularity, every sibling row a MECE split
Work iteratively
Abstract from the details; use the resulting structure to find the gaps.
- Rework is welcome — a corrective move is evidence you tested your thinking by doing. A one-pass finish with no interim findings and no course correction proves you skipped the thinking. Use revisit / reopen / group for structural fixes and rename for rewording — both kinds are counted
- Be strict with yourself about completion — don't give the board's questions a perfunctory "yes"; keep fixing until you are fully satisfied with your own work
Procedure
Follow this recursive procedure. Each step invokes either a script or a natural-language function defined below.
There are always blind spots and unknown unknowns. Take deliberate measures to root them out.
<natural-language-functions> removal_test: (x: node) → (op: + | × | t, head: element) Remove one element. What happens to x? (if more than one op fits x, choose according to the purpose) it shrinks but remains → +, head = the element removed it vanishes → ×, head = the variable that is easiest to measure (the rest is determined as the quotient) it is a different thing before and after a turning point → t, head = the period up to the turning point
naming: (rest: set of elements) → a name + axis | next head (element) | neither Can the rest be named as a thing in its own right (a domain term)? Is the name at the same level of abstraction as the heads? named, same level → a name, and the axis — the one cut that makes the row MECE (no single cut → neither) named, but a level above: it summarizes several → the next head, from inside it not named, but another head can be picked out → the next head neither → neither
abstract: (x: node, children: built) → abstract | fix Re-derive x from the children that now exist. Is it still x — one cut, adding back up to x, no sibling the leaves revealed? yes → abstract (say in note what the children revealed, even "nothing new") no → add(x, head) / group / reopen / revisit </natural-language-functions>
To keep the context lean through incremental updates, operate the board only through the scripts. Do not read the skill's other files (scripts/, ui/) — work from the scripts' outputs alone. Path = scripts/tool/ + the file name from the tables below. Example: Run from your project working directory as-is (never cd into the skill): the board lives in <cwd>/structuring, and the Workbench and the critic find it there.
Tools for structuring: <tool-like-scripts> new.ts {"whole":"the whole being split","purpose":"what it is for","horizon":"the time horizon"} with --critic add "critic":true tree/leaf.ts {"note":"one action for this leaf"} tree/abstract.ts {"note":"what the children revealed about x"} confirms the row "Next:" points to, seen from its children (all complete); once every child of a row is abstracted, the row itself is asked tree/cut.ts {"note":"result of the removal test","op":"+","head":"the head element"} op = + | × | t tree/name.ts {"note":"why the rest can be handled as one","rest":"the name given to the rest","axis":"what divides the siblings (≤ 20 chars; + and t rows)"} tree/add.ts {"note":"why this is the next head","head":"the next element"} with "x":"n1": inserts the head into a closed row on the way to the current node (an ancestor still being walked) — a settled row needs reopen first ↑ leaf / cut / name / add may carry an optional "gloss":"the one word or number that pins the name down (売上高 → 1,200億, 武力による抑止 → 力で押さえる); ≤ 20 chars. Name = bare term, gloss = which/how much, note = why" tree/revisit.ts {"note":"why revisit (c1 etc. if from user feedback)","x":"n3"} x = node id (its subtree is discarded) tree/rename.ts {"note":"why rename","x":"n3","name":"new name"} replaces only x's name (shape unchanged); with "axis" instead of "name": rewords the axis of x's row; with \"gloss\" (alone or with name): sets x's gloss tree/reopen.ts {"note":"why reopen","x":"n3","drop":["n4"]} reopens the closed row under x (heads intact, rest parked — name it with the same name to restore it with its subtree. drop removes the listed heads too; optional) tree/group.ts {"note":"why they can be handled as one","xs":["n3","n4"],"name":"group name"} wraps heads of the same row in an intermediate node (subtrees intact) tree/undo.ts {"note":"why undo"} undoes the last move and restores the board (cannot undo a move made by the user) </tool-like-scripts>
Tools for working with the user: <tool-like-scripts> show.ts {} / {"journal":10} = the 10 most recent journal entries the user's comments and UI-originated moves appear at the top of every tool response under "From the user:" serve.ts {} / {"stop":true} = stop starts the user-facing Workbench. Open the URL it prints in the in-app browser (or the default browser if there is no in-app browser) ack.ts {"id":"c1","reply":"what you did or decided"} replies to a comment / question and closes it. Do not wrap up while any is left unanswered ask.ts {"text":"question","options":[{"text":"the one you recommend","why":"one line"},{"text":"the alternative"}]} asks for a fact or judgment only the user can supply. Lay out the options and continue under options[0]; the user answers with one tap via chat or the Workbench. When the answer arrives, update the tree with it, then close with ack </tool-like-scripts>
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/structuring/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/yodakeisuke/structuring-skill/structuring)