agentleFS
ExagrowExagrow/dashboard-workshop/.claude/skills/inverse-set-kondo/SKILL.md

inverse-set-kondo

Skill0 starsChanged 37 days agoMITView on GitHub

Audit a set of rules, workflows, configs, checks or code by notionally emptying it and making every piece earn its way back. Use when something has accumulated and nobody is sure what is still needed, or when someone says "run the inverse-set Kondo", "declutter this", "what can we delete", "audit these rules", "do we still need all of this", "clean up the config". With no target named, it describes the method and stops.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not pick one yourself. Say what the method is in two or three sentences, ask what set to audit (a folder of workflows, a ruleset, a …

    line 20

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

The inverse-set Kondo

Not "what should we remove". Everything currently in place goes into a box and is notionally already gone. Then take each piece out, one at a time, and ask whether we actually want it in the world we are building now. If nothing about it earns its way back, it goes in the delete box.

Existing is not a reason to keep. Neither is "it was carefully built", "somebody wrote a long comment justifying it", or "removing it feels risky". Those are all arguments for reading it carefully before deciding, not for keeping it.

If no target was named

Do not pick one yourself. Say what the method is in two or three sentences, ask what set to audit (a folder of workflows, a ruleset, a config file, a list of data quality checks), then stop. Everything below applies only when there is a set to audit.

How to run it

Enumerate exhaustively first, by reading, not from memory. List every item in the set before judging any of them. Read the actual rules, workflow files, config keys or functions. A Kondo done from recollection audits your recollection.

Then, for each item, answer three questions in this order:

  1. What does it assert? In one sentence, in plain terms. If you cannot say what it asserts, that is already a finding.
  2. What breaks without it? Concretely, naming the failure. "Less safe" is not an answer.
  3. Does that thing still matter, and is this still the place to assert it? Often the invariant is real and has moved somewhere better, or is now guaranteed structurally, or is asserted twice and only one copy works.

Sort into three boxes, and say which:

  • Keep: with the reason it earned its place, which must not be "it is already there".
  • Change: it asserts something real in the wrong place, or with the wrong scope, or at the wrong time. Say what the new shape is.
  • Delete: nothing about it survives the three questions. Say what used to justify it and why that justification no longer reaches.

Watch for these four patterns, because they account for most of what a Kondo finds:

  • A rule that can only be satisfied by a flow nobody uses any more. It looks like enforcement and is a locked door with the key handed out.
  • The same invariant asserted twice, where one copy is structural and the other is checked. Keep the structural one.
  • A check whose failure nobody can see. It costs runtime and buys nothing.
  • Something whose stated justification is circular: it exists because of a condition that only exists because of it.

Where this sits

The other skills in this set judge what a person sees. This one judges what has piled up behind it.

  • jobs-quote-ux asks you to cut a screen to the essential. The Kondo is the same question asked of rules, checks, config and code.
  • ux-heuristics names the on-screen version in its eighth heuristic, aesthetic and minimalist design.
  • closed-loop-visual-feedback is how to confirm nothing a person relies on went missing: after deleting, render the thing and look.

Output

One table, ordered by box: everything deleted first, because that is the part worth reading. Columns: the item, what it asserts, the verdict, and the reason. Then a short count of what the set weighed before and after.

State plainly anything you could not verify and what it would take to settle it. A Kondo that guesses at one item's purpose has quietly kept or deleted it for no reason.

Propose the deletions; do not carry them out until the person agrees. Deleting is the one step here that is hard to take back.


Created by Exagrow AI Consulting. Provided under the MIT License.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.