agentleFS
randommoniclerandommonicle/claude-skills/enumerate-the-set/SKILL.md

enumerate-the-set

Skill27 starsChanged 10 days agoApache-2.0View on GitHub

A change, reset, guard or proof that touches some members of a set silently leaves the rest as they were, and the member nobody listed is the one that breaks. Name the set, list it from its source (the type, the schema, the catalogue, the directory, the enum) rather than from the change in front of you, then decide every member explicitly in the code and in a test; give every start condition a matching end; and ask a reviewer which member is not covered rather than whether the change is right. Triggers when resetting, clearing, migrating, widening or tearing down state; when handling a kind, type, status, mode or variant and the handler mentions some of them; when writing a guard, filter or visibility rule over a set of statuses; when a feature has more than one anchor, leg, write or exit path; and whenever the words "and the others", "etc", "the main case" or "the normal path" appear. Does not fire on the places a single fact is duplicated (blast-radius-grep) or on the ordering of external events (unordered-events).

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:

    List it from its source, never from the change: Read the enum, the type, the schema, the catalogue, the directory, the switch statement.

    line 29

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

Enumerate the set

Eight failures, one shape: somebody changed, reset, guarded or proved some members of a set and the rest kept their old behaviour without saying so. An episode reset named four fields and left the event cursor pointing into the previous episode. A handover step consumed the pending wait request, because only the normal step kind had been tested. A visibility rule listed the statuses it would show and missed one, so work in that status became invisible. Two accounting anchors each closed honestly and the general-ledger leg was never written at all, so client money was absent from every accounting report. A relay supervisor was proven on its clean exit and on no other.

The set was always knowable. It was in a type, an enum, a schema, a directory listing, a state diagram: somewhere a program could have printed it. In each case it was taken instead from the change under the author's nose, which contains exactly the members that happened to come to mind.

The rules

  1. Name the set out loud before you touch it. "The fields of the episode state file." "The step kinds." "The statuses a writer can produce." "The exit paths of this supervisor." "The accounting legs of a receipt." "The anchors over this feature." If you cannot name the set in one phrase, you do not yet know what you are changing, and the member you have not thought of is already excluded.
  2. List it from its source, never from the change. Read the enum, the type, the schema, the catalogue, the directory, the switch statement. Prefer a listing a command produces to one you typed. A list assembled from the diff in front of you reproduces the diff's blind spot. Where the source is a hand-kept list, that list's own omissions are the risk (prove-it-can-pass rule 6).
  3. Decide every member explicitly, including the ones you are leaving alone. "Unchanged" is a decision and is recorded as one, in a comment or a table, so the next reader can tell a considered exclusion from an oversight. An exhaustive switch with no default, a discriminated union the compiler checks, or a test parameterised over the listed set turns the decision into something a build can enforce (enforce-invariants-in-build).
  4. Every member gets a case, not just the representative one. A test over the normal step kind proves the normal step kind. The handover step was the one that broke. Parameterise over the enumerated set so that adding a member to the source makes the suite demand a case the only arrangement where the next member added cannot be forgotten. Library 44 is this rule inside a single test: a can-fail proof on one alternative of a pattern says nothing about the pattern's other alternatives.
  5. A start names its end, and an open names its close. Anything that begins a state (a fight start, a lock, a tuning override, an episode, a held resource) is written together with the condition that ends it, and the end is tested. Mod-settings tuning outlived its colony and a fight start was widened with no matching end: both are a set of size two with one member written.
  6. Every leg of a multi-leg write is a member. A transaction with two sides, an action with a ledger entry, a status change with an audit row: the set is the legs, and a leg that is never written fails silently because each individual step reports success. Where money is involved this is the expensive one: two anchors closed honestly while the general-ledger leg was never written, and client money was absent from every accounting report.
  7. Exit paths are a set. Clean exit, error exit, timeout, kill, parent death, restart. A supervisor, teardown, cleanup handler or finally block proven on the clean path is proven on one member of six. List the paths and drive each one.
  8. Two constraints over the same set must be read together. A pair of rules can each be reasonable and jointly permit nothing: two sound constraints whose permitted transitions do not intersect meant the documented reversal could not be made by anybody. When a second constraint is added over a set the first already governs, compute the intersection and check the operations you intend are still in it.
  9. Ask a reviewer the enumerating question. "Is this right?" invites a reader to check the members in front of them. "Which member of this set is not covered?" forces them to fetch the set. Give them the set's source in the request, and the same question works on yourself.

Relationship to the neighbouring skills

blast-radius-grep enumerates the places one fact lives; this skill enumerates the members one change must decide. unordered-events enumerates (event, state) pairs for external signals and explicitly excludes in-process calls; this skill covers the in-process sets it leaves out: fields, kinds, statuses, legs, exit paths, anchors, constraints. prove-it-can-pass rule 6 is this rule applied to a gate's own curated configuration. enforce-invariants-in-build is how a decision over a set stops being prose.

Provenance

Rimagent 2026-10-04 (an episode reset named four fields and kept the event cursor), 2026-10-12 (a handover step consumed the pending wait request; only the normal step kind was tested), 2026-10-13 twice (a fight start widened with no matching end; a visibility rule listed the statuses it showed and missed between_steps), 2026-10-16 (mod-settings tuning outlived its colony). PropOS 2026-09-21 (two anchors each closed honestly and the general-ledger leg was never written, so client money was absent from every accounting report), 2026-09-21 (two sound constraints whose permitted transitions do not intersect, so the documented reversal cannot be made), 2026-09-28 (a relay supervisor proven on its clean exit only). Library 44 is the same shape inside a test and stays a prove-it-can-fail row rather than being duplicated here. docs/SKILL_PROPOSALS_2026-11-01.md, Tier 1 item 2, admitted by Ben 2026-11-05. The PropOS instances are summarised as that document records them; the originals are in that repository's own lessons file and were not read for this build.

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.