agentleFS
marinvchmarinvch/Cortex/skills/define-terms/SKILL.md

define-terms

Skill1 starsChanged 18 days agoMITView on GitHub

Sharpen a repo's domain language — challenge fuzzy terms, and write its CONTEXT.md glossary and ADRs. Triggers — "what should we call this", "these two words mean the same thing", "record this decision", "write an ADR".

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Create each file when it first has content: CONTEXT.md when the first term is settled, docs/adr/ when the first decision qualifies.

    line 38
  2. Very specific:

    When CONTEXT-MAP.md exists, each context keeps its own CONTEXT.md beside its code (src/billing/CONTEXT.md).

    line 42
  3. Very specific:

    Hold the user to the glossary: When they use a term differently from its written definition, point it out at once and ask which meaning is …

    line 48
  4. Distinctive:

    Pin down vague words: When a word could mean two things, name both and ask which: "by *member* do you mean the person or their seat on the …

    line 51
  5. Distinctive:

    Test boundaries with cases: When two concepts are being related, invent a specific case that falls between them and ask where it belongs.

    line 53
  6. Very specific:

    Check claims against the code: When the user says how something works, read the code.

    line 55
  7. Distinctive:

    If any one is missing, do not offer it.

    line 75

≈ “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

/define-terms — one word per concept, written down when it settles

A codebase decays when one concept carries two names, or one name covers two concepts. This ritual is for the moments the model is being changed: a term is disputed, a boundary is unclear, a decision is being made. Looking a word up in an existing glossary is not this ritual; any ritual does that in passing.

[[codebase-design]] supplies the words for the shape of the code; this ritual supplies the words for what the code is about. /setup-repo stamps the context layer, and this ritual keeps the language in it precise.

Which "context" this is

WhereWhat it holds
a vault's context/the user's vault, privatethe user: priorities, values, how they work
a repo's CONTEXT.mdthe target repo, committedthe repo's vocabulary: a glossary and nothing else

This ritual writes only the second. It never edits a vault's context/, and it never carries an employer's or client's terms into a personal vault. That is the firewall in templates/vault-AGENTS.md, not a preference.

Files in the target repo

CONTEXT.md            the glossary, at the root; most repos need nothing more
CONTEXT-MAP.md        only when the repo has several bounded contexts
docs/adr/NNNN-*.md    decisions; adr/ instead when docs/ is the source of a published site

Create each file when it first has content: CONTEXT.md when the first term is settled, docs/adr/ when the first decision qualifies. An empty glossary is worse than none, because it looks maintained.

When CONTEXT-MAP.md exists, each context keeps its own CONTEXT.md beside its code (src/billing/CONTEXT.md). Work out which context the current discussion belongs to, and ask when it is not clear.

While the discussion runs

  • Hold the user to the glossary. When they use a term differently from its written definition, point it out at once and ask which meaning is right: "CONTEXT.md says a refund returns the whole payment; you are describing a partial one."
  • Pin down vague words. When a word could mean two things, name both and ask which: "by member do you mean the person or their seat on the plan?" Propose one canonical term.
  • Test boundaries with cases. When two concepts are being related, invent a specific case that falls between them and ask where it belongs.
  • Check claims against the code. When the user says how something works, read the code. If it disagrees, say so: "you said a booking can move between rooms, but Booking.room is set once in the constructor."
  • Write the term the moment it settles. Edit CONTEXT.md in the same turn. Updates saved for the end of the session do not get written.

CONTEXT-FORMAT.md has the entry format and the multi-context layout.

CONTEXT.md holds definitions only. No implementation notes, no specs, no decisions, and no general programming terms, however often the project uses them. A term earns an entry when it means something particular in this domain.

When a decision earns an ADR

Offer one only when the decision is all three:

  1. Costly to undo. Reversing it later takes real work.
  2. Puzzling without the story. Someone reading the code later would ask why it was done this way.
  3. A choice between real options. Something else was viable and lost for a stated reason.

If any one is missing, do not offer it. A cheap decision gets reversed, an obvious one raises no question, and a decision with no alternative has nothing to record.

ADR-FORMAT.md has the layout, the numbering and the kinds of decision that usually qualify. A repo's ADRs are not a vault's decisions/log.md: that log holds the user's personal decisions and stays in the vault.

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.