agentleFS
runpodrunpod/runpod-plugins-official/.claude/skills/add-concept/SKILL.md

add-concept

Skill49 starsChanged 48 days agoApache-2.0View on GitHub

Add or change a node in the Runpod concept graph (plugins/runpod/skills/runpod-usage/concepts/*.yaml), including process nodes that explain how something works step by step. Use when asked to add a concept, add a rule to one, document how a Runpod flow works, or fix a node. Covers duplicate checks, choosing the kind, finding public sources, writing the file, validating, and checking the graph. Pairs with the write-guide skill, which keeps the skills and docs linked to the concepts.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 6 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    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. Distinctive:

    Every file is public-facing.

    line 15
  2. Distinctive:

    Search ids, names and aliases before creating anything:

    line 20
  3. Distinctive:

    If the idea is a fact about an existing concept, add a rule to that file instead of a new node.

    line 28
  4. Very specific:

    If it is a new field or state of an existing resource, add it to that file's fields or states.

    line 29
  5. Distinctive:

    Create a new file only for a thing a customer or agent names on its own, or for a flow that spans concepts.

    line 30
  6. Distinctive:

    Write only what a public source supports:

    line 54
  7. Distinctive:

    If no public source supports a claim, leave it out.

    line 63
  8. Very specific:

    Don't use # comments: pnpm format rewrites the file from its data and refuses a file with a comment, since it would drop it.

    line 88
  9. Very specific:

    Every concept must be covered by at least one skill, reference doc or golden path.

    line 118
  10. Very specific:

    Commit the concept file, plus any other files you linked from or to, with a Conventional Commits message that says what the node explains, …

    line 125

≈ “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

Add a concept to the graph

Each node is one YAML file in plugins/runpod/skills/runpod-usage/concepts/ (called concepts/ below). The format, kinds, relation types and writing style are defined in concepts/README.md. Read it before writing; this skill is the procedure, not the format.

Every file is public-facing. Everything in it must be safe to show a customer and backed by a public source. The validator enforces this.

1. Check it doesn't already exist

Search ids, names and aliases before creating anything:

grep -ril "<term>" plugins/runpod/skills/runpod-usage/concepts/
cd ontology/tools && pnpm build:sqlite && pnpm query concept "<term>"   # resolves names and aliases
pnpm query search "<term>"                                     # finds rules that mention it
  • If the idea is a fact about an existing concept, add a rule to that file instead of a new node.
  • If it is a new field or state of an existing resource, add it to that file's fields or states.
  • Create a new file only for a thing a customer or agent names on its own, or for a flow that spans concepts.

2. Pick the kind

If it is…kindExample
Something with its own id and CRUD in the APIresourcepod, template
Part of a resource with no CRUD of its owncomponentworker, container-disk
Read-only reference datacataloggpu-type, data-center
Something you do with a resourcecapabilitypod-ssh-access
Runpod hardware a resource runs oninfrastructuremachine
Charges or creditbillingpod-billing
A cap or quotalimitapi-rate-limit
A Runpod SDK or CLItoolflash
Something Runpod references but doesn't ownexternalcontainer-image
How something works, in order, across conceptsprocesspod-deployment

Set is_a for a kind-of parent (cpu-pod is_a pod). Set part_of for the whole it belongs to (worker part_of serverless-endpoint). Top-level resources, catalogs, billing roots, tools and limits sit part_of: runpod-platform.

3. Find public sources first

Write only what a public source supports:

SourceWhereCite as
REST v2 spectestdata/runpod-migrate/v2-openapi.jsonsource: rest-v2-spec, ref: Schema.field
Public docsthe runpod/docs repo, or docs.runpod.iosource: public-docs, url: https://docs.runpod.io/<path>
Skills in this repoplugins/runpod/skills/source: skill, path: plugins/runpod/skills/...
A read-only request you ranthe live APIsource: live-probe, ref: the request, seen: the date

If no public source supports a claim, leave it out. Don't cite internal documents, internal ids or private repositories. Don't create billable resources to verify a claim without asking first.

4. Write the file

Copy the shape of a similar file (pod.yaml for a resource, machine.yaml for infrastructure, pod-deployment.yaml for a process) and fill in:

  1. id (kebab-case; the file is <id>.yaml), name, kind, summary (one or two customer-facing sentences), product, is_a, part_of, aliases.
  2. surfaces: the REST v2 schema and paths, GraphQL names from the public reference, MCP tools, runpodctl commands, console page.
  3. fields and states if the concept has them. With a REST v2 schema, every field must exist on that schema or its create/update request.
  4. relations, from the closed set in concepts/README.md.
  5. rules: one fact per rule, with an id of <concept-id>.<slug>. Write one to three sentences an agent can act on. Point applies_to at the fields, states or actions the rule constrains. Put related rules on other concepts in see: instead of restating them.

For a process, add steps in order. Each step has an id, title, description, the concepts it involves and the rules that govern it. Steps link to rules that already exist on the concepts. Put only the facts that belong to the flow as a whole on the process itself, such as pod-deployment.stock-is-not-your-machine.

Style: plain present tense, code identifiers in backticks, "Runpod" spelled correctly, and no words like "currently" or "not yet". The full list is in concepts/README.md under "Writing style".

Don't use # comments: pnpm format rewrites the file from its data and refuses a file with a comment, since it would drop it. Put the context in a note on the evidence instead.

5. Validate and look at it

cd ontology/tools
pnpm format              # canonical style; CI runs format:check
pnpm validate --strict   # must report 0 errors and 0 warnings
pnpm scan                # internal links, email addresses and keys; must report 0 errors
pnpm test
pnpm build:graph         # open ../build/graph.html

Common validator errors:

ErrorFix
... is not a conceptA relation, is_a, part_of, field ref or step names a missing id.
applies_to field:x, which ... does not declareAdd the field, or point at one that exists.
mentions an internal ...Restate the rule from a public source, or delete it.
is time-boundSay what holds, and put the date in evidence as seen.
YAML does not parseQuote any item inside [ ] or { } that contains a comma, [], {}, : or #.

In the graph, find the node, check that its links make sense, and read its panel once as a customer would.

6. Link a guide

Every concept must be covered by at least one skill, reference doc or golden path; pnpm check:guides fails otherwise. Follow the write-guide skill (.claude/skills/write-guide/SKILL.md) to extend an existing doc or write a new one, and to check that the docs already covering the concept still agree with its rules.

7. Commit

Commit the concept file, plus any other files you linked from or to, with a Conventional Commits message that says what the node explains, for example docs(concepts): explain why a stopped pod may not restart.

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.