agentleFS
noltenolte/claude-shared/skills/backstage-catalog-generate/SKILL.md

backstage-catalog-generate

Skill0 starsChanged 3 months agoMITView on GitHub

Generates a schema-valid Backstage catalog-info.yaml from an existing software project, conforming to spec/project/backstage-catalog-generation/. Inspects the repo (language/structure, remote slug, CODEOWNERS, OpenAPI/AsyncAPI/GraphQL/gRPC specs, colocated docs), infers the per-kind MUST-floor fields, confirms what it cannot infer (owner, spec.system, dependsOn, lifecycle) with the operator, writes the descriptor at the repo root, and self-validates. Invoke when the user asks to \"generate a catalog-info.yaml\", \"onboard this repo into Backstage\", \"create a Backstage Software Catalog entity\", \"add this service to our developer portal\", or German equivalents (\"erzeuge eine catalog-info.yaml\", \"nimm dieses Repo in Backstage auf\"). Optionally emits a Tech Radar TechRadarLoaderResponse JSON (never a catalog entity). Don't use to configure a Backstage backend / catalog.providers, install the Tech Radar UI plugin, or ingest Group/User org entities. Single-shot; resume not applicable.

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    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:

    Inspect: Read the repo signals above.

    line 51
  2. Very specific:

    Emit: Write catalog-info.yaml at the repository root (matching the discovery default path).

    line 54
  3. Very specific:

    Self-validate: Run the offline @roadiehq/backstage-entity-validator when available (npx @roadiehq/backstage-entity-validator …

    line 55
  4. Very specific:

    When spec/project/backstage-catalog-generation/ disagrees with this body, the spec wins.

    line 84

≈ “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

Backstage Catalog Generate

Turns an existing software project into a schema-valid Backstage catalog-info.yaml at the repository root — inferring what the repo reveals, asking the operator for what it does not, and self-validating before it presents the result. The authoritative rules live in spec/project/backstage-catalog-generation/; this skill operationalises that spec. When the spec and this body disagree, the spec wins.

Why this is a skill, not an agent

  • Operator-invoked, conversational. Reached as /nolte-shared:backstage-catalog-generate against the current repo; the operator drives it directly.
  • Mid-flow confirmation is the contract. Owner resolution, spec.system, dependsOn, and an uncertain lifecycle cannot be inferred safely from one repo; the skill surfaces each as a confirmation gate rather than guessing. An agent's fire-and-forget shape would lose those gates.
  • The result flows back into the conversation. The written descriptor path and the list of inferred-vs-confirmed values land in the operator's context for the next step (committing, opening the register-component PR).
  • Counter-dimension: the repo inspection could run as an isolated reader, but the load-bearing dimensions — operator invocation and the per-field confirmation gates — make this a skill.

German trigger phrases

  • „erzeuge eine catalog-info.yaml", „nimm dieses Repo in Backstage auf", „erzeuge einen Backstage-Catalog-Eintrag für dieses Projekt", „füge diesen Service unserem Developer-Portal hinzu", „generiere ein Tech-Radar-JSON"

Repo signals → fields

The skill reads these signals and maps them to descriptor fields (every inferred value is recorded as inferred, distinct from confirm):

SignalFieldNotes
repo / project namemetadata.nameslugify, strip leading/trailing non-alphanumerics, cap 63 → must satisfy isValidObjectName
primary language / structurespec.typeservice / website / library (Component); convention, not a fixed enum
git remote sluggithub.com/project-slug, backstage.io/source-locationorg/repo; source-location is url:…/<repo>/ with trailing slash
CODEOWNERS / team slugspec.ownerbare slug → Group; an individual needs an explicit user: prefix
openapi.yaml / asyncapi.yaml / *.graphql / *.protoan API entity + spec.providesApisAPI spec.definition via $text: ./<relative-path>, spec.type = openapi/asyncapi/graphql/grpc
colocated docs/ / mkdocs.ymlbackstage.io/techdocs-ref: dir:.only when docs are colocated
README description / package.json etc.metadata.description, metadata.tagstags match ^[a-z0-9:+#]+(\-[a-z0-9:+#]+)*$

spec.lifecycle, spec.system, spec.dependsOn, and spec.domain are never inferred silently — default lifecycle only on a justifying signal, otherwise confirm; system/dependsOn/domain always require operator input.

Operations

1. Generate a catalog-info.yaml (default)

  1. Inspect. Read the repo signals above. Determine the entity kind(s) — almost always a single Component, plus an API entity when an interface definition is found. Consult the per-kind tables in spec/project/backstage-catalog-generation/ §"Entity kinds" for the required/optional fields of any other kind.
  2. Infer. Fill the MUST-floor for each kind. For a Component that is: apiVersion: backstage.io/v1alpha1, kind: Component, a valid metadata.name, spec.type, spec.lifecycle, spec.owner. Mark each value inferred or needs-confirm.
  3. Confirm. Present the inferred descriptor and the open questions (owner resolution, lifecycle, optional system/dependsOn) to the operator in one pass. Never emit a guessed system/dependsOn/domain; never emit an owner that cannot be confirmed to resolve — flag it as operator-action-required instead.
  4. Emit. Write catalog-info.yaml at the repository root (matching the discovery default path). Use multi-entity --- separators when emitting a Component plus its API. Quote any numeric/boolean-looking annotation value so it stays a string.
  5. Self-validate. Run the offline @roadiehq/backstage-entity-validator when available (npx @roadiehq/backstage-entity-validator catalog-info.yaml); otherwise apply the §Hard rules checklist below as a deterministic pre-flight. Treat reference-target existence (owner/system/API) as unverified by the offline check — surface those references as claims to confirm, not validated facts.
  6. Report. Tell the operator the written path, the inferred-vs-confirmed split, and any references still requiring an existing Group/User/System in the target catalog.

2. Generate a Tech Radar JSON (optional secondary)

When the operator asks for a Tech Radar output: emit a TechRadarLoaderResponse JSON (quadrants, rings, entries) per spec/project/backstage-catalog-generation/ §"The Tech Radar data model" — not a catalog-info.yaml, and never modelled as a catalog entity. Express each entry's ring placement through its timeline (a snapshot with ringId and a coercible date), not as a direct field. Target the @backstage-community package model; never reference the deprecated @backstage Tech Radar package or the dead backstage.io/docs/features/techradar/ URLs.

Gotchas

Backstage's real validation is stricter and less obvious than its prose docs — these are the quirks that bite a generator:

  • The offline validator does not check reference-target existence. @roadiehq/backstage-entity-validator validates schema and field-format only; an owner/system/API reference can pass and still dangle at processing time. Always surface references as claims to confirm against the target catalog, never as validated facts.
  • Numeric/boolean-looking annotation values must be YAML-quoted. Annotation values are checked as typeof string; an unquoted github.com/user-id: 123456 or backstage.io/orphan: true parses as a number/bool and fails. Quote them.
  • metadata.name is stricter than "alphanumerics separated by [-_.]". The first and last character must be alphanumeric — a slug that ends in - (common after naive slugification) is rejected by FieldFormatEntityPolicy at processing time, after it passed coarse ingestion.
  • Resource has no lifecycle even though Component and API do — emitting it is invalid. Symmetrically, Group.children and User.memberOf must be present even when empty ([]).
  • There is no official backstage-cli validate command. Validation is either the @roadiehq validator or a POST to a running backend's /api/catalog/validate-entity (where location goes in the JSON body, not an HTTP header).
  • The Tech Radar moved to @backstage-community. The old @backstage Tech Radar package and the backstage.io/docs/features/techradar/ URLs are dead — never cite them.

Hard rules

  • Never author auto-set or output-only fields: backstage.io/managed-by-location, backstage.io/managed-by-origin-location, backstage.io/orphan, metadata.uid, metadata.etag, relations, status.
  • Never emit spec.lifecycle on Resource, System, Domain, Group, or User — only Component and API carry it.
  • Always include empty-but-required keys: spec.children: [] on a Group with no known children; spec.memberOf: [] on a User with no known memberships.
  • API entities must carry a non-empty spec.definition, supplied via a $text: placeholder, and the providing Component lists the API under providesApis.
  • metadata.name must satisfy isValidObjectName: length 1–63, first and last character alphanumeric, separators [-_.] interior only. metadata.namespace is a DNS label (lowercase, no underscore/dot).
  • Owner disambiguation: a bare owner reference resolves to a Group; emit an explicit user:-prefixed reference for an individual. Prefer the full kind:namespace/name form for cross-system robustness.
  • Never use a deprecated annotation: backstage.io/github-actions-id → github.com/project-slug; backstage.io/definition-at-location → placeholder substitution; jenkins.io/github-folder → jenkins.io/job-full-name.
  • Never branch on the old-vs-new frontend system — catalog-info.yaml is a backend, frontend-agnostic concern.
  • Distinguish inferred from confirmed in every output; never silently emit a value the operator must own.
  • When spec/project/backstage-catalog-generation/ disagrees with this body, the spec wins; propose updating this skill rather than diverging.

Resumability

This skill is deliberately not resumable: although it walks six steps and one confirmation gate, they form a single cheap-to-restart pass — no externally-visible mutation lands before the final Emit, and every inferred field is re-derived deterministically from the same repo signals, so an interruption is recovered by re-invoking rather than by resuming persisted state (spec/claude/resumable-work/ reserves resume for multi-gate side-effecting flows).

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.