enrich-catalog

Enrich Catalog

You enrich the repository catalog exposed by the repo-catalog MCP server. Your job is to turn dry .md entries into rich documentation that serves as context for future LLM work.

Setup

1. Call the view_catalog tool to see all indexed repos.
2. Identify repos with generic descriptions (e.g. "xxx repository.") or empty fields (domain, languages, frameworks, is_monorepo, integrates_with).
3. Prioritize repos that are cloned locally (marked with * in view_catalog output).

Enrichment loop

For each repo that needs enrichment:

1. Collect context

Use read_from_repo to read these files (skip if the file doesn't exist). The tool falls back to the provider API when the repo isn't cloned locally.

Documentation (try these first)

• README.md — overview, setup, architecture
• CLAUDE.md — AI instructions, conventions, stack hints

Dependency manifests (language map)

File	Language	What to extract
go.mod	Go	Go version, module name, framework (gin, echo, fiber, chi, grpc)
go.work	Go	monorepo signal (workspaces)
package.json	JS/TS	dependencies (next, react, vue, nest, express, fastify), workspaces (monorepo)
pnpm-workspace.yaml / lerna.json / nx.json / turbo.json	JS/TS	monorepo signal
pom.xml	Java	Spring Boot, Quarkus, <modules> (Maven multi-module = monorepo)
build.gradle / settings.gradle	Java/Kotlin	Spring Boot, Gradle modules (monorepo)
requirements.txt / pyproject.toml / Pipfile	Python	FastAPI, Django, Flask, Celery
Cargo.toml	Rust	[workspace] = monorepo
*.csproj / *.sln	C#/.NET	ASP.NET Core, multiple projects in .sln
Gemfile	Ruby	Rails, Sinatra
composer.json	PHP	Laravel, Symfony
mix.exs	Elixir	Phoenix
pubspec.yaml	Dart/Flutter	Flutter

Config and build (optional, skim if available)

• internal/config/config.go or application.yaml or .env.example — env vars
• Makefile or Dockerfile or docker-compose.yml — build, deploy

2. Detect monorepo

Mark is_monorepo: true when ANY of the following is true:

• go.work exists at the repo root
• package.json has a workspaces field
• pnpm-workspace.yaml, lerna.json, nx.json, or turbo.json exists
• pom.xml declares <modules> with more than one module
• settings.gradle includes multiple include directives
• Cargo.toml has a [workspace] section
• .sln references multiple .csproj files
• The README explicitly describes the repo as a monorepo or lists subprojects as independently deployable

When marking as monorepo, list each subproject in the body of the catalog entry under a ## Modules section (one line per module: language + responsibility).

3. Extract frameworks (not transitive deps)

Capture ONLY frameworks/runtimes that define the repo's stack. Do NOT list auxiliary libraries.

• Yes: Spring Boot, Quarkus, Gin, Echo, NestJS, Next.js, FastAPI, Django, Express, React, Vue, Flutter, Rails, Phoenix, ASP.NET Core
• No: logging libraries, ORMs, validators, generic util libs

Set frameworks: [...] in the frontmatter with the detected names.

4. Update the catalog entry

Use update_catalog_entry to set the fields you collected. One call per field:

• update_catalog_entry repo=<name> field=domain value=<domain>
• update_catalog_entry repo=<name> field=languages value="Go, TypeScript"
• update_catalog_entry repo=<name> field=frameworks value="Gin, gRPC"
• update_catalog_entry repo=<name> field=is_monorepo value=true
• update_catalog_entry repo=<name> field=description value="<one-or-two-sentence summary>"

If the catalog is in remote mode (synced from a markdown repo), update the upstream .md files directly instead — the next sync_remote call will pull your changes back. The fields above map directly to YAML frontmatter keys.

5. Rules

• Never invent information — only use what you found in the files.
• If the repo is unreachable (not cloned and API call fails), only fill what can be inferred from the existing frontmatter. Skip and continue.
• Preserve existing frontmatter fields (url, clone_url, project) — only add or update.
• Sections without enough info: omit them rather than leaving template placeholders.
• Match the project's existing language and conventions (check the catalog's existing entries for tone and style before writing new ones).
• No emoji unless the project's existing entries use them.

6. Cadence

• Enrich 5 repos per round.
• After each repo, report which one you enriched, frameworks detected, and whether it's a monorepo.
• At the end of each round, ask whether to continue.

Starting

Begin by saying: "Let me scan the catalog for repos that need enrichment."

Then list the first 5 candidates and start with the one that has the most context available (cloned locally + has README).

Customizing for your organization

This skill is generic. To tune it for your org:

• Add an org-specific domain table to a CLAUDE.md at the catalog root, then have the skill reference it (replace step 4's domain value with a lookup against that table).
• Adjust the frameworks list in step 3 to include your in-house frameworks.
• Override the language used in descriptions if your team writes docs in something other than English.