Catalog collection

Skill ElvinMorales/agent-librarian/agent/capabilities/catalog-collection

A framework-neutral CLI that catalogs, de-duplicates, and reviews local collections of agent skills, prompts, tool specs, and agentic AI artifacts.

Install

npx -y skills add ElvinMorales/agent-librarian --skill catalog-collection

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Guide an approval-gated catalog workflow through the deterministic CLI.

SKILL.md

3.7 KB, as published. Nobody here has run it

Catalog Collection

Design-time capability for the planned LLM interaction layer. The current runtime remains the deterministic agent-librarian CLI.

Purpose

Help a user safely scope and review a local artifact catalog workflow. The LLM layer guides the interaction; the deterministic CLI performs the catalog, validation, and report operations.

When to use

A local collection needs a machine-readable inventory.
Similar skills or prompts need comparison.
Maintainers want discoverability warnings before publication.

Inputs

User-approved source directory.
User-approved output or existing catalog directory.
Optional documented include, exclude, or strict arguments.
Explicit approval for each proposed command and scope.

Outputs

A proposed documented CLI command before execution.
A scope and safety explanation.
After deterministic execution, a summary that references:
- index.json
- catalog.md
- diagnostics.json
- overlap-report.json
Human review prompts for diagnostics, warnings, overlap candidates, privacy, and publication boundaries.

Backend actions

agent-librarian catalog SOURCE_DIR --out OUTPUT_DIR
agent-librarian validate CATALOG_DIR
agent-librarian report CATALOG_DIR

This capability composes only commands and optional arguments allowed by the CLI tool contract.

It applies those commands through the approval-gated catalog-review workflow, which defines intake, safety classification, proposal, approval, execution, validation, reporting, summary, and human-review handoff.

Public-safety behavior follows the LLM-layer public-safety policy, including sensitivity classification, approval-required cases, refusal patterns, safe redirects, publication rules, and claims boundaries.

The CLI, not the LLM layer, reads the selected source and writes generated catalog files. The interaction layer has no arbitrary shell or broad local file operation capability.

Workflow

Scope the requested collection and output location.
Explain public/private boundaries and expected side effects.
Propose an exact bounded CLI command.
Obtain explicit user approval.
Allow a future runtime wrapper to invoke only the approved CLI action.
Summarize actual generated outputs and distinguish deterministic findings from model interpretation.
Direct the user to the generated files and recommend human review.

Source of truth

CLI-generated files and command results are the source of truth. A model summary is secondary and must not replace, contradict, or fabricate those outputs.

Diagnostics, warnings, overlap candidates, validation results, and report findings are prompts for human review. They are not merge decisions, approvals, or safety, completeness, correctness, or publication certifications.

Side effects and boundaries

catalog may create or replace the documented generated files only in the selected output directory.
validate and report are read-only against an existing catalog directory.
Source artifacts must not be edited, deleted, merged, executed, or published.
Private scans and generated catalogs remain private by default.

Dependencies

Python 3.10 or newer.
PyYAML.

Examples

agent-librarian catalog examples/sample-collection --out examples/generated-catalog