alterlab-ieu/alterlab-fc-skills

1. This repository is a prompt library, not an application runtime. Its core product is a set of 72 standalone Markdown skill files that instruct Claude how to act as domain-specific autonomous agents.

2. The top-level structure is intentionally simple: README.md explains the catalog, CLAUDE.md defines repository-wide instructions and required format rules, package.json and scripts/validate.js support validation, and the real content lives under skills/.

3. The skills/ tree is the main subsystem boundary. It is organized into six departments, each with 12 skills:

- pra — PR & Advertising

- cdm — Cinema & Digital Media

- nmc — New Media & Communication

- genai — Generative AI Production

- vcd — Visual Communication Design

- rma — Research Methods & Academic Writing

4. Each individual skill is isolated in its own folder, and the only required artifact inside is SKILL.md. That makes every skill a self-contained unit that can be loaded independently into Claude Projects or Claude Code.

5. The content model is consistent across the repo: each skill is a structured prompt that defines role, workflow, frameworks, and output expectations. The README describes the common agentic cycle: research → context reading → creation → self-review → presentation.

6. The architecture is therefore catalog + pack + atomic skill:

- catalog layer: README indexes the system

- department layer: groups skills by communication discipline

- skill layer: one prompt file per use case

7. CLAUDE.md is an important governance boundary. The README explicitly says all skills follow the AlterLab NEXUS skill architecture, and CLAUDE.md contains the mandatory template and formatting rules that contributors must follow when adding or improving skills.

8. Validation is handled as a lightweight quality-control pipeline. The scripts/validate.js script is the repo’s executable check, and .github/workflows/validate.yml runs that validation in CI so broken structure or malformed skill files can be caught before merge.

9. GitHub is used as the distribution and contribution surface rather than as an app backend. Issue templates and a pull request template support skill requests and improvements, while the workflow enforces consistency across contributions.

10. The README positions the skills as usable in two main ways: through Claude Projects by uploading SKILL.md files, or through Claude Code CLI by pointing Claude at the repository. So the repo’s output boundary is external: Claude consumes these files as instructions.

11. There are no databases, services, or compiled artifacts in the repository. The architecture is documentation-driven and file-based, with the main data flow being: contributor edits SKILL.md → validation checks the repository rules → Claude loads the skill file as an agent prompt.

12. For an architecture diagram, the key nodes are: repository docs and governance, department folders, atomic skill files, validation script, and GitHub Actions CI. The arrows are mostly from contributors into skills, validation over the skill tree, and Claude consuming the Markdown files as executable instruction sets.