manage-content-collectionslisted
Install: claude install-skill Fractera/Agent-Engineering-Infrastructure
# manage-content-collections
The ONE way to do content CRUD over the site's co-located collections — **deterministic
file operations, NO code generation.** You supply DATA (a content object); the tool writes
the `_data/*` files, copies the thin route skeleton, and regenerates the post list.
Scenario #1 of 8 (4 file-system + 4 database); the file-system set is here.
This skill is **self-sufficient**: plain file operations through one tool. It does NOT
depend on Hermes, on memory, or on any other agent.
## The mental model (read before acting)
- A **group** = a tab/collection (`app/[lang]/<tab>/` — news, blog, documentation). A
**page** = one post in a tab (`app/[lang]/<tab>/<slug>/`). Both are **co-located static
folders**; the post list (`_list.generated.ts`) is auto-derived, never hand-written.
- **🧊 Phase 1: structure is TAKEN, never generated.** A page is a **CLONE of a frozen stub**
(a `compose --samples` post that already has the vetted block structure). You do **NOT**
build the body. You pass only **light metadata** — `slug` + optional `title`/`date`/`tags` —
and the tool clones the stub under the new slug. **Passing a body (`blocks`) is REFUSED.**
Authoring real text into the frozen slots is **Phase 2 (step 155)**, a later step — not here.
- **6 operations = one tool**, discriminated by `operation × target`:
| operation | target | what you pass | what happens |
|---|---|---|---|
| create | group | `tab`, `labels`, `format?`, `languages?`, `sampl