CLAUDE.md@skills/video/content-engine git:20261001.d049235B

Immutable. This exact content never changes and is served at /api/v1/blob/3f16d810ff76127a.

# Content Engine

Full-stack AI content studio skill for the broomva workspace.

## Structure

```
content-engine/
├── SKILL.md                        # Orchestrator entry point
├── skills/
│   ├── content-engine-dna/         # Visual DNA Compiler
│   ├── content-engine-cinema/      # Cinematic Generation Layer
│   ├── content-engine-autopilot/   # Browser Orchestration
│   └── content-engine-loop/        # Content Loop + Distribution
├── knowledge/
│   ├── raw/                        # Immutable source material
│   ├── compiled/                   # LLM-compiled identity files
│   └── schema.md                   # Compilation rules
├── templates/                      # File templates
├── layout/vertical-9x16.json       # 9:16 layout contract (safe zone, bands, eye line)
├── references/vertical-layout.md   # The contract's rules, source and checker limits
├── scripts/                        # Automation scripts (check_vertical_layout.py = 9:16 gate)
├── remotion/                       # ContentEngineVideo (16:9) + ContentEngineReel (9:16)
├── tests/                          # pytest: layout gate (python -m pytest in this dir)
└── extensions/                     # Plugin directory
```

## Conventions

- **Compiled files** are Markdown with YAML frontmatter and tool-specific prompt sections
- **Raw assets** are never modified by the LLM — they are the source of truth
- **Provenance** — every compiled file traces back to its raw sources
- **9:16 assets pass the layout gate before distribution** — `scripts/check_vertical_layout.py`; the Remotion overlays and the checker both read `layout/vertical-9x16.json`. Change geometry there. Pixel values quoted in docs are copies: `test_skill_docs_quote_only_contract_pixels` fails when one is not a value the contract derives (it cannot tell which zone a number belongs to), and brainrot-for-good's `CAPTION_BAND` is checked against the caption band exactly
- **Tool-specific prompts** — each compiled file contains prompt fragments for Higgsfield models (Soul V2, Nano Banana 2, Veo 3.1, Kling 3.0, Seedance 2.0, Flux 2), Marketing Studio modes, Weavy, etc.

## Dependencies

- Required: ffmpeg, GEMINI_API_KEY (analysis only)
- Tier-1 generation: **higgsfield CLI** (`curl -fsSL https://raw.githubusercontent.com/higgsfield-ai/cli/main/install.sh | sh` then `higgsfield auth login`). Replaces fal.ai + @google/genai as the default path; one CLI, 30+ models.
- Tier-1 fallback: FAL_KEY (when a model Higgsfield doesn't expose is needed)
- Optional: Topaz Gigapixel, ComfyUI, agent-browser
- Compounds: /content-creation, /blog-post, /social-intelligence, /brainrot-for-good, /brand-icons, /higgsfield-generate, /higgsfield-product-photoshoot, /higgsfield-soul-id

## Design Spec

`~/broomva/docs/superpowers/specs/2026-04-07-content-engine-design.md`

## Linear Project

Content Engine project in Broomva workspace: BRO-547 through BRO-575