cinematic-skill-creator

# Cinematic Skill Creator v4.6 (Grok 4.6 + Grok Imagine Video 1.5 Native)

You create and validate skills specifically for **Grok Imagine Cinematic Studio v3.10.0** on the **Grok 4.6** Model Layer.

Optimized for:
- `grok-4.6` + `high` — specialist design, DNA, QA, prompt craft
- `grok-4.6` + `xhigh` — multi-skill / suite architecture
- `grok-4.6` + `medium` — quick scaffold / validation
- Native Grok Imagine Video 1.5 (image-to-video, native audio, physics-aware motion, extend-from-frame)

For general-purpose skills outside this repo, use the base `skill-creator`.

## Model Layer (Grok 4.6)

| Task type                    | Preferred model | Effort |
|-----------------------------|-----------------|--------|
| Skill design and complex migration | `grok-4.6` | high |
| Multi-skill / suite architecture | `grok-4.6` | xhigh |
| Quick scaffold / validation  | `grok-4.6` | medium |

**Stack default:** cinematic+Build API/chat **`grok-4.6`**. Opt-in 1M: `grok-4.3`.  
**Registry:** `references/agents/MODEL_LAYER_v4.6.md`

```yaml
model_compatibility:
  - grok-4.6
preferred_model: grok-4.6
reasoning_effort: high
context_window: 500000
knowledge_cutoff: 2026-02-01
```

## When to Activate

- User wants a new cinematic skill (v4.6 Model Layer)
- User asks to validate an existing skill or migrate legacy skills from v4.5 / v9-4p5 to Grok 4.6
- User says `CREATE CINEMATIC SKILL`, `SCAFFOLD SKILL`, `VALIDATE SKILL`, `UPDATE SKILL TO v4.6`, `MIGRATE AGENT TO 4.6`, `CREATE V4.6 AGENT`, or `ADD MODEL LAYER`

## Conventions (Non-Negotiable)

1. **Location** — `<repo-root>/.grok/skills/<kebab-name>/`
2. **SKILL.md** — YAML frontmatter with `name` + `description` (single line, no colon-space, max 1024 chars)
3. **No human docs** — No README/CHANGELOG inside skill dirs
4. **Role Card link** — Production agents must reference `references/agents/<card>.md`
5. **Version** — New skills declare Model Layer **v4.6**. Studio product banner stays **v3.10.0** unless a bump is approved.
6. **Scripts** — Prefer `python3 tools/cinematic_studio_cli.py` over duplicating pipeline logic when that CLI exists.
7. **Model Layer** — Every skill must contain the standard Model Layer table and `model_compatibility` declaration (see template). Do not emit v9-4p5 IDs.

## Workflow

### Step 1 — Scaffold (or Migrate)

```bash
mkdir -p .grok/skills/<skill-name>/{scripts,references}
```

For legacy migration (v4.5 / v9-4p5 → v4.6):
- Replace the Model Layer section using the current template
- Map chat-expert → `high`, multi → `xhigh`, auto → `medium`
- Point to `MODEL_LAYER_v4.6.md`
- Drop dead `tools/models.py · models verify` lines

### Step 2 — Write / Update SKILL.md

Follow `references/skill_template.md` (v4.6 template).  
Keep body focused. Always include:

- Model Layer table (`grok-4.6` + effort)
- `model_compatibility` YAML block
- Grok Imagine Video 1.5 Native section when the skill touches video, audio, or sequence work

### Step 3 — Validate

```bash
bash /root/.grok/skills/skill-creator/scripts/validate-skill.sh .grok/skills/<skill-name>/
```

Also run the local cinematic validator if present:

```bash
bash .grok/skills/cinematic-skill-creator/scripts/validate_skill.sh .grok/skills/<skill-name>/
```

### Step 4 — Register

- Add/update row in `AGENTS.md` if user-facing
- Update `references/agents/AGENT_INDEX.md` if agent-related
- Ensure Role Card (if any) references MODEL_LAYER_v4.6

## Skill Types

| Type            | Example                        | Requires Role Card | v4.6 Notes |
|-----------------|--------------------------------|--------------------|------------|
| Agent skill     | `sequence-director`, `studio-director` | Yes | Model Layer + Video 1.5 protocols |
| Pipeline skill  | `cinematic-sequence-extender`  | No | 1.5 + `grok-4.6` effort |
| Tool skill      | `ai-video-upscaler`, `character-dna-extractor` | No | Add effort guidance where reasoning is involved |
| Meta skill      | `cinematic-skill-creator`      | No | This skill — maintains v4.6 layer |

## Reference Files

- `references/skill_template.md` — v4.6 blank SKILL.md scaffold
- `references/skill_conventions.md` — Naming, structure, Model Layer rules, migration guide
- `references/agents/MODEL_LAYER_v4.6.md` — Canonical model profiles and routing rules

## Quick Commands

- `CREATE CINEMATIC SKILL <name>`
- `SCAFFOLD SKILL <name>`
- `UPDATE SKILL TO v4.6 <path>`
- `MIGRATE AGENT TO 4.6 <path>`
- `ADD MODEL LAYER <skill>`
- `VALIDATE SKILL <path>`

---

*Cinematic Skill Creator v4.6 — grok-4.6 + reasoning_effort · Imagine Video 1.5 Native*