SoraSAFE

CLI tool for configuring and monitoring Claude Code

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: "sora"
description: "Use when the user asks to generate, remix, poll, list, download, or delete Sora videos via OpenAI\u2019s video API using the bundled CLI (`scripts/sora.py`), including requests like \u201cgenerate AI video,\u201d \u201cSora,\u201d \u201cvideo remix,\u201d \u201cdownload video/thumbnail/spritesheet,\u201d and batch video generation; requires `OPENAI_API_KEY` and Sora API access."
author: openai
---


# Sora Video Generation Skill

Creates or manages short video clips for the current project (product demos, marketing spots, cinematic shots, UI mocks). Defaults to `sora-2` and a structured prompt augmentation workflow, and prefers the bundled CLI for deterministic runs. Note: `$sora` is a skill tag in prompts, not a shell command.

## When to use
- Generate a new video clip from a prompt
- Remix an existing video by ID
- Poll status, list jobs, or download assets (video/thumbnail/spritesheet)
- Batch runs (many prompts or variants)

## Decision tree (create vs remix vs status/download vs batch)
- If the user has a **video id** and wants a change → **remix**
- If the user has a **video id** and wants status or assets → **status/poll/download**
- If the user needs many prompts/assets → **create-batch**
- If the user asks for two versions with a small change (same shot, different subject/detail) → **create** the base, then **remix** for the variation
- Otherwise → **create** (or **create-and-poll** if they need a ready asset in one step)

## Workflow
1. Decide intent: create vs remix vs status/download vs batch.
2. Collect inputs: prompt, model, size, seconds, and any input reference image.
3. If batch: write a temporary JSONL under tmp/ (one job per line), run once, then delete the JSONL.
4. Prefer CLI augmentation flags (`--use-case`, `--scene`, `--camera`, etc.) instead of pre-writing a structured prompt. If you already produced a structured prompt file, pass `--no-augment` to avoid double-wrapping.
5. Run the bundled CLI (`scripts/sora.py`) with sensible defaults (see references/cli.md). For long prompts, prefer `--prompt-file` to avoid shell-escaping issues; pair with `--no-augment` if the prompt is already structured.
6. For async jobs, poll until complete (or use create-and-poll).
7. Download assets (video/thumbnail/spritesheet) and save locally.
8. Remove intermediate files created during invocation (for example `prompt.txt`, `remix_job.json`, temp JSONL). If the sandbox blocks `rm`, skip cleanup or truncate the files without surfacing an error.
9. Iterate with a single targeted change per prompt.

## Authentication
- `OPENAI_API_KEY` must be set for live API calls.

If the key is missing, give the user these steps:
1. Create an API key in the OpenAI platform UI: https://platform.openai.com/api-keys
2. Set `OPENAI_API_KEY` as an environment variable in their system.
3. Offer to guide them through setting the environment variable for their OS/shell if needed.
- Never ask the user to paste the full key in chat. Ask them to set it locally and confirm when ready.

## Defaults & rules
- Default model: `sora-2` (use `sora-2-pro` for higher fidelity).
- Default size: `1280x720`.
- Default seconds: `4` (allowed: "4", "8", "12" as strings).
- Always set size and seconds via API params; prose will not change them.
- Use the OpenAI Python SDK (`openai` package); do not use raw HTTP.
- Require `OPENAI_API_KEY` before any live API call.
- If uv cache permissions fail, set `UV_CACHE_DIR=/tmp/uv-cache`.
- Input reference images must be jpg/png/webp and should match target size.
- Download URLs expire after about 1 hour; copy assets to your own storage.
- Prefer the bundled CLI and **never modify** `scripts/sora.py` unless the user asks.
- Sora can generate audio; if a user requests voiceover/audio, specify it explicitly in the `Audio:` and `Dialogue:` lines and keep it short.

## API limitations
- Models are limited to `sora-2` and `sora-2-pro`.
- API access to Sora models requires an organization-verified account.
- Duration is limited to 4/8/12 seconds and must be set via the `seconds` parameter.
- The API expects `seconds` as a string enum ("4", "8", "12").
- Output sizes are limited by model (see `references/video-api.md` for the supported sizes).
- Video creation is async; you must poll for completion before downloading.
- Rate limits apply by usage tier (do not list specific limits).
- Content restrictions are enforced by the API (see Guardrails below).

## Guardrails (must enforce)
- Only content suitable for audiences under 18.
- No copyrighted characters or copyrighted music.
- No real people (including public figures).
- Input images with human faces are rejected.

## Prompt augmentation
Reformat prompts into a structured, production-oriented spec. Only make implicit details explicit; do not invent new creative requirements.

Template (include only relevant lines):
```
Use case: <where the clip will be used>
Primary request: <user's main prompt>
Scene/background: <location, time of day, atmosphere>
Subject: <main subject>
Action: <single clear action>
Camera: <shot type, angle, motion>
Lighting/mood: <lighting + mood>
Color palette: <3-5 color anchors>
Style/format: <film/animation/format cues>
Timing/beats: <counts or beats>
Audio: <ambient cue / music / voiceover if requested>
Text (verbatim): "<exact text>"
Dialogue:
<dialogue>
- Speaker: "Short line."
</dialogue>
Constraints: <must keep/must avoid>
Avoid: <negative constraints>
```

Augmentation rules:
- Keep it short; add only details the user already implied or provided elsewhere.
- For remixes, explicitly list invariants ("same shot, change only X").
- If any critical detail is missing and blocks success, ask a question; otherwise proceed.
- If you pass a structured prompt file to the CLI, add `--no-augment` to avoid the tool re-wrapping it.

## Examples

### Generation example (single shot)
```
Use case: product teaser
Primary request: a close-up of a matte black camera on a pedestal
Action: slow 30-de

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.