Rawdestiny Labs Contact / Inquiries

Rawdestiny AI Research

From generated output
to reviewed work.

Claude-assisted narrative tools and open production QA. We build the evidence layer that helps a small studio examine a real film, question a reviewer and check a repair.

Install OpenFilmQA Read the architecture
Open source / v0.2

OpenFilmQA

A local CLI and portable review skill for Claude, Codex and other agents. Prepare review packets, record evidence-linked observations, compare versions and evaluate a reviewer against private human labels.

Source code on GitHub

Install the Claude skill

npx --yes --package=github:BaoPham9x/OpenFilmQA openfilmqa install --claude

Install for other skill agents

npx --yes --package=github:BaoPham9x/OpenFilmQA openfilmqa install

Installs directly from GitHub using npm’s npx. npm registry publication is pending. Installation does not call a model or overwrite an existing skill.

Nine stages.
One evidence trail.

The workflow covers the whole film. The tool records what a reviewer actually inspected, so unsupported motion or sound cannot silently become a pass.

  1. Lock source & scope

    Identify the exact film, current references, storyboard and approved plan.

  2. Read the director’s intent

    Trace essential action, props, camera geography and the feeling the scene should leave.

  3. Inspect actual frames

    Check identity, outfit, set, scale and visual faults against the right references.

  4. Watch action & joins

    Review continuous motion and neighboring shots for contact, completion and continuity.

  5. Watch & listen

    Check story clarity, pacing, performance, sound and emotional landing.

  6. Adjudicate each finding

    Have another reviewer or the owner confirm or reject the claim from actual media.

  7. Repair & recheck

    Compare exact versions and adjacent shots, including new damage from a repair.

  8. Check delivery

    Decode the export and verify its specification. Keep technical and creative approval separate.

  9. Evaluate the reviewer

    Measure missed faults, false alarms and harmful repairs against private human-labelled cases.

Claude today.
Claude API next.

We use Claude to plan production, build tools and reason through proposed faults against scene specifications. The portable skill gives Claude a reproducible review workflow.

Open reviewer contract

Attach a compatible CLI reviewer through a documented JSON request/response contract. The supplied Codex and Antigravity CLI adapters currently inspect images. Continuous motion and sound remain unreviewed unless a capable tool actually watches and listens.

Planned Claude API reviewer

Our next integration will connect prepared evidence to Claude’s API and validate structured findings. We will measure false alarms, missed faults and repair quality against independently labelled cases before claiming reliable automatic judgment.

The v0.2 engine currently makes no paid model calls. Reviewer execution is opt-in and uses the maker’s authorized account.

Read the adapter contract

Real production
contexts.

Little Bao House and Sir Bao give our narrative workflows actual characters, action, cut continuity and sound to review.

Explore Little Bao House Explore Sir Bao