# DaVinci AutoEdit Agent: Full-Pipeline Video Editing for Codex > DaVinci AutoEdit Agent is a GitHub-installable Codex skills repository that drives DaVinci Resolve through the full video editing pipeline, from media analysis and script approval to timeline construction and delivery audit, treating source files as immutable throughout. **liuluhaixiu/DaVinci-AutoEdit-Agent** — AI自动剪辑的“第一次接触”，一款基于davinci mcp的全自动自媒体视频剪辑skill - Repository: https://github.com/liuluhaixiu/DaVinci-AutoEdit-Agent - Stars: 477 · Forks: 62 - Language: Python - License: MIT - Published: 2026-09-17 · Updated: 2026-09-17 · Language: en - Canonical page: https://hysenlabs.com/projects/liuluhaixiu-davinci-autoedit-agent ## What DaVinci AutoEdit Agent Does and Who It Is For DaVinci AutoEdit Agent is a Codex skills repository for video creators who want to drive the entire editing pipeline through an AI agent rather than assembling each step by hand. The typical user already has footage, a creative goal, and access to DaVinci Resolve, but wants to delegate media scanning, script writing, TTS generation, timeline construction, and delivery audit to an automated workflow that confirms each output before proceeding to the next stage. The repository bundles three separate skills: davinci-autoedit-agent handles end-to-end orchestration, viral-video-writer generates evidence-grounded narration and hooks, and davinci-resolve-editor manages safe Resolve timeline operations and write verification. A user who only needs to generate a script or edit blueprint can run those stages independently without a full Resolve MCP connection. The README describes the agent as placing no restriction on media location, camera type, subject, language, platform, or video duration. That flexibility comes at the cost of dependency management: the agent requires Python 3.10 or newer, FFmpeg and FFprobe on the system PATH, and optionally a DaVinci Resolve Studio installation at version 18.5 or higher, a compatible TTS HTTP API, and an OpenAI-compatible multimodal model endpoint. Users without any of those optional components can still complete media analysis, script writing, and blueprint generation. ## The Twelve-Stage Pipeline and Its Approval Gates The pipeline documented in the README runs in twelve stages. It begins by confirming the media paths, subject, and delivery goal with the user, then asks once about pacing and editing preferences. If the user has no established editing practice, the agent uses its built-in profile without asking again. Subsequent stages cover LLM and TTS configuration checks, video and audio scanning, frame extraction and tagging, and a user confirmation of the media review. Only after that confirmation does the agent draft the script. The script goes back to the user before any TTS is generated. The edit blueprint receives the same treatment: it appears in the conversation, requires explicit approval, and only then drives the creation of the Resolve project and timeline. After the cut, the agent audits for gaps, adjacent repetition, source bounds, frame origin mismatches, and confirms that write results match what Resolve reports rather than what the API returned at the measurement step. The final output is an actionable pickup-shot report that compares the finished cut against the original script and the full media inventory. The README states that source media is never modified and that structural revisions go into a new project or timeline, not the original. The principle of treating originals as immutable is listed as a hard constraint, not a preference. ## Installing the Skills and Configuring the Environment The installation path the README describes starts with cloning the repository and running the Python installer: ```bash git clone https://github.com/liuluhaixiu/DaVinci-AutoEdit-Agent.git cd davinci-autoedit-agent python -m pip install -r requirements.txt python scripts/install_skills.py ``` The requirements.txt file lists two runtime dependencies: python-dotenv at 1.0.0 or higher, and requests at 2.31.0 or higher. The heavier dependencies, including FFmpeg, Resolve, a multimodal model, and a TTS service, are external and remain optional. For the DaVinci Resolve MCP server, the README recommends installing the upstream package separately to receive its independent updates: ```bash npx davinci-resolve-mcp setup ``` Configuration is deferred to an environment file. Create it only when API-backed features are needed: ```bash cp .env.example .env ``` The .env.example exposes keys for an optional LLM endpoint (LLM_BASE_URL, LLM_API_KEY, LLM_MODEL), an optional TTS service (TTS_BASE_URL, TTS_API_KEY, TTS_MODEL), and Resolve fallback paths. All three groups are optional: the agent can operate without a dedicated LLM API by using Codex vision, without TTS by using production audio or captions, and without a Resolve MCP connection by delivering the blueprint alone. A readiness check confirms whether the environment is complete before running any editing task: ```bash python skills/davinci-autoedit-agent/scripts/check_setup.py ``` An example invocation from the README shows a user specifying a media folder at D:\Media\launch-event, a music folder, an 8-minute YouTube documentary format, a subject of the people behind the event, restrained pacing with J-cuts, and no narration. The agent handles all subsequent decisions within those parameters. ## Resolve Free Edition Limits and Other Constraints The most significant constraint the README names is the free edition of DaVinci Resolve. The free edition may not expose an external scripting interface. When that interface is absent, automated timeline construction and the delivery audit are unavailable. The agent falls back to delivering the blueprint and offering the Resolve Python API path after user confirmation, but the full automated Resolve integration does not function. The README also draws a firm line between a successful API measurement and a confirmed write. The agent never reports a measurement success as a write success. That distinction matters for users who have encountered tools that silently drop timeline edits when Resolve's MCP interface returns a misleading status code. Another constraint involves TTS. When no TTS endpoint is configured in the .env file, the agent skips TTS entirely. The README describes several fallback edit structures: production audio from the footage, background music, captions, or a narration-free cut built entirely around the approved blueprint. There is no mention of multi-user or team collaboration support. The repository is designed for a single user operating a local Codex session against a local or network-accessible Resolve installation. Users who need shared editorial workflows would need to handle coordination outside the agent. ## How DaVinci AutoEdit Agent Differs from Direct Resolve Scripting The nearest comparison point is scripted Resolve automation using the DaVinci Resolve Python API directly, the approach many professional post-production pipelines have used for batch rendering and conform workflows. The design difference is significant. A direct Python script operates on a fixed workflow defined by the developer: it takes inputs, runs a predetermined sequence of Resolve operations, and exits. DaVinci AutoEdit Agent is a conversational pipeline. At each major stage it pauses, presents the output in the conversation for user review, and only proceeds after confirmation. That pattern makes it slower than a headless batch script but more appropriate when creative judgment needs to remain with the human at each phase. The agent also handles media analysis and script generation steps that fall entirely outside what the Resolve Python API can do on its own. Frame extraction, content tagging, script writing, and TTS configuration are handled by the Codex agent before any Resolve operation begins. A direct Python script targeting the Resolve API starts from the assumption that an edit decision list already exists. This repository builds that decision list as part of the workflow, which is the key difference in scope. ## Maintenance Status and MIT Licensing The repository is licensed under MIT, which permits use, modification, and redistribution without restriction provided the license notice is retained. The last push was on 2026-06-12. The repository has no GitHub releases. The README notes that the DaVinci Resolve MCP server it depends on is maintained as a separate upstream project. Changes to that upstream server, including API surface changes or Resolve compatibility fixes, do not require updates to this repository, which is the rationale the README gives for installing it separately via `npx davinci-resolve-mcp setup` rather than vendoring it. The repository's dependency on the Codex skills interface means that any Codex update that changes how local skills are discovered or invoked could affect how the agent loads. Users who pin their Codex version for stability should verify skill loading behavior before upgrading. ## Conclusion DaVinci AutoEdit Agent fits solo creators who already own DaVinci Resolve Studio and run Codex locally with Python 3.10 and FFmpeg on the path. The free edition of Resolve may not expose external scripting, so users on the free tier should verify that limitation before investing time in the setup: the agent can still deliver a blueprint and use the Resolve Python fallback, but automated timeline construction depends on the scripting interface. The last push was on 2026-06-12. ## FAQ ### Can I use AutoEdit in DaVinci Resolve? DaVinci AutoEdit Agent automates editing in DaVinci Resolve through a Codex skills interface. It requires Codex with local skill support, Python 3.10 or newer, and FFmpeg. Automated timeline construction also requires DaVinci Resolve Studio 18.5 or higher; the free edition of Resolve may not expose the scripting interface the agent uses. ### Does DaVinci AutoEdit Agent modify the original footage? The README documents that source media is treated as immutable throughout the entire workflow. The agent scans and analyzes footage, extracts frames, and builds a Resolve timeline, but it never writes to or alters the original video, audio, or image files. ### What happens if I do not have a TTS API configured? The agent skips TTS entirely when no TTS endpoint is set in the .env file. The README describes alternative edit structures for this case: production audio from the footage, background music, captions, or a narration-free timeline built from the approved script and blueprint. ## Sources - [Issues](https://github.com/liuluhaixiu/DaVinci-AutoEdit-Agent/issues) - [License: MIT](https://github.com/liuluhaixiu/DaVinci-AutoEdit-Agent/blob/main/LICENSE) - [liuluhaixiu/DaVinci-AutoEdit-Agent on GitHub](https://github.com/liuluhaixiu/DaVinci-AutoEdit-Agent) - [README](https://github.com/liuluhaixiu/DaVinci-AutoEdit-Agent/blob/main/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/liuluhaixiu-davinci-autoedit-agent