Premiere

Operating Adobe Premiere Pro from a command-line agent normally is not possible, because the editor exposes its automation surface through ExtendScript rather than a shell interface. Flue closes that gap by acting as a local shell-to-application bridge: a coding agent pipes ExtendScript into premiere_bridge.py and receives structured JSON back, following the path agent shell -> premiere_bridge.py -> ExtendScript -> JSON result. That contract lets an agent inspect the open project, enumerate sequences, clips, and markers, make bounded edits, and run small ExtendScript steps without any MCP server in between.

Reach for this when a human wants work performed inside Premiere itself rather than on exported files - reviewing the state of a timeline, adjusting elements, or scripting a repetitive task. The recommended workflow begins by reading FLUE.md in the skill directory, which documents how Flue works, the bridge contract, a worked ExtendScript example, and safety expectations. If Flue is already installed, the operator also consults the installed flue SKILL.md and the premiere_adapter APP.md for the operational contract before acting. If Flue is not installed, the agent tells the human and offers the install command pip install flue && flue setup, but never installs without explicit approval in the current session.

The same stdin/stdout contract extends well beyond Premiere: one Flue install also covers Photoshop, After Effects, Illustrator, InDesign, Audition, Blender, Houdini, 3ds Max, Unity, and Microsoft Office through their respective adapters. Because Flue acts inside professional desktop software, the guidance emphasizes treating the human as the driver, preferring small and inspectable steps, and avoiding destructive actions unless explicitly requested. Flue is MIT-licensed and reviewable on GitHub and PyPI before use, so teams can audit the bridge before granting it control of their editor.