Montaj

What it does

Montaj is an agent-focused video editing toolkit that lets agents run end-to-end editing workflows (probe, trim, waveform-based silence removal, transcription, captions, materialize encodes and background removal) either via a local CLI or an HTTP API. It outputs trim specs for non-destructive edits, supports transcription and caption generation, and includes steps for common edit tasks so an agent can decide the right sequence and parameters based on user intent.

When to use it

Use Montaj whenever the user requests video editing, transcription, captioning, or format conversions for short-to-medium clips. It is ideal for workflows that benefit from automated silence trimming, filler removal, and scripted pipelines (e.g. social clips, short-form captions, explainer videos). Choose HTTP mode when a local serve API is available; fallback to CLI when headless.

What's included

• Scripts: CLI step examples and commands are documented in the skill (probe, trim, cut, materialize_cut, transcribe, caption, normalize, resize). (has_scripts: false in repo metadata but command examples are present in the SKILL.md)
• References: Workflow templates and sub-skill pointers (serve, parallel, select-takes, overlay, write-overlay, etc.).
• Instructions: Detailed core loop, run modes (HTTP/CLI/MCP), step-by-step commands, and deviation rules for when to skip steps like captions or silence removal.

Compatible agents

Best suited for agents with CLI or HTTP integration capability (Copilot/Codex-style or local agent runtimes that can run shell commands and HTTP calls).