The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates video processing by invoking various external CLI tools. It uses ffprobe for media metadata extraction, python3 for project management via the editor.py tool, and npm/npx to run rendering and QA processes. These operations are essential to the skill's primary function of video editing.
[INDIRECT_PROMPT_INJECTION]: The skill implements a revision loop that fetches reviewer comments from an external canvas URL using read-notes.mjs. This ingested content is then imported into the project state to guide modifications, representing a surface where external instructions could influence agent actions.
Ingestion points: External review canvas URLs processed by read-notes.mjs (referenced in references/project-tools.md).
Boundary markers: The instructions do not specify any delimiters or warnings to isolate ingested review notes from the agent's core instruction set.
Capability inventory: The skill has broad capabilities including file system writes, execution of shell commands (python, npm, ffprobe), and video rendering via hyperframes.
Sanitization: No explicit sanitization, filtering, or validation of the ingested review JSON is described.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx to execute the hyperframes rendering engine, which involves fetching and running packages from the npm registry. It also interacts with external REST APIs such as arcads-broll for asset generation and the video-review-canvas service for delivery and feedback collection.
[DYNAMIC_EXECUTION]: The orchestration workflow involves the runtime generation of storyboards and composition manifests (JSON, HTML, and JavaScript). These dynamically generated files are then interpreted by the hyperframes engine to produce final rendered media.

video-edit-pipeline