The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including user-provided footage, transcripts, scripts, and style reference links. This creates a surface where malicious instructions embedded in these assets could attempt to influence the agent's behavior during planning or asset generation stages.\n
Ingestion points: The agent reads from 01-transcript/words.json (derived from external sources) and processes reference videos/images in 03-assets/refs/ and scripts during Stage 0 and Stage 2.\n
Boundary markers: The skill uses strong directives ("Never invent a palette", "tokens.css is law") and quality gates (G0-G7) to constrain the agent's output and maintain consistency.\n
Capability inventory: The skill orchestrates file system writes across multiple directories (edit/00-source to edit/07-master) and invokes external execution engines such as Remotion, Manim, and After Effects.\n
Sanitization: There are no explicit instructions for sanitizing or escaping the content of transcripts or scripts before they are used to generate the plan.json or processed by sub-agents.\n- [REMOTE_CODE_EXECUTION]: The instructions for Stage 1 specify the use of npx hyperframes transcribe. Using npx initiates a download and execution of the specified package from the npm registry if it is not already available in the local environment, which introduces a dependency on external code at runtime.\n- [COMMAND_EXECUTION]: The skill frequently invokes shell commands and external tools to perform its tasks, including whisper for transcription and various engines (Remotion, Manim, Blender) for rendering. While these are appropriate for the task, they represent the skill's operational capabilities on the host system.

motion-pipeline