The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied product information and narrative descriptions to generate custom video projects, creating an attack surface for data-embedded instructions.
Ingestion points: User-provided product metadata, feature lists, and narrative beats described in SKILL.md.
Boundary markers: The workflow specifies using AskUserQuestion to confirm the beat list before code assembly, serving as a manual review checkpoint.
Capability inventory: Writing React code to the file system and executing build commands (remotion render) along with utility scripts (make-sfx.sh, verify.sh).
Sanitization: No automated sanitization of user strings is specified, but content is rendered to static video frames in a controlled environment.
[EXTERNAL_DOWNLOADS]: The onboarding guide instructs the agent to clone the skill repository from GitHub and install standard project dependencies.
Evidence: ONBOARDING.md contains git clone commands targeting the author's GitHub repository and npm install for framework dependencies.
[COMMAND_EXECUTION]: The skill executes local scripts and system tools to process assets and generate video files.
Evidence: scripts/make-sfx.sh synthesizes audio using FFmpeg and an inline Python script. scripts/verify.sh captures stills and analyzes audio metadata for quality assurance.
[DYNAMIC_EXECUTION]: The agent is instructed to generate React-based Remotion code at runtime to assemble the project structure and logic based on the user's specific product story.
Evidence: SKILL.md defines a workflow where the agent scaffolds a project and wires <Sequence> and <Composition> components dynamically.

launch-video