The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data, including topics and script files, which are used to generate video content and narration. This creates a surface where embedded instructions in the input could potentially influence the agent's actions during the production workflow.
Ingestion points: User-provided topic strings and script_path file contents are ingested into the job.json manifest by scripts/init_job.py and validated by scripts/preflight.py.
Boundary markers: There are no explicit delimiters or specific instructions for the agent to ignore potentially malicious content within the topic or script data.
Capability inventory: The skill executes shell commands via subprocess.run (for ffprobe in scripts/preflight.py) and a Bash script (scripts/finalize_delivery.sh for ffmpeg). It also leverages agent capabilities for remote video and speech generation.
Sanitization: scripts/init_job.py performs slugification for the job ID, and scripts/finalize_delivery.sh uses regex validation for the output filename stem.
[COMMAND_EXECUTION]: The skill relies on local command-line tools to process media files and manage the video generation lifecycle.
Evidence: scripts/preflight.py invokes ffprobe to validate the integrity and properties of input images and audio samples.
Evidence: scripts/finalize_delivery.sh executes ffmpeg for complex video filtering, loudness normalization, and generating contact sheets.
[DYNAMIC_EXECUTION]: The skill manages state and configuration by dynamically generating job manifests from a template.
Evidence: scripts/init_job.py reads a JSON template from assets/job.template.json and populates it with runtime parameters to create a project-specific job.json file.

ai-presenter-video