The Agent Skills Directory

[PROMPT_INJECTION]: The skill reads content from external markdown files (script.md and edit-notes.md) to identify clips and generate captions, which creates an attack surface for indirect prompt injection.
Ingestion points: The agent context is populated with data from workspace/video-projects/<slug>/script.md and workspace/video-projects/<slug>/edit-notes.md.
Boundary markers: The instructions lack boundary markers or explicit directives to ignore instructions that might be embedded within the project scripts or edit notes.
Capability inventory: The skill is granted Bash and Write tool permissions, allowing the agent to execute the rendering script and modify files in the workspace.
Sanitization: There is no evidence of sanitization, filtering, or validation of the text ingested from the script files before it is used to generate command arguments or rendered captions.
[COMMAND_EXECUTION]: The skill invokes a local script using the Bash tool to process video data.
Evidence: The skill executes the command helpers/render.py --edl <short-N.edl.json> --vertical -o workspace/exports/<slug>/<slug>-short-<N>.mp4 in the Process section of SKILL.md.

shorts-producer