The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external video files and provides natural language summaries (e.g., narrative, summary) to the agent. This creates an attack surface where malicious visual or text-based content within a video could be used to influence the agent's subsequent decisions.
Ingestion points: The agent is instructed to use agent-motion timeline and other analysis commands on untrusted recording files as specified in SKILL.md and references/commands.md.
Boundary markers: The skill does not define delimiters or provide instructions for the agent to treat the CLI output as potentially untrusted data.
Capability inventory: The agent has access to Bash, Read, Grep, and Glob tools. The analysis output includes a next_steps field that suggests commands for the agent to execute, creating a multi-step chain that could be influenced by malicious input.
Sanitization: There is no evidence of sanitization or filtering for the natural language summaries generated by the CLI tool. Additionally, analysis flags like --ffmpeg and --ffprobe (in references/commands.md) could potentially be abused if an agent is tricked into pointing them at malicious executables via an injection payload.

agent-motion