The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to read sensitive authentication data from the local file path ~/.config/vos/credentials and the VOS_API_KEY environment variable to authorize operations.
[COMMAND_EXECUTION]: The skill relies on executing several shell commands, including the vos CLI and curl for interacting with the platform's API at vos.so.
[INDIRECT_PROMPT_INJECTION]: Untrusted user-provided prompts are interpolated directly into shell command arguments (--prompt) and JSON payloads for API requests without guidance on technical sanitization or shell-escaping.
Ingestion points: User-provided animation prompts described in the workflow of SKILL.md.
Boundary markers: None provided for the construction of shell command strings.
Capability inventory: Execution of subprocesses via the vos CLI and curl within the instruction set.
Sanitization: Instructions advise the agent to omit personal identifiers like names or emails but do not specify requirements for escaping shell metacharacters or validating input against command injection patterns.

vos-authoring