The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The Preamble in SKILL.md instructs the agent to execute telemetry/version_check.py and telemetry/telemetry_init.py. These scripts are not provided in the skill's source files, which means the agent is being told to run unverifiable code that may or may not exist in the host environment.
[COMMAND_EXECUTION]: The skill requires the execution of Python scripts (scripts/audit_edit_boundaries.py and scripts/validate_delivery.py) to process data and validate outputs. This provides a mechanism for local code execution based on the skill's instructions.
[EXTERNAL_DOWNLOADS]: The skill's core functionality involves fetching video content and transcripts from external URLs provided by the user, which involves network communication with third-party services.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content (videos and transcripts), creating a surface for indirect prompt injection attacks where malicious instructions hidden in the data could influence the agent.
Ingestion points: External video URLs, uploaded files, and transcripts processed as described in SKILL.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the ingested transcript data as potentially untrustworthy or to ignore embedded instructions.
Capability inventory: The agent has the ability to execute shell commands (via the provided Python scripts) and perform network requests to fetch video data.
Sanitization: The skill does not mention any sanitization, filtering, or validation steps for the textual content derived from the video transcripts.

video-content-engine