The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses vendor-specific tools (postplus CLI) and standard system utilities (ffmpeg, ffprobe) to process media and extract frames. Instructions strictly forbid the agent from attempting to install missing dependencies or requesting elevated permissions, which reduces the risk of privilege escalation.
[EXTERNAL_DOWNLOADS]: The skill facilitates fetching media from social platforms (TikTok, Instagram, YouTube, etc.) using the CLI's specialized adapters. This design abstracts network operations and limits the agent's direct interaction with raw external data during acquisition.
[INDIRECT_PROMPT_INJECTION]: The skill inherently processes untrusted external media content that may contain embedded adversarial instructions. It manages this risk through explicit instructions that mandate distinguishing visible facts from subjective claims and requiring human review for critical operations.
Ingestion points: External URLs and local file paths passed to the --video and --source parameters within SKILL.md and references/video-analysis.md.
Boundary markers: Documentation requires the agent to explicitly label observations and maintain evidence-backed reports to prevent hallucinated or injected content from being treated as fact.
Capability inventory: The skill is limited to media transformation and analysis CLI commands.
Sanitization: The agent is instructed to use vendor-provided readiness diagnostics (postplus doctor) rather than executing arbitrary repair or installation commands.

media-analysis