The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the 'openclip-agent' Python package during the setup phase. The SKILL.md file provides commands to install or upgrade this package using uv, pipx, or pip. While the instructions advise the agent to ask for user consent, this process introduces external code dependencies into the local execution environment.- [DYNAMIC_EXECUTION]: The skill implements a 'self-extending toolbox' system that allows the agent to generate, register, and execute new scripts at runtime. The 'oc-toolsmith' subagent is tasked with authoring new tools, which are registered via 'oc toolbox new' and executed via 'oc toolbox run'. Although the skill describes a verification and auditing process ('oc toolbox promote') intended to gate these tools before they are shared, the capability for an agent to generate and execute its own source code represents a significant dynamic execution risk.- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the automated processing of video transcripts derived from user-provided content.
Ingestion points: Transcripts are generated by the 'oc stt' tool from video audio and are subsequently consumed by subagents such as 'oc-cut-proposer', 'oc-hook-finder', and 'oc-subtitle-agent'.
Boundary markers: No explicit boundary markers or delimiters are defined for the transcript data when it is passed to the worker agents.
Capability inventory: Subagents have access to the 'oc' CLI, which provides extensive capabilities for file system manipulation, video rendering, and script management within the toolbox.
Sanitization: The skill does not describe any sanitization or filtering of the transcript text to prevent instructions embedded in the video's audio from being interpreted as commands by the subagents.- [COMMAND_EXECUTION]: The skill frequently executes shell commands through the 'oc' CLI and 'ffmpeg/ffprobe' to perform video analysis, transcoding, and editing tasks.

oc