The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from local video projects which could potentially contain adversarial instructions designed to influence the agent.
Ingestion points: Project content, including transcripts, captions, and media metadata, is loaded into the agent's context via the read_project tool as specified in references/editing-workflow.md.
Boundary markers: The instructions do not implement boundary markers or explicit instructions for the agent to ignore commands embedded within the project data.
Capability inventory: The agent possesses significant capabilities to mutate project files using editor tools, specialized generation tools (e.g., image-gen, voice, music), and the review_edit_session tool for final application.
Sanitization: There is no evidence of sanitization or filtering of external content before it is interpolated into the agent's workflow.
[DYNAMIC_EXECUTION]: The skill implements a dynamic capability extension mechanism through the load_skill tool.
According to SKILL.md, invoking load_skill allows the agent to discover and register new tools from the live MCP server at runtime, such as talking-head-guide or shader-gen.
This allows the skill to modify the agent's available toolset based on the connected editor's state and configuration.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for the user or agent to manage its own lifecycle through external repositories.
SKILL.md and references/known-errors.md include commands such as npx skills update openchatcut and npx skills add 0xsline/OpenChatCut to fetch and update the skill's logic.
These references target the vendor's official repository for standard skill maintenance and versioning.

openchatcut