The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data through DaVinci Resolve interchange formats and project files.
Ingestion points: Processes .drt, .drp, .otio, .edl, AAF, and .xml files as described in SKILL.md.
Boundary markers: No explicit instruction-level boundary markers are defined for the content of ingested files to prevent the agent from following instructions embedded in metadata.
Capability inventory: Possesses capabilities including file system writes, project database patching, and media relinking.
Sanitization: Mentions a 'sanitize' step for XML imports to preserve link integrity, but does not specify validation against malicious instructions.
[COMMAND_EXECUTION]: The skill invokes external system tools and specialized utilities to perform its tasks.
Evidence: Mentions execution of ffmpeg for frame comparisons and the drt utility for handling project version management.
[DYNAMIC_EXECUTION]: The skill performs direct manipulation of application databases and uses runtime introspection to manage API compatibility.
Evidence: Performs direct patching of the DaVinci Resolve project_db when the application is closed and uses Python's dir() function to probe API capabilities at runtime.

resolve-conform