The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes vendor-provided Ruby scripts, including combine_visual_transcripts.rb and export_to_fcpxml.rb, to concatenate transcript data and export editorial sequences. These scripts are run via bundle exec and system shell commands.
[PROMPT_INJECTION]: An indirect prompt injection surface exists as the agent processes visual transcripts to make editorial decisions.
Ingestion points: Combined visual transcript JSON files and library.yaml metadata.
Boundary markers: No specific delimiters or instruction-ignore warnings are used when reading transcript content.
Capability inventory: Shell script execution, file system management, and subagent tasking via the Task tool.
Sanitization: No explicit sanitization of transcript text is performed before it is analyzed by the agent.
[SAFE]: The implementation follows security best practices by using YAML.load_file with a restricted set of permitted classes, effectively mitigating risks associated with unsafe data deserialization.

roughcut