The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local Ruby scripts (process_footage.rb, library.rb, script_extractor.rb) and system commands (tail, caffeinate) to manage the video processing lifecycle. These commands use variables for library names and filenames to identify target assets.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes audio transcripts derived from external video files. This introduces a surface for indirect prompt injection where spoken content could theoretically influence the agent's behavior, though the skill employs several mitigations.
Ingestion points: Audio transcript JSON files (e.g., libraries/<library>/transcripts/) in SKILL.md and agent_prompt.md.
Boundary markers: The sub-agent prompt defines specific markdown headers (## Overview, ## Key Visuals, ## Notable Dialogue, ## B-Roll) to constrain the output format and reduce the likelihood of the agent following instructions embedded in the transcripts.
Capability inventory: The system includes capabilities for file writing (summaries), metadata updates to library configurations, and subprocess execution for video processing tools.
Sanitization: The script_extractor.rb utility is used to pre-process raw transcripts into clean text segments before they are presented to the Sonnet analysis agent, removing timestamps and non-dialogue metadata.

analyze-video