The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute local Ruby utility scripts (summary_skeleton.rb and visual_script_extractor.rb) via the shell to process data and generate file structures. While these scripts are bundled with the skill, invoking shell commands with variable path arguments (<visual_transcript_path>) presents a standard risk surface for command injection if the underlying platform does not sanitize the inputs.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) because it ingests and processes external data (video transcripts) that could contain malicious instructions.
Ingestion points: The agent reads transcript data from JSON files specified by absolute paths at runtime in both SKILL.md and agent_prompt.md.
Boundary markers: There are no boundary markers, XML tags, or explicit 'ignore instructions' warnings used when the transcript content is presented to the agent.
Capability inventory: The agent has significant capabilities, including filesystem writes (via the Edit tool) and the ability to execute shell commands as part of its defined workflow.
Sanitization: The Ruby scripts perform basic JSON parsing but do not sanitize or validate the text content (dialogue and visual descriptions) before it is passed to the LLM for summarization.

summarize-video