The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including video transcripts, draft scripts, and analytics screenshots. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the data provided by the user.\n
Ingestion points: The Procedure section specifies that the skill processes user-provided ideas, transcripts, and analytics screenshots.\n
Boundary markers: The instructions explicitly direct the agent to treat these inputs as untrusted and to ignore any embedded commands, referencing an external SECURITY.md.\n
Capability inventory: The skill allows the agent to execute specific Python scripts (youtube.py, registry-events.py) and write data to the memory/ directory.\n
Sanitization: The skill relies on instructional safety boundaries rather than technical filtering or escaping mechanisms.\n- [COMMAND_EXECUTION]: The skill uses local scripts to perform core functionality such as data retrieval and registry updates.\n
Evidence: SKILL.md directs the agent to use scripts/connectors/youtube.py for YouTube Shorts analytics and to execute registry-events.py with specific operation parameters (propose) to update channel and claim registries.

short-video-scripter