The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as YouTube transcripts and comments, when invoking research tools. This creates a potential surface where malicious instructions embedded in external content could influence the agent's behavior during the planning phase.
Ingestion points: External data gathered via creator-stack:research and the YouTube MCP tools (get_video_transcript, get_video_comments) is stored in research.md and then read during the planning workflow.
Boundary markers: The skill does not explicitly define delimiters or instructions to ignore embedded commands within the ingested research data.
Capability inventory: The skill has the capability to write to local markdown files (plan.md) and invoke several other automated skills (creator-stack:title, creator-stack:thumbnail, creator-stack:hook, creator-stack:write).
Sanitization: There is no evidence of specific sanitization or filtering logic applied to the external content before it is processed by downstream skills.
[SAFE]: File system operations are constrained to a specific directory convention (./youtube/episode/), which limits the scope of file access and modification to the project's YouTube-related folders.
[SAFE]: The use of YouTube MCP tools and the delegation of tasks to specialized creator-stack skills are standard practices for this type of orchestration skill and align with its primary purpose.

plan-video