The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external TikTok sources, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Video captions extracted via the caption_extract_url tool (referenced in references/winning-content.md) and product metadata/descriptions fetched via product_search (referenced in references/opportunity-scan.md).
Boundary markers: While references/principles.md establishes clear labeling for the "evidence chain" (e.g., "MCP observation"), the skill does not define technical delimiters to separate this external data from the system prompt or agent instructions.
Capability inventory: The skill allows the agent to perform multiple search and extraction operations across product, creator, video, and shop data using the KOLSprite MCP.
Sanitization: There is no evidence of sanitization or filtering for the content ingested from the external platforms.
[EXTERNAL_DOWNLOADS]: The skill connects to a remote MCP service to perform its core functions.
Evidence: The configuration in agents/openai.yaml points to https://mcp.kolsprite.com/mcp. This is a vendor-owned resource provided by the skill's author to support the intended research capabilities.

tiktok-shop-growth-research