The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided conversation text to generate structured Notion pages, as described in SKILL.md and demonstrated in evaluation files. While this creates a surface for indirect prompt injection, the risk is mitigated by instructions directing the agent to summarize, clean, and structure the data rather than verbatim transcript processing.
Ingestion points: Raw conversation transcripts and chat history processed at runtime.
Boundary markers: The instructions do not define specific delimiters for separating conversation data from processing instructions.
Capability inventory: The skill uses Notion MCP tools (notion-search, notion-fetch, notion-create-pages, notion-update-page) to manage content in a user's Notion workspace.
Sanitization: The skill relies on natural language instructions for the agent to "Remove conversational filler" and extract key points, acting as a functional filter for embedded instructions.
[COMMAND_EXECUTION]: SKILL.md contains a 'Workflow' section that includes shell commands for setting up and logging into the Notion MCP. These are transparent instructions for the user to perform initial configuration manually and are not executed silently by the agent.
[EXTERNAL_DOWNLOADS]: The configuration in agents/openai.yaml specifies a dependency on the official Notion MCP server (https://mcp.notion.com/mcp). This is a legitimate integration with a well-known service.

notion-knowledge-capture