The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of an automation that will research topics and publish results by interpolating user-provided inputs into a prompt template. \n- Ingestion points: SKILL.md (Step 3) collects research topic, keywords, and quality rules from the user. \n- Boundary markers: The setup instructions lack requirements for delimiters or warnings to prevent the model from executing instructions embedded in the research topic. \n- Capability inventory: The automation leverages Tavily MCP for searching and Notion MCP for writing, as defined in SKILL.md (Step 1). \n- Sanitization: No input validation or escaping is applied to user-provided research parameters before prompt construction. \n- [DATA_EXFILTRATION]: The skill transmits task configuration to an external endpoint at ${OPENHANDS_HOST}. \n- The target URL and API key are sourced from the platform's runtime services. \n- Since the skill is authored by the platform vendor 'openhands', this communication with the vendor's automation infrastructure is a standard operation and does not represent an unauthorized exfiltration attempt.

research-brief