The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project learnings retrieved from an external InfraNodus graph. while this ingestion point represents a surface for indirect prompt injection, the risk is mitigated by explicit instructions to verify learnings against current project code and documentation. Ingestion points: Data is retrieved from the get_project_learnings MCP tool. Boundary markers: The skill requires manual confirmation for data updates and provides specific admission criteria for insights. Capability inventory: The skill is limited to graph-based knowledge retrieval and storage; it lacks tools for direct filesystem modification or arbitrary command execution. Sanitization: Both the agent instructions and the backend server are designed to filter out secrets and sensitive credentials.
[DATA_EXFILTRATION]: The skill transmits project architectural metadata (such as file paths and module relationships) to a vendor-controlled service. This is the intended purpose of the skill and is strictly governed by user consent: the graph must be explicitly enabled per project, and every proposed batch of learnings must be reviewed and approved by the user before transmission. All traffic is directed to official vendor endpoints (mcp.infranodus.com).

project-learnings