The Agent Skills Directory

[SAFE]: No malicious patterns detected. The skill instructions are focused on providing a legitimate way to query repository structure.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses the ~/.deeplake/memory/graph/ directory. This is a hidden path in the user's home directory, but it is the standard location for storage used by the author's (activeloopai) tools. No access to sensitive user credentials or network exfiltration was found.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes AST-derived data from the repository.\n
Ingestion points: Metadata and structural info read from ~/.deeplake/memory/graph/ via cat commands.\n
Boundary markers: None present; the agent treats the files as authoritative structural data.\n
Capability inventory: Uses the Bash and Read tools for file access.\n
Sanitization: No validation is performed on the content of the graph files. While this creates a surface for indirect injection if the codebase being mapped is malicious, the impact is limited to informational structural queries.

hivemind-graph