The Agent Skills Directory

[DATA_EXPOSURE]: The skill documentation proactively identifies a potential data leakage scenario inherent to the service's architecture. It notes that in shared workspaces, memory queries might return data from other peers if not properly scoped. It provides clear remediation guidance: using unique HONCHO_WORKSPACE_ID values for different user groups to ensure isolation.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a bridge for external data, which introduces a surface for indirect prompt injection if the stored memory contains malicious instructions.
Ingestion points: Data enters the agent's context through query_memory.py and get_context.py via the Honcho API.
Boundary markers: The returned data from get_context and query_memory does not include specific delimiters or 'ignore' instructions for the LLM.
Capability inventory: The skill itself is limited to memory operations and does not possess high-risk capabilities such as arbitrary command execution or file system modification.
Sanitization: The skill does not perform sanitization on the content retrieved from the external memory store.
[CREDENTIALS_UNSAFE]: The skill handles API keys through standard and safe practices, recommending the use of .env files and python-dotenv rather than hardcoding credentials. The documentation uses appropriate placeholders.

honcho-memory