agent-memory-discipline
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to automatically query and ingest data from external memory backends (files, MCP servers, or hosted services) before acting on tasks. This behavior creates a surface for indirect prompt injection where malicious instructions stored in the memory could influence the agent's actions.
- Ingestion points: Step 1 in
SKILL.mddefines triggers for memory recall, such as starting project work or choosing tools. - Boundary markers: The skill lacks instructions for the agent to treat recalled content as untrusted or to use specific delimiters to isolate it from the system prompt.
- Capability inventory: The skill is designed to work with any memory tool or MCP server provided by the agent's environment for read/write operations.
- Sanitization: There is no mention of sanitizing or validating the content retrieved from memory before the agent processes it.
- [SAFE]: The skill provides explicit safety instructions to the agent to prevent data exposure. It specifically directs the agent not to save secrets, tokens, passwords, or personal data into memory entries, mitigating the risk of credential exposure in long-term storage (Step 2 and Security & Safety Notes in
SKILL.md).
Audit Metadata