memos-memory
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on the
memosCLI tool to perform all operations, executing subcommands such assearch,add,extract,chat, andkbto manage memory and documents. - [PERSISTENCE]: The uninstallation routine described in
references/memos-uninstall.mdincludes a--pathflag that modifies shell configuration files (~/.bash_profileand~/.zshenv) to remove MemOS-managed PATH entries. While this is an environment modification pattern, it is presented here as a cleanup mechanism during a user-initiated uninstallation. - [INDIRECT_PROMPT_INJECTION]: The skill facilitates the capture of user and assistant messages for long-term storage and later retrieval into the agent's context. This creates a surface for indirect prompt injection if malicious instructions are persisted in the memory database.
- Ingestion points: Conversation history captured in
references/memos-add.md. - Boundary markers: None explicitly defined to separate retrieved memory context from instructions.
- Capability inventory: The skill has extensive shell execution capabilities via the
memosCLI across all operational files. - Sanitization: The instructions do not specify any sanitization or validation of the text content before it is stored or after it is retrieved.
Audit Metadata