memory

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Uses Bash and awk scripts to extract sections from task management files and perform file operations like backups and template-based generation.\n- [DATA_EXFILTRATION]: Accesses the ~/.claude-mem/settings.json configuration file to identify and integrate with local memory systems. This path is outside the project root and is used to detect the presence of specific memory management tools.\n- [PROMPT_INJECTION]: Presents a surface for indirect prompt injection by processing external data from memory systems and user-controlled files without sanitization.\n
  • Ingestion points: Data retrieved from external Memory Control Protocol (MCP) tools (harness_mem, serena) and contents of project-level files like Plans.md.\n
  • Boundary markers: No delimiters or instructions are used to separate ingested content from system prompts in the documentation templates.\n
  • Capability inventory: Includes file system modification (Write, Edit), command execution (Bash), and interaction with memory toolsets.\n
  • Sanitization: No validation or escaping mechanisms are applied to the ingested data before it is incorporated into project files or the agent context.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 10:24 AM