claude-mem-coded-assistant

Pass

Audited by Gen Agent Trust Hub on Mar 5, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [NO_CODE]: The skill is comprised entirely of Markdown documentation and instructions. No executable scripts, binaries, or configuration files that trigger code execution are included in the skill package.\n- [PROMPT_INJECTION]: The workflow described in the skill creates a surface for indirect prompt injection by ingesting data from an external memory source and converting it into instructions in the project's primary control file.\n
  • Ingestion points: Data enters the agent's context through observations retrieved via the claude-mem MCP tools (search, timeline, get_observations) as detailed in claude-mem-usage.md.\n
  • Boundary markers: The skill instructions do not specify the use of boundary markers or protective wrappers when incorporating memory observations into CLAUDE.md or MEMORY.md.\n
  • Capability inventory: The skill facilitates modification of key local project files (CLAUDE.md and MEMORY.md). It does not define capabilities for subprocess execution, external network requests, or broad filesystem access.\n
  • Sanitization: No programmatic sanitization or validation of external content is included; the skill relies on the agent's distillation process to exclude malicious or sensitive data.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 5, 2026, 03:28 AM