memory-management

Pass

Audited by Gen Agent Trust Hub on Mar 31, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it interprets and acts upon data stored in external files (working memory and deep memory).
  • Ingestion points: CLAUDE.md and the memory/ directory files (glossary.md, people/.md, projects/.md) are read at runtime to decode user requests and influence agent behavior.
  • Boundary markers: No delimiters or instructions are used to distinguish between system-trusted instructions and potentially untrusted content within the memory files.
  • Capability inventory: The agent uses retrieved memory to interpret intent, draft communications (e.g., Slack/Email), and execute project workflows.
  • Sanitization: No sanitization, escaping, or schema validation is performed on the content of the memory files before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 31, 2026, 02:03 AM
Security Audit — agent-trust-hub — memory-management