codex-memory-curator

Pass

Audited by Gen Agent Trust Hub on Jul 6, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill is designed for administrative cleanup of the agent's memory store. It implements a strict 'Human-in-the-loop' model, requiring explicit approval and providing a preview of changes before any file modifications occur.
  • [DATA_EXFILTRATION]: While the skill processes potentially sensitive data such as credentials stored in memory, it includes specific redaction logic in scripts/scan-memory-risks.mjs and instructions in SKILL.md to ensure secrets are never printed in full. This is handled as a security feature within its intended use case.
  • [COMMAND_EXECUTION]: The skill uses local Node.js scripts for file operations. These scripts (inventory-memories.mjs, scan-memory-risks.mjs, backup-memories.mjs) are non-interactive, use only standard Node.js libraries, and are bundled with the skill for transparency.
  • [SAFE]: The skill manages an indirect prompt injection surface (reading external memory files and modifying them) by enforcing strict validation rubrics, mandatory manual verification of all changes, and automated data redaction. 1. Ingestion points: memory files (SKILL.md). 2. Boundary markers: Classification rubrics in references/classification-rubric.md and conflict resolution rules. 3. Capability inventory: Backup and edit operations. 4. Sanitization: Redaction logic in scripts/scan-memory-risks.mjs.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 6, 2026, 04:21 PM
Security Audit — agent-trust-hub — codex-memory-curator