codex-memory-curator
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill operates entirely within the local environment using standard Node.js libraries. No external dependencies are fetched or executed, and no network operations are initiated.\n- [SAFE]: Potential exposure of sensitive data in Codex memories is mitigated by a dedicated risk scanner (
scripts/scan-memory-risks.mjs) that redacts credentials, tokens, and PII using regex patterns before reporting findings to the agent or user.\n- [COMMAND_EXECUTION]: The skill invokes internal Node.js management scripts for file operations. These scripts implement defensive programming techniques, such as anchored directory traversal checks, symlink rejection, and secure file opening flags (O_NOFOLLOW), to ensure file system operations remain scoped and safe.\n- [INDIRECT_PROMPT_INJECTION]: As the skill audits durable user state (Codex memory), it is exposed to data from previous sessions. It mitigates indirect injection risks through a structured classification rubric, atomic review requirements, and mandatory human approval for all mutations.
Audit Metadata