claude-memory-curator
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious patterns were detected in the skill instructions or associated scripts. The skill demonstrates high-quality security engineering, including atomic file operations, symlink protection, and automated redaction of sensitive data.\n- [COMMAND_EXECUTION]: The skill utilizes local Node.js scripts (inventory-claude-memory.mjs, scan-claude-memory-risks.mjs, backup-claude-memory.mjs) to manage and curate memory files. These scripts are self-contained and use standard Node.js modules for local maintenance tasks.\n- [INDIRECT_PROMPT_INJECTION]: The curation process involves reading untrusted repository files like CLAUDE.md and .claude/rules.\n
- Ingestion points: Reads project-specific and user-level Claude Code context files from the local filesystem.\n
- Boundary markers: Implements strict report templates and explicit approval workflows to prevent accidental obedience to embedded instructions.\n
- Capability inventory: Limited to reading, scanning, and writing backups/edits for specific Claude context surfaces.\n
- Sanitization: Automatically redacts tokens, secrets, and private identifiers using regular expression scanning in scan-claude-memory-risks.mjs before presenting content to the agent or user.
Audit Metadata