claude-memory-curator
Pass
Audited by Gen Agent Trust Hub on Jul 6, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a well-scoped utility for managing Claude Code's memory and instructions. It implements safety gates, such as mandatory backups and user approval before any file modifications.
- [DATA_EXFILTRATION]: While the skill accesses potentially sensitive Claude context files (instructions, settings, and auto-memory), it does not perform any network operations. It includes a risk scanning script (scripts/scan-claude-memory-risks.mjs) specifically designed to redact tokens, secrets, and private identifiers from its output.
- [COMMAND_EXECUTION]: The skill utilizes several internal Node.js scripts for its operations. These scripts are confined to the local filesystem, use only built-in Node.js modules, and are scoped to project and user directories related to Claude Code.
- [PROMPT_INJECTION]: As a tool for auditing instruction files, the skill naturally processes content that could contain indirect prompt injections. It addresses this risk through structured workflows, specific safety rules in SKILL.md, and processing content in bounded chunks with built-in redaction.
- Ingestion points: Reads instructions from CLAUDE.md, .claude/rules/, and auto-memory topic files.
- Boundary markers: Instructs the agent to process content in bounded chunks and use the redact function.
- Capability inventory: Capabilities are limited to local file system access and execution of internal Node.js scripts.
- Sanitization: Implements redaction logic to mask sensitive data identified during scanning.
Audit Metadata