memory-palace-red-queen

Warn

Audited by Gen Agent Trust Hub on May 15, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes git commands using child_process.execSync to retrieve repository metadata for context detection in commands/context.md. Additionally, archived code in evolutions/hypothesis-001-hooks/automated-system.js uses require() with computed paths to load configuration and data files from the user's memory directory.
  • [PROMPT_INJECTION]: The system presents a vulnerability to indirect prompt injection. 1. Ingestion points: User memories containing arbitrary text are stored via commands/store.md. 2. Boundary markers: Not utilized in sub-agent prompt templates (e.g., subagents/examiner.md). 3. Capability inventory: Orchestration of multiple sub-agents and access to the local filesystem and network. 4. Sanitization: No filtering or validation is applied to stored memories before they are used as prompts for specialized AI agents.
  • [EXTERNAL_DOWNLOADS]: The skill recommends installation via npx to execute code from the public npm registry.
  • [DATA_EXFILTRATION]: The skill includes a feature in features/export-import/gist.js to export local memory palaces to GitHub Gists, allowing potentially sensitive local data to be transmitted to a remote service.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 15, 2026, 07:17 AM
Security Audit — agent-trust-hub — memory-palace-red-queen