hermes-war-room-ui

Fail

Audited by Gen Agent Trust Hub on May 17, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The installation instructions direct the user to download a compressed archive from an external GitHub repository (github.com/Naroh091/hermes-war-room) that is not part of a recognized trusted organization.
  • [REMOTE_CODE_EXECUTION]: The skill executes code from the downloaded external archive directly using Node.js (node .output/server/index.mjs).
  • [COMMAND_EXECUTION]: The server-side logic (e.g., in server/api/operatives/create) utilizes shell execution (execAsync) with parameters such as 'slug' and 'cloneFrom' that are derived directly from user-provided API request bodies, creating a high-risk command injection surface.
  • [DATA_EXFILTRATION]: The skill requires access to and reads from sensitive local directories, specifically ~/.hermes, which contains the agent's database (kanban.db) and configuration profiles (SOUL.md).
  • [PROMPT_INJECTION]: The skill processes untrusted external data, creating an indirect prompt injection surface.
  • Ingestion points: Untrusted data enters the agent context through the mission message API (/api/missions/:id/message) and task summaries retrieved from the database (SKILL.md).
  • Boundary markers: No delimiters or instructions to ignore embedded commands are present in the described logic.
  • Capability inventory: The system has capabilities for subprocess execution (execAsync) and file system access across all scripts (SKILL.md).
  • Sanitization: No validation or escaping of the external input is demonstrated in the provided code snippets.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
May 17, 2026, 06:22 PM
Security Audit — agent-trust-hub — hermes-war-room-ui