Skip to content

Hermes security blocked install #513

Description

@asheem22

Summary

Hermes's security scanner blocked it — DANGEROUS verdict, 50 findings. Most are false positives from the scanner being overly cautious with documentation files and subprocess calls that are legitimate for a research tool. The two real flags worth noting:

  • HIGH injection in SKILL.md:455 — the skill tries to override agent behavior mid-run with CRITICAL: instructions. That's a real concern.
  • CRITICAL exfiltration in scripts/lib/env.py — reads environment variables. Likely for API keys, but worth knowing.

You have a few options:

  1. Override the security scan — use hermes config set security.tirith_enabled false, install, then re-enable. Bypasses the block.
  2. Manual install — clone the repo and symlink just the SKILL.md, not the full scripts. Gives you the skill text without executing the scripts.
  3. Skip it — I can do similar research using my built-in web tools (Reddit via web, HN via Algolia API, YouTube via yt-dlp directly). Less polished but no security risk.

My honest recommendation: Option 3 for now. The injection flag in SKILL.md is a real red flag — it tries to hijack agent behavior after research completes. Not something I'd want running autonomously in a cron. We can build a clean research skill ourselves later that hits the same sources without the prompt injection risk.

Steps to Reproduce

Try install on Hermes

Expected Behavior

Should have installed no issues

Error / Traceback


Install Method

Claude Code plugin

OS

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions