Executive brief
NousResearch hermes-agent is an AI agent framework. A security vulnerability in its skill-viewing tool allows an attacker to bypass directory restrictions and access files they should not be able to see. This could lead to the exposure of sensitive information, such as API keys, configuration files, or environment secrets stored on the host system.
Technical details
A path traversal vulnerability exists in NousResearch hermes-agent version 2026.5.29.2 within the `skill_view` function of `tools/skills_tool.py`. The root cause is the improper validation of the `Name` argument, which is joined with the trusted search root using `search_dir / name` before any containment checks are performed. An attacker can provide a manipulated path (e.g., `../outside-skill`) to select a directory outside the intended `~/.hermes/skills/` boundary. If the target directory contains a `SKILL.md` file, the agent will treat it as a valid skill directory, allowing subsequent file reads (via `file_path`) to leak sensitive files like `.env` from the host. The issue has been addressed in commit 56f833efa427ccb444c0f9ad1759af1012f2124d.
Affected products
- NousResearch hermes-agent 2026.5.29.2
Timeline
- 2026-06-06: patched: Fix merged in pull request 40566
- 2026-07-06: advisory: NVD publication date
References
- https://github.com/NousResearch/hermes-agent/
- https://github.com/NousResearch/hermes-agent/commit/56f833efa427ccb444c0f9ad1759af1012f2124d
- https://github.com/NousResearch/hermes-agent/issues/38643
- https://github.com/NousResearch/hermes-agent/pull/40566
- https://vuldb.com/cve/CVE-2026-14783
- https://vuldb.com/submit/847502
- https://vuldb.com/vuln/376373