Awesome Hermes Plugins › Plugin Collections & Developer Resources

mauricemohr88-debug/hermes-plugin-guard

⭐ 16 Python added to this list on 2026-10-05 repository created 2026-07-23

Hermes Plugin Guard is an unofficial, community-built static analysis tool for reviewing third-party Hermes Agent plugins before enabling them. Distributed as a standalone CLI (hpg, installable from PyPI via pipx) and, separately, as a native review-only Hermes plugin, it treats target plugin code strictly as data: files are read and pattern-matched, never imported or executed. The scanner checks plugin manifests, Python source, declared dependencies, and basic repository hygiene for a defined set of high-risk patterns, including direct subprocess calls that bypass Hermes' own terminal-approval path, dynamic code execution and unsafe deserialization, access to sensitive filesystem paths, destructive filesystem operations, disabled TLS verification, network listeners bound to all interfaces, undeclared secret environment variables, privileged plugin registration and middleware surfaces, work performed at import time, likely committed credentials, mutable remote dependencies, remote scripts piped directly into a shell, and gaps such as missing tests or project policy files. The standalone CLI runs entirely locally, makes no network calls, and uploads nothing. The native Hermes integration is more constrained by design: it exposes a single read-only model tool that accepts only an already-installed, currently-disabled plugin, applies a fixed high-severity rule set, and returns at most twenty findings as opaque rule IDs, severities and bounded line numbers, deliberately omitting filenames, source text, and other content that a model-facing response should not carry, since that response can become part of the active conversation sent to the configured model provider. The project is explicit that static analysis is a first fast pass, not a substitute for code review, and that current Hermes releases have no admission-gate hook, so the guard cannot enforce a scan before every plugin activation; it documents this boundary directly rather than overstating what the tool can guarantee. It targets Hermes plugin authors and operators who want a repeatable pre-enablement check, in CI or locally, before trusting a new plugin with the permissions of the Hermes process.

https://github.com/mauricemohr88-debug/hermes-plugin-guard

hermes-pluginsecurity-scannerstatic-analysisplugin-reviewdeveloper-tool