Lydite

From its catalog listing
Hand a GitHub issue and the command that proves it done to Lydite from the chat. Claude Code workers do the job, Lydite replays your proof itself, and a pull request comes back with a receipt (cost, workers, retries, the commit that was reviewed and replayed, and what was not replayed). Also hands Hermes the lessons Lydite keeps for a repository, read-only, has Lydite check a plan file before a launch, and turns a hermes-security findings export into one Lydite plan per finding you choose (prove it by a failing test, then fix it).
Good to know before installing
after your approval of each launch (the prompt shows the repository, work, exact proof command or, for a plan, its header proof (cut if long) and SHA-256, profile and target branch; a value holding a control or formatting character, and a plan in which a task carries its own proof, are refused before any prompt; "always" covers only that exact request: repo, profile, issue or objective, target branch, dry run or launch, and proof or plan bytes), starts the already-installed `lydite` binary (argument list, no shell) as a detached background process with a closed environment list plus the provider key you set; Lydite runs git, gh and Claude Code on the host with your GitHub login; under a `mur = exige` profile (the default) the workers' commands run in Claude Code's sandbox and srt replays your proof inside the wall, `mur = natif` does the same with Lydite's own bubblewrap wall in place of srt, and one of the two is required for `--issue` and for a plan written from a findings report; under a profile you accepted with `mur = demande` the workers' commands may run outside the sandbox and see the provider key, and your proof is replayed directly on the host; it reaches GitHub and the model provider of its profile; the status, lessons and plan-check tools run the binary without approval, as does the read-only wall check (`lydite profil`) before a findings plan's prompt; the findings tool starts nothing; reads the plan file, the repository's `.agents/skills` lessons and the findings report you name, and writes the plans of the findings you choose under the plugin's data folder; no telemetry.
From its catalog listing.
How to install
- Install it. Hermes asks for any API keys it needs, then asks whether to turn it on.
hermes plugins install lydite - Said no? Turn it on later:
hermes plugins enable lydite - Restart Hermes so it loads the plugin.
Some plugins also need a sign-in or settings: see Full setup & settings. Memory plugins ask whether to become your memory provider instead. Steps follow the official plugins guide (opens in a new tab).
- Category
- Coding & developer tools · Automation & multi-agent
- Maintained by
- aoricloud (Community)
- Needs
- No API keys declared
- Adds tools
lydite_lancerlydite_etatlydite_leconslydite_planlydite_findings- Adds hooks
pre_tool_call- Works on
- linux
- Needs Hermes
- 0.23.0 or newer
- Version
- 0.4.0
- In the catalog
- 2026-11-11, updated 2026-11-12
Known issues (from its catalog listing)
- Linux only: Lydite's sandbox wall holds there.
- Needs the lydite binary of the same release as this plugin (attested release of github.com/aoricloud/lydite), Claude Code, gh and srt installed (bubblewrap and lydite-reseau instead of srt under mur = natif), and a Lydite profile accepted by a human.
- Some names are still French: three tool names (lydite_lancer, lydite_etat, lydite_lecons), the plugin settings (binaire, profil, cle_nom, cle) and a few lines Lydite writes, such as impossible (déclarée). Lydite's commands have English words, and its usage and the lines about the pull request and CI are in English; pull request receipts are in English. The progress lines of a launch are still in French.
Using the Hermes desktop app? See where plugins live

Full setup & settings ↗Source code ↗
More in coding & developer tools
All coding & developer tools plugins
New Hermes Agent updates every day on X. Follow @HermesWatcher