Tell Hermes to run an experiment with Codex then review the result with Claude Code, all from a single prompt.
Coordinate multi-agent coding tasks across separate working directories without switching between terminals.
Route bug reproduction, diff review, and code generation to different AI tools based on their strengths automatically.
Requires Hermes Agent and at least one supported coding tool such as Codex or Claude Code already installed and working.
Hermes Code Bridge is a plugin for Hermes Agent that turns Hermes into a coordinator for local AI coding tools. If you already use command-line coding assistants such as Codex, Kimi Code, Claude Code, OpenCode, or Gemini CLI, this plugin lets Hermes direct them on your behalf rather than you having to switch between terminals and copy prompts manually. The core idea is that different coding tools are good at different things. Codex tends to handle focused implementation tasks well. Kimi Code has a large context window suited to messy or very large codebases. Claude Code is often used for careful review passes. Without something to coordinate them, a developer has to act as the router: copy a prompt into one terminal, wait, copy the output into another, and so on. Hermes Code Bridge gives that routing job to Hermes. A typical request looks like: tell Hermes to use Codex to run an experiment, then ask Claude Code to review the result, and come back with a summary of what changed, what tests ran, and what risks remain. Hermes dispatches the work, monitors the coding agents as they run, collects their output and any file changes, and presents the evidence in one place. Installation is a single command that installs the plugin from the repository URL. The plugin also ships as a standalone skill file if you want to add the capability without the full plugin wrapper. Once installed, you trigger it with a slash command followed by a plain-English description of what you want done and which tool to use. The safety rules built into the skill are explicit: Hermes runs the tool you actually asked for rather than substituting something else, confirms risky actions before dispatching, does not touch session internals or databases belonging to the coding agents, and reports failures honestly rather than inventing a successful result. The README includes a table of supported backends and a set of example prompts covering common workflows such as bug reproduction, diff review, and multi-agent coordination across separate working directories.
← xuyang-liu16 on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.