A Claude Code skill that routes hands-on implementation work (coding, refactors, known-repro bug fixes, test writing, CI fixes, mechanical migrations, bulk codebase reads) to the local Hermes Agent CLI, while keeping design, spec-writing, review, and verification in Claude.
The idea: Claude's tokens are metered and expensive; the backend Hermes runs on (often a flat-rate coding model) is usually faster and cheaper for typing out code once the spec is clear. Claude's real edge is judgment — so Claude designs and freezes a spec, Hermes builds it, and Claude reviews the diff before anything ships.
This is adapted from the codex-first skill pattern by @steipete, retargeted from a direct codex exec invocation to the fuller hermes chat CLI.
- Claude Code
- The
hermesCLI installed and configured (hermes statusshould show a working model/provider)
Copy or clone this repo into your project's skills directory:
git clone https://github.com/dax8it/hermes-first.git .claude/skills/hermes-firstOr just drop SKILL.md into .claude/skills/hermes-first/SKILL.md in your project.
Once installed, Claude Code will consult this skill automatically whenever a task in your repo reads like a work order — a frozen spec to build, a script to write, a migration to run — rather than something requiring design judgment. See SKILL.md for the full routing rules, invocation pattern, and verification steps.
MIT — see LICENSE.