← use cases · runtime
Running the generator on Codex instead of Claude Code
Switch the headless agent runtime from Claude Code to Codex with a single config value.
Who it's for
Teams standardised on a different agent runtime, or anyone who wants to compare generation quality across runtimes.
The problem
skillterm should not lock you to one agent vendor. The generator logic is runtime-agnostic; only the binary that does the reasoning changes.
How skillterm does it
- 1
Install the alternative runtime (npm install -g @openai/codex).
- 2
Set runtime = "codex" in ~/.skillterm/config.toml, with its binary path, model, and timeout.
- 3
Run skillterm generate as usual, or override per-invocation with the provider flag.
- 4
The SKILL.md format and the Tab-time behaviour are identical regardless of which runtime wrote the file.
The outcome
The same skill-generation workflow on a different agent runtime — configuration, not a fork.