Llmtpl

llmtpl

[](https://github.com/ryokwkm/llmtpl/actions/workflows/ci.yml) [](https://github.com/ryokwkm/llmtpl/releases) [](LICENSE)

**English** | [日本語](README.ja.md)

Turn AI agent config features on and off, one line at a time.


llmtpl manages the config files an **AI coding agent** reads: the instructions file, the long rules it points at, hook scripts, and the settings that register them. It keeps everything one feature needs in **one directory**, and every repository turns that feature on or off with a single line.

logcheck = true

That line brings in the instructions, the rules, the script, and the registration. Flip it to `false` and all of it leaves — nothing to undo by hand.

Install

brew install ryokwkm/tap/llmtpl

Or download a binary from [Releases](https://github.com/ryokwkm/llmtpl/releases), or with Go:

go install github.com/ryokwkm/llmtpl@latest   # needs Go 1.24+

From source, `make build` writes `./llmtpl`. `make install` puts it in `~/.local/bin` and installs zsh completions (macOS/zsh assumed). `llmtpl completion zsh|bash|fish` prints them individually.

Try it

Everything below runs as written, and the output is real. This README covers getting started; the details of the spec, and the reasoning behind them, are in [docs/reference.md](docs/reference.md).

Put one feature in one directory

A one-feature directory is a **bundle**; a project that receives config is a **target**. Those are the only two words. The directory name — `logcheck` — becomes the flag name, so creating a directory creates a flag; there is no registry to update.

demo/
├── llm-tpl/