Toha
Generate projects and files from templates.

Toha is a project scaffolding tool and Rust crate. A template contains an interview and files to render with Jinja. Toha plans the output before it writes, so you can inspect a dry run and handle conflicts.
The same interview engine serves terminal users, scripts, agents, and Rust callers. Scripts and agents can answer one JSON Schema batch at a time and resume a staged interview in another process.
- One interview for people, scripts, agents, and Rust callers
- Resumable async interviews and headless answers files
- Typed Jinja values and rendered file paths
eachfile generation and static files- Template hooks gated by trust
- Git template sources, short names, and aliases
- Embedded agent skills through
toha skills
Read the Toha documentation for guides and the full contract.
Install
Homebrew:
APT:
|
|
RPM:
Arch Linux (AUR):
GitHub Releases also
provides Linux and macOS tarballs, Windows zip archives, and SHA256SUMS for
direct installation.
First run
Toha ships a small demo template inside the binary. From any directory, offline, with nothing installed:
Answer Note title and Topic. To inspect the file plan without writing
it, use toha apply toha-demo ./preview --dry-run.
toha-demo is a reserved fallback name. If you install or alias a template as
toha-demo, that template resolves instead.
Using Toha from Rust
Add the library without the CLI dependencies:
From this repository checkout, drive the same interview with an identity-bearing answers document. The document names the template it answers, which Toha compares by exact string equality with the formal name the caller has established before it evaluates any answer:
use Path;
use ;
use ;
use ;
use canonical_target;
let template_root = new.canonicalize.unwrap;
let template = load.unwrap;
let expected = template_root.to_str.unwrap;
let target = canonical_target.unwrap;
let seed = Seed ;
let Asking = start.unwrap else ;
// The document must name the formal template it answers.
let text = json!.to_string;
let Accepted =
answer_document_once.unwrap
else ;
let Complete = interview else ;
assert_eq!;
In-memory callers that build answers directly still use Pending::answer with
RawAnswers and the raw headless walk; the envelope is only for external JSON.
See CONTRIBUTING.md for development and changes.