wile CLI

Autonomous AI coding agent that ships features while you sleep.

Install

``bash
bunx wile --help
`

Configure a project

From your project repo:

`bash
bunx wile config
`

This creates:

- .wile/secrets/.env for required credentials
- .wile/.env.project for env vars forwarded into the container (or set WILE_ENV_PROJECT_PATH)
- .wile/.gitignore to ignore secrets/, screenshots/, and logs/
- .wile/prd.json (empty) and .wile/prd.json.example

Set WILE_REPO_SOURCE=local in .wile/secrets/.env to run against the current directory without GitHub.
When WILE_REPO_SOURCE=local, GitHub credentials are optional.
Set WILE_MAX_ITERATIONS in .wile/secrets/.env to change the default loop limit (default: 25).
Set CODING_AGENT=CX to use Codex CLI, CODING_AGENT=OC to use OpenCode (OpenRouter), CODING_AGENT=GC to use Gemini CLI, otherwise CODING_AGENT=CC uses Claude Code.

Run Wile

`bash
bunx wile run
`

Optional flags:

- --repo : override repo URL
- --max-iterations : default 25
- --test: run the mocked agent in Docker without GitHub/Claude

Logs

Each run writes a session log to .wile/logs/run-YYYYMMDD_HHMMSS.log.

Publish

Release (bump version + build):

`bash
./scripts/release-cli.sh patch
`

Publish:

`bash
./scripts/publish-cli.sh
``