Skip to main content
Edit

Start here

Give your coding agent a task. Shaka guides it through implementation, checks, independent review, and a pull request you can understand and decide whether to merge. Start with a small fix in a project you know.

Examples use $shaka in Codex. Use /shaka in Claude Code, Cursor, or OpenCode; in Pi, load the installed Shaka skill. See coding agents.

1. Install Shaka​

Ask your coding agent:

Install Shaka from https://github.com/shakacode/shaka.

Prefer to run the commands yourself? Follow manual installation.

The agent checks for Ruby 3.4 or later, Git, and an authenticated GitHub CLI. It installs one dedicated checkout outside your projects and links the skill directly into it. The default location is ~/.agents/shaka; you can choose another directory, such as ~/agent-tools/shaka.

Keep the installation outside your project repositories. Use separate checkouts for development so the installation stays ready for updates.

The skill prompts you when an update is available. Ask your agent to update Shaka.

Start a new chat if the skill does not appear. Use a source installation: the published gem is a name-reservation prerelease without the current workflow.

Try $shaka without a task for a guided welcome and setup check. Or ask:

$shaka doctor

Doctor reports what is ready, which supported reviewer CLIs are on PATH, and where to install missing ones. It gives setup advice even before this repository is configured. Finding a CLI does not verify sign-in, quota, or a working review. An extra provider can supply independent review; installing every CLI is optional. The diagnostic changes no settings. Give the agent a task when you are ready.

2. Choose individual or team setup​

Open a chat in your project and give Shaka a task:

$shaka <task or issue link>

If Shaka is not configured, the agent asks which setup you want. Individual setup keeps settings out of commits in a project without team settings. To choose it upfront:

$shaka <task or issue link>
Set up Shaka just for me. Go.

The agent finds your test and lint commands, explains AI review choices, and defaults to Ask. Individual setup continues into the task without a team setup PR. Host permissions may require approval for protected Git writes. See individual support and acceptance gaps.

For shared team settings, ask:

$shaka Set up Shaka for this repository.

The agent prepares shared settings and scripts in a setup PR. They connect Shaka to your project's commands and record its review and merge choices. Review the choices and merge this first setup PR yourself on GitHub. To try its settings before adoption, select the settings PR for a task. If the repository is already configured, skip this step. You can ask the agent to change settings later; you do not need to maintain them by hand. See repository setup for details.

After merging shared reviewer settings, check whether your signed-in AI CLI accounts can run the selected reviewer models and effort levels. This catches account or model-access problems before the agent reaches code review. Ask:

$shaka Probe this project's configured reviewers.

This live probe may consume quota or cost money. It keeps your chosen settings. The agent sends each supported configured reviewer one short prompt and reports which attempts succeeded. If an attempt fails, it explains the problem and asks you how to proceed.

3. Start a task​

$shaka Fix search when the query contains an apostrophe. Go.

Replace the example with a small outcome or an issue link. Go lets the agent start with its current model and effort settings. It handles routine work and asks when a missing choice, access problem, or consequential decision needs you.

For a different outcome, choose an example:

Look for a PR explaining what changed, what was tested and reviewed, and any remaining gaps. With Ask, you decide whether to merge; Auto lets the agent merge after the required checks, review, and approvals. Independent opening-check choices shipped in #425 after #420. Merged #426 added settings previews. Working with Shaka covers these choices, useful prompts, feedback, and resuming a task.

Use a personal fork​

To customize Shaka and contribute improvements upstream:

Fork https://github.com/shakacode/shaka and install my fork.

For later maintenance, ask the agent to upgrade your installation. To evaluate an unmerged workflow change, try a Shaka PR on real work.