Case study
An AI reviewer found two bugs. A developer merged anyway.
Two useful review comments went unanswered among 83 conversation items. See how clearer summaries and follow-up can help a developer act on what matters.
The sandwich model
Keep human judgment around the agent loop.
What you get
Know what was checked before you merge.
You should be able to assess the work without reconstructing an agent’s conversation. A Shaka PR puts the result, verification, and decisions in front of you.
What changed, and why?
A clear description of the result, with a code walkthrough that explains the implementation choices.
How was it checked?
Tests, independent review, and evidence tied to the tested commit. UI changes include desktop and mobile before-and-after screenshots.
What needs my decision?
Review findings, missing checks, and delivery decisions stay visible. Longer results, available usage and cost, and notes for resuming work are expandable.
Explore real Shaka PRs → See their tests, review findings, code walkthrough, and reported verification gaps. See a before-and-after UI example →
Inside a Shaka PR
See the workflow in real pull requests.
Follow the numbered callouts to see verification, independent review, and code walkthroughs in recent Shaka PRs. Open each source to read the full context.
shakacode/shaka · PR #372
See what was checked.
- Evidence tied to a commit
The check table names the tests, assertions, and commit that passed.
- More than automated tests
Desktop and mobile captures, review results, and a product checkpoint appear alongside the checks.
- Details when you need them
Expand usage and cost, execution provenance, or repository settings without losing the summary.
shakacode/shaka · PR #350
Local adversarial review
- Know who reviewed what
Each round names the commit, reviewer, model, effort, and review criteria.
- Track findings through repair
The first two rounds each record a fix. Later findings are documented, with the full reports available below.
- Read the outcome first
The summary says why the loop ended and what remains. Reported cost is labeled as an estimate.
shakacode/shaka · PR #325
Understand the change and its limits.
- An explanation linked to code
The walkthrough connects the before-and-after behavior to the implementation and its regression test.
- Risks and rollback stay visible
The reviewer can see which installations change behavior and what reverting would restore.
- Verification gaps are explicit
The full-suite row reports three failures that also occurred on main locally. A passing focused test does not hide them.
Cropped GitHub screenshots captured October 1, 2026. Callouts explain the recorded results; they are historical examples, not a claim about any PR’s current readiness.
For developers
One workflow, adapted to your project.
Open-source projects
Give maintainers contributions they can assess: a focused change, its verification evidence, and answers to important review findings.
Company repositories
Use your existing checks and review rules. Start with the agent preparing a PR for your decision; configure automatic merging within your limits when appropriate.
Independent projects
Spend less time directing the process. Try one task, use the checks your project needs, and improve the workflow as you go.
How it works
Describe the task. Shaka supplies the steps.
Describe the outcome
Give Shaka a task or issue. The skill guides your coding agent through planning, implementation, verification, review, and delivery.
Use your repository’s checks
Repository settings supply the commands and reviewers. The agent tests and reviews locally, fixes problems, and then pushes the PR.
Configure your repository →Review the result
The PR brings back the evidence and decisions. With Ask, you decide when to merge. With Auto, the agent merges after required checks and approvals, within your configured limits.
Working with Shaka →Optional: two AI reviewers from different providers read each commit before it’s pushed; their findings are merged into one list, so each problem is fixed once.
Your project is growing
Built something with AI? Take the next step.
A personal experiment can become a tool other people rely on. As users and saved data become part of the picture, add checks that protect their time, data, and trust.
Start with one change in a GitHub repository. Use Shaka to prepare a PR, try the result, and understand the checks and review before you merge.
Evidence from real use
Finding a bug is only the first step.
Get started
Install. Configure. Give it a task.
Shaka needs Ruby 3.4 or later, Git, an authenticated GitHub CLI, and a coding agent that can load skills and run commands. The guide gives you prompts for installing Shaka and configuring your repository.
For engineering teams
ShakaCode can help your team
Shaka is open source. ShakaCode helps engineering teams choose AI coding tools, set review and verification rules that match their projects, and improve how developers work with agents.