A focused developer-workflow agent that turns one small GitHub issue into a clear review package: accepted criteria, selected files, change plan, code result, test output, and reviewer summary.
This project does not try to build a full IDE. It focuses on one narrow workflow: understand the issue, select context, plan a small change, run tests, and produce a result a human can review quickly.
AI coding tools can generate code quickly. The harder problem is making changes that are small, relevant, tested, and easy to review.
A useful coding agent should answer:
- What is the issue asking for?
- What are the accepted criteria?
- Which files are relevant, and why?
- What is the smallest reasonable change?
- What tests were run?
- What should the reviewer check?
The product should feel like a compact review assistant for small issues.
Core screens:
| Screen | What it shows |
|---|---|
| Issue Intake | Title, body, labels, accepted criteria |
| Repository Map | Project structure, package manager, test command |
| Selected Files | Files chosen for the change, with short reasons |
| Change Plan | Small plan before implementation |
| Code Result | What changed and where |
| Test Output | Command, result, and short notes |
| Review Summary | What the reviewer should inspect |
Build a narrow coding agent that reads a small GitHub issue, selects relevant files, writes a change plan, produces a small result, runs tests, and writes a review summary.
issue -> criteria -> repo map -> selected files -> plan -> result -> tests -> review summary
The first demo should show that the system makes code review easier instead of hiding the work.
| Priority | Workstream | Outcome |
|---|---|---|
| P0 / MVP | Supported issue types and demo repo | The agent has a narrow and believable scope |
| P0 / MVP | Issue intake and repo map | The agent understands the task before editing |
| P0 / MVP | Relevant file retrieval | The agent avoids loading the whole repository blindly |
| P0 / MVP | Showcase path | One demo path from issue intake to review summary |
| P1 | Plan and minimal code result | The change is reviewable before tests run |
| P1 | Test runner and one repair note | Test feedback becomes part of the loop |
| P1 | Draft PR summary | Human reviewers can inspect the result quickly |
| P2 | Interview notes and demo script | The project is easy to explain to AI devtools teams |
- Select a small GitHub issue.
- Show the accepted criteria.
- Show the repository map and selected files.
- Generate a small change plan.
- Show the result.
- Show test output.
- Show a reviewer-friendly summary.
- Context engineering for coding tasks.
- Narrow scope control and file selection discipline.
- Test-aware development loop.
- Clear communication for human reviewers.
- Strong fit for AI devtools and developer productivity roles.
Planning and scaffolding. Issues are used as the implementation roadmap. The next build target is the P0 MVP showcase path.