Wukong Code
Make coding agents work like engineers
A software development methodology for coding agents.
Discovery, design, TDD, review, verification. Beyond generating code, finish the engineering.
26 top-level skills · 10 development environments · zero core dependencies · MIT licensed
-
Clarify goals, boundaries, and acceptance criteria
-
Compare approaches and obtain approval
-
Break down tasks, risks, and verification commands
-
RED → GREEN → REFACTOR
- RED
- GREEN
- REFACTOR
-
Check requirement fit and code quality
evidence attached -
Run fresh tests, builds, and reproductions
PASS -
You choose the final integration path
A failed check returns work to the stage that needs evidence
The problem
AI writes fast. Reliable delivery is another matter.
The code is generated before the requirement is clear.
The bug is not located, so every change is a guess.
The agent declares success before the tests run.
The missing engineering process: design, tests, verification.
Core capabilities
Reliable engineering, by default
-
Design before coding
Clarify the goal, compare approaches, then start.
-
Real test-driven development
See the failure first, then write the smallest implementation.
-
Systematic debugging
Reproduce, gather evidence, find the root cause, then fix.
-
Review throughout
Keep checking requirement fit and code quality.
-
Verify before declaring done
Without fresh tests or builds, it is not done.
-
Triggered by the task
Skills pick the workflow from context. No repeated prompts.
Delivery evidence
From idea to reliable delivery
Simple tasks stay simple. Complex tasks get constraints.
$ wukong run --task feature
discovery ............. ✓design approval ....... ✓implementation plan ... ✓TDD (RED) ............. ✓TDD (GREEN) ........... ✓code review ........... ✓ evidence attachedfinal verification .... ✓ PASSmerge / PR ............ ready
Product design
Product Design, same workflow
From research, audits, and visual exploration to implementation and Design QA.
- 01 · VISUAL EXPLORE Choose a direction
- 02 · BUILD Implement the real interface
- 03 · DESIGN QA Compare the rendered result
Project dashboard
- Requirements review To organize
- Interaction states Unspecified
- Design checks To review
Project dashboard
- Requirements review Verified
- Interaction states Covered
- Design checks Passed
Method comparison
More than a prompt template
| Dimension | Ordinary prompt | Wukong Code |
|---|---|---|
| Trigger | Depends on manual reminders | Triggers from the task |
| Scope | Focuses on code generation | Covers the delivery workflow |
| Debugging | Relies on guesses | Investigates the root cause first |
| Completion | Depends on judgment | Requires verification evidence |
Environments
Use it where you already work
Zero core dependencies. No extra tools. No overwritten commands.
- Antigravity
- Claude Code
- Codex App
- Codex CLI
- Cursor
- Factory Droid
- GitHub Copilot CLI
- Kimi Code
- OpenCode
- Pi
Next step
Code can be fast. Quality cannot depend on luck.
Let agents do the right work at the right time.
More than code—delivery you can trust.