Vibecoding Full Lifecycle Engineering Workflow

Problem Solved

Many AI coding sessions stop at runnable code: unclear requirements, blind global replacements, unbacked database writes, credentials in logs, and diagrams that cannot be read back. Gongcheng turns Vibecoding into an enforceable engineering flow: clarify the request, explore an MVP, then harden architecture, implementation, tests, deployment, and handoff.

How It Works

• Workflow orchestration: stages run through sub-skills such as gongwen, gongtan, gongxu, gongchan, and gongsheji, reducing premature specs and missed requirement checks.
• Engineering guardrails: gongyou checks change impact, gongshu grades database risk and requires backup/audit, and gongmi blocks plaintext credentials before they move into config.yaml or environment variables.
• Output standards: documents default to .docx; diagrams prefer PlantUML rendered to SVG so source remains readable; Office files use tools/excel, tools/ppt, tools/docx, and tools/pdf for deeper editing.
• Hard rules: no unanchored global replacement, no UPDATE/DELETE without WHERE, no unconfirmed git push -f, and changes must be committed, pushed, and synced to the project memory before the session ends.

Boundary And Notes

Best for teams with concrete engineering goals and a need to constrain AI behavior; not for pure demo snippets or open-ended chat. The real config.yaml is not part of the public package, so repo, SSH, database, and deployment settings must be injected locally.