One-Stop Code Change Workflow

Problem

Code changes often stall because requirements are vague, review happens after implementation, test coverage misses edge cases, and rollback context is lost. A single agent that codes and self-evaluates can easily miss affected files, failure paths, and missing test cases.

How It Works

dev-flow splits code changes into requirement confirmation, coding, checking, browser testing, and result output, and delegates critical judgments to separate sub-agents: Questioning Agent probes requirements and gaps, while Check Agent starts before coding and validates both code correctness and test plan completeness.
- After user confirmation, the workflow must continue without pausing.
- Check Agent must start before coding and receive the modified-file list via SendMessage.
- Every modification needs a corresponding test case. If agent-browser is available, browser testing is mandatory; if unavailable, the workflow must emit manual-test guidance and request confirmation.
- A test report is required, covering passed cases, missing cases, exception paths, and rollback notes.

Boundaries

It fits small-to-medium changes that need strict guardrails, especially frontend/backend features, UI interactions, and API work. It depends on AskUserQuestion, SendMessage, and agent-browser; without browser testing or user confirmation, it can only follow a fallback path. It does not replace architecture review, security audits, or CI quality gates.