The problem
A single "do everything" coding agent quickly loses the plot on real features — it conflates requirements, design, implementation, and testing into one fuzzy prompt and produces brittle results.
What I built
Agent SDLC breaks the software lifecycle into specialised agents that each own one role and hand off to the next:
- Business Analyst agent — clarifies and structures requirements.
- Designer agent — turns requirements into a concrete plan and interface.
- Developer agent — implements against that plan.
- Tester agent — verifies the result and feeds issues back.
Architecture
- Orchestrated on the GoClaw agentic framework, with explicit handoffs and shared state between roles.
- Each role has a focused prompt, tools, and success criteria — making the whole pipeline inspectable and improvable role-by-role.
Outcome
A division-of-labour approach to agentic development that produces more reliable output than a monolithic agent, because each step is scoped, verifiable, and replaceable.
What you get
If you want to apply multi-agent orchestration to a workflow (not just coding), I can decompose it into roles, define the handoffs, and build the orchestration graph.