create-meta-prompts

Multi-stage prompts for agent-to-agent workflows are created by structuring sequential research, planning, and implementation phases into an XML-rich prompt chain. This process generates a versioned SUMMARY.md and dependency-aware artifacts that downstream prompts can consume and extend automatically.

Chaining agent prompts with XML metadata requires saving each prompt in its own `.prompts/{number}-{topic}-{purpose}/` folder. This approach ensures clear provenance, enables automatic detection of existing outputs, and yields machine-friendly artifacts for downstream prompt consumption.

A ready-to-run plan for downstream prompts is generated by defining explicit dependencies and a versioned SUMMARY.md within a multi-stage prompt chain. This allows sequential agents to automatically detect and reference prior research and planning outputs to build dependency-aware execution chains.

Yes, this prompt engineering workflow supports automatic detection of existing prompts for chaining. It identifies prior research and plan outputs to build dependency-aware execution chains, ensuring each new prompt correctly references and extends the XML-rich artifacts generated by previous stages.

You can use Opencode to automate sequential research, planning, and implementation phases by describing your task to generate a multi-stage prompt chain. Opencode creates the prompt chain in the `.prompts` directory and guides execution to produce structured, machine-friendly outputs.