The greenfield agents¶
Five agents make up the Code New Project Agents Team. The orchestrator (/reversa-new) drives the other four in a fixed sequence. Each agent reads what the previous one produced and adds its own artifact.
Pipeline¶
There is a CONTINUAR checkpoint between agents. The orchestrator never advances on its own.
1. Reversa New (orchestrator)¶
Command: /reversa-new
Reads the initial brief (passed inline or asked interactively), saves _reversa_sdd/newproject-brief.md, walks the four functional agents in fixed order, and writes a checkpoint in state.json#newproject_progress after each stage.
Detects re-execution: if a pipeline is already in progress, asks whether to continue, recreate or re-run from a specific agent.
Produces: _reversa_sdd/newproject-brief.md and the orchestrator state under state.json#newproject_progress.
2. Ideator¶
Command: /reversa-ideator
Structured brainstorm with six divergent questions: root problem, value delivered, alternatives, raw audience, success metrics, dangerous assumptions. Asks one question at a time (when the engine does not support batching), waits for the answer before moving on, and never collapses the questions into a single multi-shot prompt.
Produces: _reversa_sdd/ideation.md.
3. Researcher¶
Command: /reversa-researcher
Turns the raw audience from ideation.md into one to three structured personas with journeys (entry, friction, outcome). The user chooses the number of personas; the agent only suggests based on the breadth of the audience description.
Produces: _reversa_sdd/personas.md.
4. Drafter¶
Command: /reversa-drafter
Synthesizes ideation and personas into a complete PRD: problem, success metrics, scope, non-goals, constraints, risks, open questions. Works as a synthesizer, not an interviewer: extracts everything it can from the two sources and asks at most two coverage questions to fill the most critical gaps. Anything that remains undefined is marked 🟡 [INDEFINIDO, validar com usuário].
Produces: _reversa_sdd/prd.md.
5. Spec SDD¶
Command: /reversa-spec-sdd
Decomposes the PRD into logical components and writes one SDD spec per component, with an automatic quality score (0 to 100) and gap analysis. The methodology is Pragmatic RFC plus LLM-First: structured like an RFC (Problem / Goals / Design / Edge Cases) but optimized to be consumed by humans and AI agents alike.
This agent is a vendored version of the global sdd-spec skill: it lives natively inside Reversa, reads prd.md as the primary source, writes into _reversa_sdd/sdd/, marks every spec with the 🟡 (planned) seal, and on completion hands off to /reversa-forward.
Can also be used standalone: evaluating an existing spec, or generating a single spec from any input the user provides.
Produces: _reversa_sdd/sdd/<component>.md (one per component).
Running manually¶
You rarely need to call an isolated agent. /reversa-new orchestrates everything. But if an agent failed or you want to redo a stage:
/reversa-new # detects in-progress pipeline, offers Continue / Recreate / Re-run from
/reversa-ideator # standalone, reads newproject-brief.md
/reversa-researcher # standalone, reads ideation.md
/reversa-drafter # standalone, reads ideation.md + personas.md
/reversa-spec-sdd # standalone, reads prd.md or any source the user passes
Each standalone agent verifies its own preconditions and aborts with a clear message pointing at the missing artifact.