dreamlab-ai PRD to Complete Documentation
One command turns a PRD into a complete, cross-referenced documentation set. Optionally executes the build afterwards.
When not to use
- Existing documentation validation and alignment → use docs-alignment.
- Full code implementation with quality gates and testing → use build-with-quality.
- SPARC-phase orchestration across development stages → use sparc-methodology.
- LaTeX report generation with charts and bibliography → use report-builder.
- Simple feature without formal documentation → just write code directly.
What this does
You provide a PRD; you get:
- Specification docs — requirements, user stories, API contracts, security model.
- Domain model (DDD) — bounded contexts, aggregates, entities, events, DB schema.
- Architecture (ADR) — every architectural decision with rationale.
- Implementation plan — milestones, epics, tasks with dependencies.
- Unified INDEX.md — single entry point that ties everything together.
Usage
# Generate documentation only
/prd2build /path/to/your-prd.md
# Generate documentation AND execute build
/prd2build /path/to/your-prd.md --build
- Documentation only: wait 5-10 minutes; all docs land in
docs/. - With
--build: docs generate first, then a mesh swarm executes the complete build following all ADRs and DDD artifacts.
Input PRD
$ARGUMENTS
How it runs
The skill spawns a single concurrent batch of 8 documentation agents (foreground,
parallel) — researcher, ui-designer, code-analyzer, System Architect, SPARC
Coordinator, Implementation Planner, Test Strategist, and a Documentation Integrator
that runs last to stitch the unified INDEX.md. With --build, a 6-agent mesh swarm
then executes the build in the background.
Spawn the whole batch in one message and let the Task tool block until every agent completes — don't run them in waves or add per-phase checkpoints; docs don't need them.
Claude Code only: the 8-agent batch spawn uses the Task tool. On Codex / GPT-6 Astra: run the 8 agents' prompts sequentially in one session instead — there is no equivalent built-in subagent-batch tool.
The full agent prompts and the exact execution/build batch → references/agent-prompts.md.
Output & next steps
After execution you get docs/README.md (navigation) and
docs/implementation/INDEX.md (the single source of truth for implementation), plus
the full specification/, ddd/, adr/, sparc/, implementation/, testing/,
and design/ trees.
Read docs/README.md, review docs/implementation/INDEX.md, then start building.
The complete output tree, INDEX.md contents, minimum artifact bars, customization
(quality-bar env vars, update mode), and a worked example → references/outputs-and-index.md.
Quality bars (auto-validated)
8+ specification files · 27+ ADRs · 11+ DDD files · 3+ bounded contexts · 5+ aggregates
· 20+ tasks. Adjust via PRD2BUILD_MIN_* env vars (see the outputs reference). If the
PRD is too vague, agents make explicit assumptions and document them.