ProductDesign package
From decision to build,without starting over.
The design package is how a decision reaches the code. It starts with a 26-section implementation spec generated from the concept version, its evidence, evaluations, and experiment results, then adds the published brand kit and the confirmed screens with their specs. The Builder writes the spec into docs/handoff and the brand kit and screens into docs/design, and the spec still exports as a ZIP for any coding agent.
- across six groups, plus README
- 26 sections
- completeness, consistency, citations
- 3 validators
- to one immutable concept version
- Pinned
- imported with the brand kit and screens
- Builder-ready
How it works
Generated from state,not from a chat log.
- 01
Pin the inputs
Choose a concept version. The bundle snapshots its evidence, evaluation, and experiment state at that moment.
- 02
Generate section by section
Each file has one purpose and is written from structured records, not from earlier generated prose.
- 03
Validate before export
Completeness checks required content, consistency looks for contradictions between sections, and a citation guard checks every reference.
- 04
Build from it
Open it in the Builder with the brand kit and confirmed screens, or download the ZIP for your own repository and coding agent.
Provenance
Every claim resolves.Every assumption says so.
Statements about the market or a platform cite claim IDs that resolve to SOURCES.md and the stored source behind them. Anything unsupported is labeled as an assumption, and choices are recorded in DECISIONS.md, so an agent can tell a requirement from a suggestion.
- README with provenance for the whole bundle
- SOURCES.md maps each claim ID to its source and locator
- DECISIONS.md records choices and human overrides
- Sections that do not apply say why, instead of disappearing
Capabilities
Everything an agent needs,nothing it has to guess.
Six groups, 26 sections
Product, UX, engineering, AI, quality, and execution, from the vision to implementation-agent instructions.
Completeness validator
Confirms the bundle names the target user, job, P0 scope, data model, acceptance criteria, phases, and open risks.
Consistency validator
Catches contradictions, such as an endpoint using an entity missing from the data model or a screen that relies on a non-goal.
Citation guard
Every citation is checked against the stored claims before export, so no reference points nowhere.
Regenerate one section
Rework a single file as a new revision. The rest of the bundle stays exactly as reviewed.
Brand and screens included
The Builder imports DESIGN.md, tokens, logo files, and fonts, plus each confirmed screen and its spec, into docs/design, and installs favicons and app icons.
What you getA spec an agent can execute.
The package closes the gap between research and engineering. The AI group appears only when the concept uses AI, and validation briefs and PRDs come from the same state.
- Vision, scope, requirements, and success metrics
- Flows, screens, states, and accessibility
- Architecture, schema, API contracts, and jobs
- AI behavior, prompt contracts, and an eval plan
- Test plan, acceptance criteria, and launch checklist
- A build plan and implementation-agent instructions
Hand off the decision,not just the idea.
Generate the package for your concept, then watch the Builder build against it, or open the ZIP in your own editor.