| name | blueprint |
| description | Convert a research synthesis report into an implementation-neutral product blueprint for product concept, users, workflow, product experience direction, logical architecture, MVP, evaluation, handoff, and next-stage routing. Use when the user has a research report and asks to "Create a blueprint from this research report", "Design the product based on this research", "Turn this research into a product design", "What should we build?", "Generate a product blueprint", "Convert the research to an MVP plan", "Define the product experience direction", "Define how users should experience the product", "Define user trust/control/review expectations", "What should we build next?", "Which design stages should we run next?", or "Recommend the next stages after the blueprint"; or after research-pipeline handover. Aliases: "product-blueprint", "research-blueprint", "research-to-product". Do NOT use for more research, tech stack, detailed UI/UX design or wireframes, user-story requirements, or single-paper explanation.
|
| license | MIT |
Research-to-Product Blueprint
This skill is the second stage of a research-driven development workflow.
It consumes a research synthesis report and produces a product-level
blueprint. Its job is not to summarise papers again — it converts
research findings into product-design language.
Research mechanisms / confidence-graded findings / gap classifications
→ product primitives → core capabilities → workflows
→ product experience direction → logical architecture
→ MVP boundary → evaluation strategy → technical-design handoff
→ recommended next stages
The blueprint is also the first adaptive stage-gate router after research
extraction: §19 recommends which downstream stages (architecture-design,
tech-stack-selection, ux-design, security-review, test-design, …) should run
next using RUN / SKIP / DEFER / ASK_USER — evidence-based, confidence-scored,
and overrideable. It recommends a path; it never silently expands the pipeline.
When To Trigger
- "Create a blueprint from this research report"
- "Design the product based on this research"
- "Turn this research into a product design" / "What should we build?"
- "Generate a product blueprint" / "Convert the research to an MVP plan"
- "Define the product experience direction" / "Define how users should
experience the product" / "Define user trust/control/review expectations"
- "What should we build next?" / "Which design stages should we run next?" /
"Recommend the next stages after the blueprint"
- The aliases
product-blueprint, research-blueprint, research-to-product
- Accepting the
research-pipeline handover offer at the end of its
iterative gap-closure loop (see that skill's
references/iterative-synthesis.md, "System-Design Handover")
- A
<topic-slug>-research-report.md exists and the user's intent is
product design, not more research.
Do not trigger for:
- More research (use
research-pipeline)
- Technical architecture or tech-stack selection (a later tech-arch skill)
- Detailed UI/UX design: drawing wireframes, writing screen copy, defining
exact CLI flags or MCP tool schemas, building frontend components (a later
UX-design skill). The blueprint defines UX intent, not detailed UX.
- Requirements clarification with user stories (use
req-analysis)
- Single-paper explanation (use
paper-analyzer)
Inputs
Required:
- One research synthesis report:
<topic-slug>-research-report.md
(produced by research-pipeline or a compatible research workflow).
Optional:
gaps.json — structured ACADEMIC / ENGINEERING / OUT_OF_SCOPE
classification with priorities
runs/<run_id>/summarize/synthesis_report.json — machine-readable
findings and gap objects
runs/<run_id>/screen/screened.jsonl — paper metadata for citations
runs/<run_id>/plan/query_plan.json — original topic framing
project_name, target_domain, target_users, product_direction
constraints, excluded_scopes
mvp_bias: small / balanced / ambitious
risk_tolerance: low / medium / high
output_language: english / chinese / bilingual
output_detail: concise / standard / detailed (default: standard)
The Markdown report is authoritative. JSON artifacts are
supplementary only. If JSON conflicts with the Markdown report, prefer the
Markdown report and note the conflict in §2 (Source Research
Interpretation). If JSON carries structured detail missing from the
Markdown report, use it but mark it as supplementary.
Output
One Markdown document written to the current working directory:
<topic-slug>-product-blueprint.md
If <topic-slug> cannot be inferred, derive it from the source report
filename; if no filename is available, derive a lowercase hyphenated slug
from the product thesis. If files cannot be written, emit the full
Markdown blueprint inline and state the recommended filename.
The blueprint must contain these 20 sections, in order:
- Executive Product Thesis
- Source Research Interpretation
- Target Users and System Actors
- Product Goals and Non-Goals
- Research-to-Product Translation Map
- Adopt / Adapt / Merge / Defer / Reject Decisions
- Core Product Capabilities
- Workflow Model
- Product Experience Direction
- Logical Architecture
- Conceptual Information Model
- Decision Policies
- Risk, Governance, and Safety Model
- Evaluation Strategy
- MVP Scope
- Roadmap and Future Extensions
- Open Questions and Validation Plan
- Handoff Notes for Technical Design
- Recommended Next Stages
- Traceability Appendix
Use templates/product_blueprint_template.md as the skeleton.
Formatting requirements:
- A
## Contents section at the top with internal Markdown links to all
20 sections.
- At least one Mermaid diagram for the main end-to-end workflow.
- At least one Mermaid diagram for the logical architecture.
- Additional Mermaid workflow diagrams only for complex, safety-critical,
or high-risk workflows.
- Markdown tables for translation maps, decisions, risks, evaluations,
and policies.
- Citations as
[arxiv_id] (e.g. [2312.01234]) or [Author, Year]
(e.g. [Park et al., 2023]), traceable to the source report's
## References.
Non-Goals
Do not select: a programming language, framework, database, vector
database, cloud provider, deployment platform, UI library, or concrete
package/module structure.
Do not produce: code, a database schema, a deployment plan,
vendor-specific architecture, implementation tickets, or detailed coding
tasks.
Do not produce detailed UX design: screen layouts, wireframes, visual
hierarchy, exact CLI command syntax/flags, exact MCP tool or API schemas,
copywriting, or full accessibility checklists. §9 captures UX intent only.
The boundary is: blueprint defines UX intent · architecture defines the
UX-enabling technical structure · a later UX-design skill defines the detailed
experience · implementation-plan turns it into build tasks. See
references/product-experience-direction.md.
Do not treat research gaps as solved without explicit validation.
When uncertain whether a statement is implementation-neutral, apply the
decision rule in references/borderline-cases.md: if removing the
specific term and replacing it with its purpose still conveys the
constraint, keep the conceptual version; if the statement cannot be
expressed without naming a product, vendor, or deployment model, it
belongs to the later technical-design skill.
Method
Follow the staged method. The prompt files under prompts/ carry
the detailed instructions for each stage; the manifest.json records the
task graph, gate names, and failure policy.
- Read & detect — read the source report and any supplementary
artifacts. Detect available sections and the citation style.
- Classify input quality — strong / usable / weak / insufficient
(see
references/input-mapping.md). If insufficient, STOP and emit
the standardized insufficient-input failure (do not fabricate a
blueprint). If weak, proceed but mark missing areas as assumptions
or open questions. Use prompts/00_assess_input_quality.md.
- Resolve domain — if the report spans multiple unrelated product
domains, scope the blueprint to one. Proceed with the highest-coverage
domain as a documented default; ask the user only when domains have
similar evidence coverage and would yield materially different theses.
- Extract & classify — extract mechanisms, methods, patterns,
benchmarks, assumptions, contradictions, gaps, risks, and architecture
hints. Tag each with a type and confidence grade (HIGH 🟢 / MEDIUM 🟡 /
LOW 🔴). Use
prompts/01_extract_research_items.md.
- Apply gap mapping —
ENGINEERING gaps → product requirements;
ACADEMIC gaps → validation requirements / open questions (never an
MVP requirement unless the product's purpose is to answer the research
question); OUT_OF_SCOPE → non-goal. See references/gap-type-mapping.md.
- Translate — convert research items into product primitives
(capability, workflow, policy, conceptual component, information
object, evaluation requirement, governance rule, risk control, user
interaction, lifecycle state, integration surface). Merge overlapping
primitives. Use
prompts/02_translate_to_product_primitives.md.
- Resolve ideas — decide ADOPT / ADAPT / MERGE / DEFER / REJECT for
every major idea, conservatively. Use
prompts/03_resolve_ideas.md.
- Size the MVP — apply the Round History signal (a long round history
with many remaining gaps, or
Readiness: HAS_GAPS, means a speculative
space → heavier DEFER/REJECT pressure). Structure §15 as MVP-0 (smallest
demonstrable end-to-end slice) / MVP-1 (first usable version) / Safety
Baseline / Evaluation Baseline / Deferred; keep MVP-0 minimal and do not
translate research completeness into MVP inclusion.
- Compose — generate the thesis, then all 20 sections with the
required Mermaid coverage and tables, including §9 Product Experience
Direction (UX intent only; see
references/product-experience-direction.md)
and §19 Recommended Next Stages (adaptive stage-gate routing with
RUN / SKIP / DEFER / ASK_USER decisions; see
references/adaptive-stage-gate-routing.md).
Lead the thesis with the primary
research-backed architecture, not a conditional/secondary mechanism.
Copy metadata (never invent; keep pipeline runs vs. gap-closure rounds
separate; skill version from manifest.json or unknown). Keep primary
actors/domains aligned with the thesis. Never leave a citation blank —
gap-derived items cite [Source Report: Research Gaps — <name>]. A
release gate from a MEDIUM/LOW-confidence mechanism needs a HIGH-risk
why-now justification or is downgraded. The Contents lists every numbered
section and every appendix present; keep the main body scannable (move
large tables to appendices). Respect the output_detail length budget.
Use prompts/04_generate_blueprint.md and the templates.
- Quality gates, self-repair & self-check — run the gates in
prompts/05_quality_gate.md, revise failing sections, and apply safe
wording rewrites before delivery (detect → repair → re-check) so only
judgement-needing warnings remain. Emit the actionable ## Appendix A
self-check (gate · status · finding · required action · blocks-TD)
reflecting the post-repair document. Maximum 3 revision attempts; after
3 failures, surface the specific failing gates to the user and stop. Do
not deliver an unvalidated blueprint.
Decision Categories
- ADOPT — use directly (HIGH confidence, product-critical).
- ADAPT — use with modification (MEDIUM confidence or partial fit).
- MERGE — combine related ideas into one capability.
- DEFER — valuable but not MVP-critical, or MEDIUM/LOW confidence.
- REJECT — weak evidence, out of scope, unsafe, or incompatible with
the product thesis.
- DEFER / VALIDATE — unresolved
ACADEMIC gaps that may be valuable
but are not ready to become product requirements.
Quality Gates
Pass only if ALL hold (full text in prompts/05_quality_gate.md and
tests/expected_sections_checklist.md):
- Input understanding, metadata & thesis emphasis — names the source
research question, acknowledges input quality and the targeted domain
(if multi-domain), uses copied-not-invented metadata (skill version from
manifest.json or unknown; pipeline runs and gap-closure rounds kept
separate), and leads the thesis with the primary research-backed
architecture (not a conditional/secondary mechanism).
- Traceability — every major capability traces to a research
citation (
[arxiv_id] / [Author, Year]) or a constrained explicit
design decision with rationale; otherwise mark it "Design hypothesis —
requires validation." No citation cell is blank — gap-derived items cite
[Source Report: Research Gaps — <name>].
- Implementation neutrality — no programming language, framework,
database, cloud provider, vendor, package structure, deployment
commands, code, or tickets.
- Workflow completeness — each major workflow has a trigger, inputs,
decision gates, steps (or a Mermaid flow), outputs, failure modes, and
success criteria.
- Scope control & MVP discipline — primary actors/domains match the
thesis (high-stakes domains seen only as evidence stay Secondary/Future);
§15 splits the core path into MVP-0 (smallest demonstrable) and MVP-1
(first usable), separates Safety and Evaluation baselines, and has an
explicit success definition;
ACADEMIC-gap items are excluded unless the
product validates that gap. Flag (do not auto-fail) an MVP-0 over 6
capabilities without justification, a large Phase-1 system mislabelled
MVP-0, or standard output over budget; fail only if it is no longer a
small, testable core.
- Risk honesty & release-gate confidence — HIGH-impact risks are
explicit, mitigations are realistic (never "prompt the model better"),
safety-critical deferred items are release gates, and risks from
unvalidated
ACADEMIC items are flagged. A release gate derived from a
MEDIUM/LOW-confidence mechanism is justified only by HIGH risk impact +
no cheaper control + an explicit why-now; otherwise downgrade it.
- Downstream usefulness — a technical-design agent can choose a tech
stack and plan an implementation without re-reading the papers; the
Contents section exists with valid links and lists every numbered section
and every appendix present; the main end-to-end workflow and the logical
architecture each have a Mermaid diagram.
- Product Experience Gate — §9 names a primary user, job-to-be-done,
experience thesis, and primary interaction mode; classifies each interaction
mode (primary / secondary / wrapper-integration / future surface — "AI Skill"
disambiguated, not conflated with MCP); defines trust/control/transparency
needs, human-in-the-loop, and failure/recovery where relevant; and hands UX
assumptions off to architecture. It stays UX intent (no wireframes, screen
layout, exact CLI/MCP/API syntax, or copy). Full fail/warning conditions in
references/product-experience-direction.md.
- Adaptive Stage-Gate Recommendation Gate — §19 Recommended Next Stages
exists with a complexity assessment (labelled a routing heuristic, not a
formal estimate) and a stage-recommendation table with a
Depends On
column; every stage uses a controlled decision (RUN / SKIP / DEFER /
ASK_USER); each RUN has evidence, each SKIP a reason, each DEFER a revisit
trigger, and each ASK_USER the missing information; an ASK_USER rationale
explains any absence of ASK_USER; the recommended pipeline splits a
Recommended Linear Path from a Conditional Follow-up Gates table (so
deferred conditional stages are not dropped); architecture-design is
normally RUN and architecture-update / architecture-reconciliation default
to DEFER at blueprint stage; and §9 Product Experience Direction informs the
UX/security/test decisions. The skill recommends — it does not silently expand
the pipeline. Full conditions in
references/adaptive-stage-gate-routing.md.
Fail immediately if any tech-stack choice, code, or implementation ticket
appears; if either required Mermaid diagram is missing; if open research
gaps are silently treated as solved; if risks are omitted; if the
Recommended Next Stages section is missing or uses uncontrolled decision
wording; or if the output is an unstructured essay.
References
| File | Load when |
|---|
references/artifact-contract.md | The Cross-Skill Artifact Contract the blueprint output must satisfy |
references/input-mapping.md | Mapping report sections → blueprint inputs; quality thresholds |
references/gap-type-mapping.md | Turning ACADEMIC / ENGINEERING / OUT_OF_SCOPE gaps into product actions |
references/borderline-cases.md | Deciding whether a statement is implementation-neutral |
references/product-experience-direction.md | Writing §9 UX intent; the boundary rule; clarification questions; the Product Experience Gate |
references/adaptive-stage-gate-routing.md | Writing §19 Recommended Next Stages; decision vocabulary; per-stage rules; complexity scoring; the Adaptive Stage-Gate Recommendation Gate |
references/troubleshooting.md | Insufficient / weak / multi-domain inputs, missing sections, gate failures |
templates/product_blueprint_template.md | The 20-section output skeleton |
templates/workflow_template.md | Writing a workflow with full fields + Mermaid |
templates/logical_architecture_template.md | Writing §10 conceptually (not technically) |
templates/translation_map_template.md | Writing §5 / §6 tables |
templates/evaluation_strategy_template.md | Writing §14 scenarios |
Final Reminder
This skill bridges research synthesis and technical architecture.
It answers: "What product should exist, how should it behave, and what
workflow should it implement?"
It does not answer: "What tech stack should implement it?" — that is
the next skill's job.