Business Blueprint Skill#

Use the Python scripts in this repository as the execution surface.

Output Directory#

All generated files (blueprint JSON, viewers, exports) go into projects/workspace/ — not the repository root.

bash

python scripts/business_blueprint/cli.py --plan projects/workspace/solution.blueprint.json --from "..."
python scripts/business_blueprint/cli.py --project projects/workspace/solution.blueprint.json
python scripts/business_blueprint/cli.py --export projects/workspace/solution.blueprint.json

Industry Selection#

Choose --industry from exactly one of: "common", "finance", "manufacturing", "retail". Select the closest match based on the user's domain and materials; do not invent other values.

IndustryHints contentcommonNo hints — generic domainsfinanceRisk control, credit, compliance, customer profile, etc.manufacturingProduction planning, quality, warehouse, supply chain, etc.retailStore operations, membership, POS, order fulfillment, etc.

How to Generate a Blueprint#

The AI agent is responsible for entity extraction. The Python tool handles JSON writing, visualization, and export.

Step 1: Read industry hints#

Read the seed template at business_blueprint/templates/{industry}/seed.json and get the industryHints.checklist.

Step 2: Extract entities from source text#

Using the user's source material AND the industry hints checklist, extract:

• capabilities: business capability areas (name, description)

• actors: roles/people involved (name)

• flowSteps: business process steps (name, actorId, capabilityIds, stepType)

• systems: IT systems that support capabilities

  • See references/entities-schema.md for all entity field definitions
  • See references/systems-schema.md for systems category/layer rules
  • See scripts/business_blueprint/templates/common/seed.json for field examples

Domain-knowledge mode (when seed has meta.blueprintType: "domain-knowledge", e.g. cross-border-ecommerce): entities go into library.knowledge.* (painPoints / strategies / rules / metrics / practices / pitfalls), not the architecture buckets above. The seed's industryHints.knowledgeHints.namingHints section, when present, dictates content granularity:

• strategy.name must be a 4-10 字 product-grade noun phrase (e.g. "统一归因模型", "AIGC 素材工厂"), not an action ("优化素材") or a dimension ("测款节奏"). Reference the seed's strategy_named_examples.
• painPoint.audience and strategy.audience are free-string fields tagging the primary persona ("品牌方/DTC", "平台卖家", "代运营/服务商"). Multiple values comma-separated. No standalone persona entity is required.
• metric.forecast: {direction, magnitude, unit} is optional and used for commitments to the customer ("ROAS 提升 25%"). Keep value/ benchmarkContext for current baseline; the two coexist as "now X, can reach Y".

These fields are pass-through under v2 minimal-validation: the validator does not enforce them, but renderers and pitch flows consume them.

Step 3: Write the blueprint JSON#

Write the JSON file directly to the output path. Use this schema:

json

{
  "version": "1.0",
  "meta": {
    "title": "...",
    "industry": "retail",
    "revisionId": "rev-YYYYMMDD-NN",
    "parentRevisionId": null,
    "lastModifiedAt": "ISO8601",
    "lastModifiedBy": "ai"
  },
  "context": {
    "goals": [],
    "scope": [],
    "assumptions": [],
    "constraints": [],
    "sourceRefs": [{"type": "inline-text", "excerpt": "..."}],
    "clarifyRequests": [],
    "clarifications": []
  },
  "library": {
    "capabilities": [
      {"id": "cap-xxx", "name": "...", "level": 1, "description": "...", "ownerActorIds": [], "supportingSystemIds": []}
    ],
    "actors": [
      {"id": "actor-xxx", "name": "..."}
    ],
    "flowSteps": [
      {"id": "flow-xxx", "name": "...", "actorId": "actor-xxx", "capabilityIds": ["cap-xxx"], "systemIds": [], "stepType": "task", "inputRefs": [], "outputRefs": []}
    ],
    "systems": [
      {"id": "sys-xxx", "kind": "system", "name": "...", "aliases": [], "description": "...", "resolution": {"status": "canonical", "canonicalName": "..."}, "capabilityIds": ["cap-xxx"]}
    ]
  },
  "relations": [
    {"id": "rel-xxx", "type": "supports", "from": "sys-xxx", "to": "cap-xxx", "label": "支撑"}
  ],
  "views": [],
  "editor": {"fieldLocks": {}, "theme": "enterprise-default"},
  "artifacts": {}
}

Step 4: Generate visualizations#

bash

python scripts/business_blueprint/cli.py --export <blueprint.json>

This generates SVG + HTML viewer by default. Use --format drawio|excalidraw|mermaid for other formats.

Export View Selection Policy#

Treat export view choice as a routing decision, not a styling preference.

• If a request matches a supported, standard export template, use that template.
• If there is no standard export template for the requested diagram, fall back to freeflow.
• Do not substitute swimlane, matrix, product tree, or other generic views just because they are available.
• When embedding a blueprint diagram into a report or ad hoc analysis, freeflow is the safe default unless the user explicitly asks for a supported standard template.

Step 5: Generate downstream projection#

bash

python scripts/business_blueprint/cli.py --project <blueprint.json>

This generates solution.projection.json, the canonical machine projection consumed by downstream report/slide workflows.

Workflow Decision Tree#

text

User provides raw requirements / meeting notes?
  → AI agent reads hints, extracts entities, writes blueprint JSON
  → Optionally run --project for downstream machine handoff
  → Then run --export for visualization

User needs diagram files (SVG, draw.io, etc.)?
  → --export (default: SVG + HTML viewer)

User unsure about blueprint quality?
  → --validate

User wants downstream report / slide generation?
  → --project

Commands#

CommandDescription--plan <path> --from <text>Generate empty blueprint JSON from source text (AI should prefer writing JSON directly)--project <path>Generate canonical projection JSON for downstream skills--export <path>Export SVG + HTML viewer (default), or use --format for other formats--validate <path>Validate a blueprint and print JSON results

Execution: Run directly as scripts:

bash

python scripts/business_blueprint/cli.py --plan ...
python scripts/business_blueprint/cli.py --export ...

Export Formats#

FormatFileUse Casesvg (default)solution.exports/solution.svg + HTML viewerQuick preview, embeddingdrawio``solution.exports/solution.drawioEditable diagramsexcalidraw``solution.exports/solution.excalidrawWhiteboard-style diagramsmermaid``solution.exports/solution.mermaid.mdGitHub-native rendering

Collaboration Boundary#

This skill produces semantic intermediate artifacts. Downstream skills consume them:

• report-creator consumes solution.projection.json → assembles reports
• slide-creator consumes solution.projection.json → assembles presentations
• Other skills may consume relations → generate PlantUML or other diagram syntax
• Downstream skills should never directly edit solution.blueprint.json
• solution.handoff.json is viewer-only metadata, not a downstream narrative input

Sandbox Execution#

When running in an isolated Python sandbox (Jupyter, notebook, cloud REPL) that auto-installs dependencies:

1. The sandbox uses scripts directly. Run execution scripts from the repository root: python

   import subprocess
   subprocess.run(["python", "scripts/business_blueprint/cli.py", "--export", "solution.blueprint.json"])
   
   

   • sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) — will raise NameError
   • subprocess.run(["business-blueprint", ...]) — sandbox runs Python cells, not shell
   • os.system() — same reason

Architecture Diagram Generation#

When user requests an architecture diagram (keywords: "架构图", "architecture diagram", "--export", "diagram"):

1. Read references/architecture-design-system.md for the complete design system.

2. Read the appropriate template from references/architecture-templates/ based on the user's domain:

   • AWS/Serverless/Lambda → serverless.md
   • Microservices/Kubernetes/微服务 → microservices.md
   • Other → use serverless.md as a structural reference

3. Read the blueprint JSON to extract entities and flow steps.

4. Generate a self-contained HTML file with inline SVG following the design system rules.

5. Write the output file to the same directory as the blueprint JSON.

If the request does not match one of the supported standard templates above, stay on the default freeflow export path. Do not switch to another generic view type as a fallback. If a standard template would create a squeezed, clipped, or overcrowded diagram, stop using the fixed template geometry and fall back to freeflow or a wrapped multi-row layout.

Route eligibility matrix#

Use an explicit route contract before rendering:

RouteStructural prerequisitesFirst fallbackTerminal behaviorfreeflowAny valid blueprint with at least one renderable node or relationNoneIf integrity still fails, export exits non-zero with a structural diagnostics payloadarchitecture-templateRecognizable L→R architecture shape, categorized systems, limited per-layer density, and no route-breaking overflow riskfreeflowSame as aboveposterClear layer/group structure with bounded peer density per row or wrapped-row supportwrapped poster or freeflowSame as aboveswimlaneActor-owned flow steps with meaningful lane groupingfreeflowSame as abovehierarchyStable tree/group relationship with low ambiguity in parent-child groupingfreeflowSame as aboveevolutionOrdered chronological or staged progression datafreeflowSame as above

Do not invent route heuristics ad hoc inside a renderer. Route eligibility must stay explicit and reviewable.

Generation Rules#

• Use dark mode by default ( #020617 bg + 40px grid). Only use light mode when the user explicitly asks for it.
• L→R data flow: Clients(左) → Frontend → Backend → Database(右)
• Map systems[].category to semantic colors from the design system
• Map systems[].properties.type == "aws" → AWS Region boundary box
• Map systems[].properties.type == "k8s" → Kubernetes Cluster boundary box
• Use flowSteps[].seqIndex for L→R ordering
• Component sizing: 0-1 cap = small(44px h), 2-4 = medium(80px h), 5+ = large(80px h)
• Layout must be content-driven. Never force every node in a layer into one fixed row if that creates toothpaste-style squeezing.
• When a layer has more than 3 peer nodes, or labels/features become tight, wrap into multiple rows or widen the canvas before shrinking the content.
• Render users/actors as actor labels, badges, or lane headers by default. Do not render them as ordinary system cards unless the user explicitly asks for that visual treatment.
• Legend must live in a bottom safe area and participate in canvas sizing. Never place the legend as a floating overlay in the top-right corner.
• Final SVG/HTML height must be derived from the bottom-most node, legend, summary cards, and footer plus padding. Do not use fixed-height wrappers or overflow: hidden that can clip the last row.
• Z-order: bg → grid → title → region → arrows → nodes → legend → cards → footer
• Component border: rx="8", stroke-width="2"
• Region border: rx="16", stroke-dasharray="8,4", opacity="0.4"
• Geometry-sensitive integrity checks must use the numeric thresholds from evals/export-integrity-thresholds.json, not prose heuristics.

Output#

• Single HTML file: {blueprint_stem}.html alongside the blueprint JSON
• No external dependencies (except Google Fonts CDN for JetBrains Mono)
• Opens in any browser, printable to PDF

Error Handling#

• If --validate returns errors: fix structural issues before proceeding to --export.
• If --validate returns only warnings: proceed but note the warnings in any handoff.
• If Python version < 3.12: the package will refuse to install. Use python3 -m business_blueprint.cli with system Python as fallback.
• If a specialized route fails integrity: fall back to its configured fallback route.
• If freeflow also fails integrity: export exits non-zero with a structural diagnostics payload instead of emitting a silently broken artifact.

Cross-Platform Scope#

Phase 2 does not attempt full Windows terminal parity.

Known deferred cases:

• PowerShell pipe quirks beyond documented CLI contract tests
• console-default encoding issues outside explicit UTF-8 execution paths

Accepted workaround for encoding-sensitive runs:

Use scripts directly (pure Skill, no package structure)#

subprocess.run(["python", "scripts/business_blueprint/cli.py", "--export", str(blueprint_path)])

• set PYTHONIOENCODING=utf-8 where needed