Convector Cookbook
End-to-end testable recipes for all Vandoko pipelines — team onboarding guide
Convector Cookbook
Version 0.0.1 — Initial release for team onboarding
Welcome to the Convector Cookbook. This is your starting point for learning and running every Vandoko pipeline — from client onboarding through component registry publication.
Each recipe is a testable, end-to-end walkthrough with exact commands, expected outputs, verification steps, and troubleshooting tables. Follow them sequentially for your first run, then use them as reference for production work.
New Client vs Existing Client
When /go-vandoko is invoked, the orchestrator runs a branching check at initialization. The path depends on whether prior client data exists.
Pipeline Initialization
The orchestrator checks for prior state:
No prior data found. The pipeline starts fresh:
- Generate a new
pipelineId(UUID) - Create
vandoko-pipeline-state.json - Delegate to client-interview skill
- Proceed through all 10 stages sequentially
Prior ClientProfile found in project memory. The system prompts:
Prior client data found for [Client Name] (completed [date]).
Reuse? (yes / re-interview)- "yes" → Skip INTAKE interview, load existing ClientProfile, skip to DISCOVERY (or first incomplete stage)
- "re-interview" → Clear prior state keys, run fresh interview from Q1
Interview Phase
The system asks Q1–Q5 (fast path), then offers extended questions Q6–Q20. Every question must be answered or explicitly skipped. There is no prior data to draw from.
If resume_context is provided to client-interview, it returns the existing
InterviewResult immediately without asking any questions. The interview is
entirely skipped.
Pipeline Execution
- Full INTAKE scrape (Playwright, up to 4 pages) - Full DISCOVERY research (Exa, Gemini, brand gap analysis) - All 4 decision points active — nothing pre-approved - All artifacts created fresh - Full concept gallery (3 concepts for growth, 5 for premium) - Project lineage initialized from empty
- INTAKE may be skipped (valid ClientProfile reused) - DISCOVERY partially
cached (new Exa searches still run for market changes) - BrandStrategy may
carry forward (skip Decision Point 1 if previously approved) - Resume from any
stage via
--resumeflag - Artifacts layer onto existing ones without regeneration - Lineage continues — new entries appended
Comparison at a Glance
| Aspect | New Client | Existing Client |
|---|---|---|
| First prompt | "What is the URL of the business?" (Q1) | "Prior client data found. Reuse? (yes/re-interview)" |
| Interview duration | 5–20 questions (2–10 min) | 0 questions if reused; full if re-interview |
| INTAKE stage | Full Playwright scrape (~2–5 min) | Skipped if ClientProfile reused |
| First decision point | Decision 1: Strategy approval | May start at Decision 2+ if strategy approved previously |
| Total pipeline time | 30–90 min depending on tier | 10–60 min depending on what is reused |
Recipes
Core Pipelines
New Client — Full Pipeline
Complete go-vandoko cold start with Meridian Robotics (growth tier). 10 stages, 4 decision points, full artifact generation.
Existing Client — Resume/Expand
Warm-start pipeline for returning clients. Skip interview, reuse artifacts, upgrade tier, add deliverables.
ABM Pipeline — Full Enrichment
Account-based marketing pipeline with NovaByte Systems. 6 phases from company enrichment to campaign creation.
QA Audit — Live Site
7-dimension quality audit against a live site. Visual, SEO, accessibility, brand, conversion, content, and production readiness.
Vandoko Properties
vandoko.com — Parent Brand
Build the Vandoko parent company site. Self-referential pipeline using canonical brand tokens and vandoko-theme.
vandoko.agency — Agency Site
Agency showcase that demonstrates the pipeline as a product. Portfolio from client work, QA auditing its own site.
vandoko.ai — Intelligence Platform
SaaS product site for the market intelligence platform. ABM data feeds the product, DQP corpus powers design quality.
Registry Pipelines
HTML → Registry Ingestion
Convert raw HTML to a @vandoko registry component. 5-phase pipeline: extract, analyze, convert, validate, publish.
Stitch → Registry Pipeline
Generate UI in Gemini Stitch, export HTML, feed through html-to-registry, publish to the component registry.
Client Site Component Install
Install @vandoko components with client-specific brand tokens. brand-config → design tokens → registry install.
Live Telemetry Instrument Landing
A demonstrative instrument-panel landing composed from existing vkb blocks, led by the live 3D Pylon Terrain hero with an analytics data card + oversized live figures dataviz spine.
New Client — Full Pipeline
Complete go-vandoko cold start walkthrough with Meridian Robotics at growth tier