Note
Looking for a ready-to-use Qlik plugin? A fully operational Qlik development toolkit (qlik-toolkit) is available from the Pupfish Claude Marketplace. 12 skills, 7 specialist agents, and a syntax-validation hook, installable in seconds.
This repo is different. It's an empty framework template, the architectural skeleton (orchestrator, agent and skill definitions, pipeline state schema, hook configurations) that you can use as a starting point for building your own full Qlik development pipeline with orchestrated Claude agents. Skill content and agent system prompts are placeholders here; the structure shows what belongs where.
A multi-agent development framework that accelerates end-to-end Qlik Sense application development using Claude Code, Anthropic's agentic coding tool.
This framework provides a complete .claude/ directory structure that turns Claude Code into a coordinated Qlik development pipeline. Seven specialized AI agents handle distinct phases of development, from requirements gathering through documentation, orchestrated by a central project manager that enforces quality gates and manages dependencies.
The framework produces comprehensive, domain-aware artifacts: specifications, data models, ETL scripts, expressions, visualization designs, and documentation. A Qlik developer provides architectural guidance and execution feedback at key checkpoints. The framework iterates to production-ready quality.
See CLAUDE.md for the full orchestration logic, pipeline gates, and context-passing strategy.
| Phase | Agent | Output |
|---|---|---|
| 0 | Requirements Analyst | Platform Context Document (brownfield ingestion) |
| 1 | Requirements Analyst | Project Specification Document |
| 2 | Source Profiler (skill) | Source Profile Report with architecture classification |
| 3 | Data Architect | Data Model Specification (app architecture, star schema, ETL design, cross-layer field mapping) |
| 4 | Script Developer | Complete .qvs script files with diagnostics |
| 5 | Expression Developer | Expression Catalog + runnable expression-variables.qvs |
| 6 | Viz Architect | Sheet specifications, master item definitions, manual build checklist |
| 7 | QA Reviewer | QA Review Report + data quality validation |
| 8 | Doc Writer | Full documentation set |
Brownfield-first. Most real projects inherit existing artifacts. The pipeline begins with context ingestion (existing apps, shared subroutine libraries, platform conventions, upstream data architecture) before requirements gathering.
Source architecture-aware. Different upstream architectures (Data Vault 2.0, star schema, normalized OLTP, flat files) require fundamentally different consumption patterns. The framework classifies source architecture and designs accordingly.
Human-in-loop execution validation. Script correctness, synthetic key detection, data quality discovery, and expression evaluation all require loading data in the Qlik engine. The pipeline includes explicit execution validation gates where a developer loads artifacts, reports results, and the framework iterates.
MCP-optional. The framework works without database connectivity. When MCP connections are available, source profiling and data quality validation are automated. Without them, the framework produces structured templates for manual completion.
13 domain knowledge modules loaded on-demand by agents:
Core: qlik-naming-conventions, qlik-data-modeling, qlik-load-script, qlik-expressions, qlik-security, qlik-performance, qlik-visualization
Workflow: source-profiler, platform-conventions, data-quality-validator, qlik-project-scaffold, qlik-review-checklist, qlik-deploy
Skills encode Qlik-specific patterns, anti-patterns, and best practices: cross-layer field naming strategies, dual-timestamp incremental loads for SCD2 satellites, null handling patterns, diagnostic query templates, and more.
The architectural skeleton is fully open: detailed orchestration rules, directory structure, agent definitions (with frontmatter specifying roles, tool access, and skill dependencies), skill definitions (with frontmatter and descriptions), hook configurations, orchestrator logic, and pipeline state schema. This gives you the complete blueprint for the multi-agent pipeline.
Agent system prompts and skill domain knowledge content have been developed separately and are not included in this release. Each file contains structural placeholders showing what belongs there. The framework is designed so that practitioners can author their own content using the architecture as a guide, or contact Pupfish Analytics to collaborate on complete implementation.
qlik-claude-framework/
├── CLAUDE.md # Orchestrator (pipeline management)
├── .claude/
│ ├── settings.json # Hook configurations
│ ├── agents/ # 7 specialized subagents
│ ├── skills/ # 13 domain knowledge modules
│ └── hooks/ # Validation scripts
├── inputs/ # User-provided source materials
│ ├── source-documentation/
│ ├── existing-apps/
│ ├── platform-libraries/
│ └── upstream-architecture/
└── artifacts/ # Framework-generated output
├── 04-scripts/
├── 07-qa-reports/
└── 08-documentation/
- Claude Code (CLI tool from Anthropic)
- Familiarity with Claude Code's subagent architecture, skills, and CLAUDE.md configuration
- A Qlik Sense environment (Cloud or client-managed) for execution validation
- Optionally: MCP database connections for automated source profiling
With agent system prompts and skill content in place, the workflow is:
- Clone this repo into your Qlik project directory
- Place source materials in
inputs/(existing apps, schema docs, upstream architecture docs) - Run Claude Code. The orchestrator reads
CLAUDE.mdand begins the pipeline - Provide business requirements conversationally during Phase 1
- Review and approve the Project Specification
- The framework produces data model, scripts, expressions, and viz specs
- Load artifacts in Qlik Sense at execution validation gates and report results
- The framework iterates until validation passes
- Complete documentation is generated
This framework changes what it means to do Qlik development. The developer's role shifts from writing load scripts and expressions to providing clear architectural specifications and validating execution results. That requires genuine expertise in data architecture, business requirements, and the Qlik associative engine. The mechanical work of translating architecture into code is what the agents handle.
This repo is released as a framework template. See CONTRIBUTING.md for details on forking, extending, and collaboration.
MIT License. Copyright (c) 2025 Pupfish, LLC. See LICENSE and NOTICE for details.
Created by Pupfish Analytics.
For assistance with implementation, customization, or enhancement, visit pupfish.io.