Qlik Sense App Analysis — Agent Skill

An AI Skill for exploring, analyzing, visualizing, and governing Qlik Sense apps and Qlik Cloud analytics content through the Qlik MCP server.

Point Claude/ ChatGPT at a Qlik Cloud tenant with the Qlik MCP server connected, and this skill teaches it how to find the right app, prefer governed definitions over ad hoc ones, avoid Qlik's specific silent-failure traps, and — when asked — build dashboards or author governed master items, glossary terms, and data products.

It works out of the box against any tenant. It's designed to get better the more you extend it with your own apps and domains — see Extending the skill below.

Why this exists

Anthropic's self-service analytics write-up frames most wrong answers a data agent gives as one of three failure modes: entity ambiguity (which table/field is "revenue"?), staleness (is this still true?), and retrieval failure (the right answer existed but the agent never found it). Their fix against a SQL warehouse is a stack you build by hand: canonical datasets, a semantic layer, lineage, and skills that route the agent to a governed answer before it writes any raw SQL.

Qlik MCP tools address the same three failure modes — but several of the layers the warehouse world has to build by hand already exist natively in Qlik Cloud:

Failure mode	Warehouse approach (build it yourself)	Qlik MCP approach (already there)
Entity ambiguity	Curate canonical dbt models, hope people don't duplicate them	Data Products (curated, documented datasets) and Master Items (governed dimensions/measures) — list before you build
Staleness	Freshness/completeness checks you write and maintain	qlik_get_dataset_freshness, qlik_get_dataset_trust_score, qlik_get_dataset_quality_computation_status are native tool calls
Retrieval failure	Distill a query corpus into reference docs by hand	qlik_search, qlik_list_sheets, qlik_get_sheet_details surface existing apps/charts directly — reuse before rebuilding
Lineage / "where did this come from"	Reconstruct the transformation graph from dbt manifests	qlik_get_lineage is a direct tool call (still one hop at a time — call it recursively)
Business glossary	An internal wiki or knowledge graph you pipe in yourself	A structured Business Glossary with a draft → verified → deprecated lifecycle, native to the platform
"Raw SQL" fallback	Hand-written SQL, reviewed adversarially before trusting it	qlik_create_data_object with a Qlik expression (set analysis) — Qlik performs the calculation, so the review is about expression correctness, not aggregation logic

And two things are genuinely new in the Qlik MCP world, with no warehouse-skill equivalent:

• Authoring, not just querying. The warehouse skill only reads. Qlik MCP tools can also create governed objects — master dimensions/measures, sheets, charts, data products, glossary terms — so this skill is part analyst, part steward, and every creation action carries a "does this already exist?" check.
• Stateful selections. SQL queries are stateless; Qlik selections persist across tool calls until cleared, and a selection on a value that doesn't exist fails silently. That's a class of silent failure SQL never has.

qlik-sense-app-analysis/SKILL.md lays out this mapping in full and turns it into operating rules.

Installing

The deployable skill is the qlik-sense-app-analysis/ folder in this repo (it contains SKILL.md and references/). Install it whichever way your AI client loads skills:

• Zip upload (claude.ai / Claude Code skill upload): download qlik-sense-app-analysis.zip from the Releases page, or rebuild it yourself — see Building the zip.
• Folder drop / plugin: copy the qlik-sense-app-analysis/ folder into your Claude client's skills directory, or add this repository as a plugin source.

Requires a Qlik MCP server connection with the tools listed in qlik-sense-app-analysis/references/qlik-mcp-tool-reference.md. The skill degrades gracefully if some tool categories aren't available (e.g. no Data Products or Business Glossary licensed), but app discovery and analysis tools are expected to be present.

Extending the skill

The generic version works immediately, but for a real deployment you'll usually want to point it at your own apps. Copy qlik-sense-app-analysis/references/app-profile-TEMPLATE.md to references/<name>.md per app or business domain and fill in the app ID, the master items and data products already available, and any field-naming quirks specific to that app. SKILL.md checks for a matching profile before falling back to live discovery — no other changes required. This is intentionally the same pattern Anthropic's article uses for per-domain warehouse reference docs: narrow the search space with curated files before the agent goes looking on its own.

Structure

qlik-sense-app-analysis/                     — repository root
├── README.md
├── LICENSE
└── qlik-sense-app-analysis/                 — the skill (the deployable unit)
    ├── SKILL.md                             — the skill itself
    └── references/
        ├── qlik-mcp-tool-reference.md       — full tool catalogue by category
        ├── expression-gotchas.md            — Qlik expression / set analysis syntax rules
        ├── governance-workflows.md          — master item / glossary / data product lifecycle rules
        ├── visualization-guidelines.md      — chart-type selection, rendering fallbacks, dashboard/report pattern
        ├── design-rationale.md              — the reasoning behind the skill's structure
        └── app-profile-TEMPLATE.md          — copy per app/domain to extend the skill

The prebuilt qlik-sense-app-analysis.zip is published on the Releases page rather than committed to the repo. The dist/ folder is a local build artifact (git-ignored) produced by the command below.

Building the zip

From the repository root:

zip -r dist/qlik-sense-app-analysis.zip qlik-sense-app-analysis

or on Windows PowerShell:

Compress-Archive -Path qlik-sense-app-analysis -DestinationPath dist/qlik-sense-app-analysis.zip -Force

The zip's top-level folder is qlik-sense-app-analysis/ containing SKILL.md, which is what skill uploaders expect.

Contributing

Pull requests that add well-documented app or domain profiles, correct tool behaviour as the Qlik MCP server evolves, or tighten up the gotchas/governance references are welcome. Please keep customer-specific detail (app IDs, internal field names) out of anything merged upstream — those belong in a private fork's references/ folder, not in the shared template.

License

MIT — see LICENSE.