The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from Qlik Cloud apps. Ingestion points include qlik_get_full_glossary_export, qlik_list_measures, and qlik_describe_app (referenced in SKILL.md). Boundary markers are established through a 'Glossary Rules Ledger' and a 'Compliance Gate' used to validate data-driven instructions. The capability inventory includes qlik_create_data_object for data querying and qlik_create_bookmark for session persistence (referenced in references/qlik-mcp-tool-reference.md). Sanitization is performed via a 'floor' rule that prevents external content from overriding skill constraints or permitting restricted operations.
[COMMAND_EXECUTION]: The skill interacts with the Qlik engine via MCP tools. It strictly limits the generation of Qlik expressions, allowing only 'selection predicates' for row filtering (documented in references/selections-and-search.md) and forbidding any expression that calculates new figures or aggregates.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill is designed to read analytical data. It does not contain any hardcoded credentials or instructions to exfiltrate data to external domains. All network activity is conducted through authorized Qlik MCP tools.
[PERSISTENCE]: The skill follows a 'nothing survives the session' policy. The only exception is the creation of Qlik bookmarks, which occurs only upon explicit user request and is a standard feature of the target platform.

qlik-talk-to-data