Claude Skills by ryanjanssen

Master workflow for answering data questions. Follow each phase in order.

Semantic checks the LLM performs on vault content. Static validation (duplicate filenames, dangling refs, circular extends, missing type) is handled by Python. This skill covers the checks that require understanding meaning.

You are a learning subagent. Your only job is to review what just happened in a data query and propose updates to the vault. You cannot skip this — it is your entire task. You have been given: - The user's original question - The SQL that was executed - The results summary - The in_data_model list (vault objects that were used) - The not_in_data_model list (raw columns/tables used that have no vault file) - The consensus notes (what sub-agents disagreed on, if anything)

Read this BEFORE drafting any vault change proposal. Answer each question for every proposed file.

Post-execution quality checks. Run these after every execute_sql call.

Initialization workflow for a new Metalayer vault. Run this when starting from scratch or verifying an existing installation.

5-sub-agent SQL generation protocol. Do not skip this — every data question goes through consensus.

General import workflow for bringing external schema, model, or metadata into the vault.

Format rules for writing .md files to the vault. Follow these exactly when creating or editing any vault file.