Schema Sync Structured

Fix an agentic NL2SQL system after a database schema change. Inspect the schema contract, agent prompts, and vector database chunks for stale field references and update them.

DSmmartin Updated

File contents

Schema Sync — Structured

After a database schema migration, the following files and systems may contain stale field references.

Files to inspect

  • agentic_system/schema/github_events.yaml — machine-readable schema contract
  • agentic_system/agents_core/nl2sql/prompts/system.md — NL2SQL agent system prompt
  • agentic_system/agents_core/rag/prompts/system.md — RAG agent system prompt
  • ChromaDB collections: schema_docs, qa_examples — vector RAG chunks

What to look for

  1. Read the live ClickHouse schema: SELECT name, type FROM system.columns WHERE table = 'github_events' ORDER BY position
  2. Find references to old field names and SQL predicates in the files above.
  3. Update each file to use the new field names, types, and predicate syntax.
  4. For ChromaDB: find chunks with stale: True in their metadata and update their content.

Validation

Run python scripts/validate_schema.py to check whether the YAML contract matches the live schema.

DSmmartin/skills_talk_agent_camp_2026/tree/main/dev_tools/skill_examples/01_structured commit c28f5a8d73

Frequently asked questions

npx skillmds@latest add dsmmartin/schema-sync-structured