METADATA-AWARE CODE GENERATION
Generate data code that knows what it can break.
ContextForge checks what a data change could break before an AI coding agent writes or ships the change.
It uses DataHub metadata to understand dependencies, governance and ownership, generates the code and tests, blocks risky autonomous action, and remembers approved decisions for future agents.
1. Change request
DEMO DATAHUB GRAPH2. Context used
MCP-READYDECISION TRACE
3. Evidence → impact → policy → action
This is an auditable decision summary, not hidden model chain-of-thought.
4. Impact-aware output
WAITINGRun an analysis to generate artifacts.The AI can prepare the change, but it cannot approve a risky production change by itself.
RUN COMPLETE
5. ContextForge Run Report
DataHub evidence analyzed
Artifacts generated
- ✓ dbt compatibility migration
- ✓ schema tests
- ✓ evidence-linked PR summary
Governance decision
Context memory
Approved decision retained in browser demo memory. Live DataHub write-back is implemented separately in the public CLI.
REAL DATAHUB WRITE-BACK PATH
6. Approved decision → DataHub context
The public ContextForge CLI contains the real MCP mutation path. It requires live DataHub mode plus explicit --approve --write-back authorization, then calls the official save_document tool.
Approve a run to preview the exact write-back payload.Real persistence requires configured DataHub credentials and write permission. The public repo includes the implementation, safety contract, and example payload.
CONTEXT MEMORY
7. The next agent inherits the decision
ContextForge remembers that legacy_segment was retained as a compatibility alias while downstream consumers migrate to customer_tier.
Browser Judge Mode uses local storage to demonstrate the user experience. Live mode persists the approved rationale to DataHub as a Decision document.
WHY THIS MATTERS
Your coding agent knows code. ContextForge teaches it your data system.
A syntactically valid change can still break dashboards, violate governance rules, surprise owners, or silently change business logic. ContextForge turns metadata into operational constraints and human-review boundaries.
Know the blast radius
Lineage exposes downstream dependencies before code is generated.
Carry governance into code
Ownership, PII and trust signals become generation constraints and review gates.
Make AI decisions auditable
Code, tests, policy and review evidence stay linked instead of disappearing inside a prompt.
ONE PATTERN, MORE THAN DATA
A context-aware change engine.
Data engineering
Schema, lineage, ownership, governance and usage become safe code-generation context.
Cybersecurity
The same pattern can gate IAM, detection, dependency and infrastructure changes by blast radius and authority.
Live game systems
Player-data schemas, telemetry, economy tables and backend changes can inherit dependency context before deployment.
Data engineering is the hackathon proving ground. Cyber and gaming are architecture expansion paths, not claimed integrations in this demo.
JUDGE TAKEAWAY