Skip to content

Add databricks-ai-gateway skill#471

Open
Praneeth16 wants to merge 1 commit intodatabricks-solutions:mainfrom
Praneeth16:add-databricks-ai-gateway-skill
Open

Add databricks-ai-gateway skill#471
Praneeth16 wants to merge 1 commit intodatabricks-solutions:mainfrom
Praneeth16:add-databricks-ai-gateway-skill

Conversation

@Praneeth16
Copy link
Copy Markdown
Contributor

Summary

  • Adds a new databricks-ai-gateway skill covering both V1 (per-endpoint config) and V2 Beta (enterprise control plane)
  • V1 coverage: guardrails (PII, safety, keywords, topics, LLM judge, custom), rate limits (QPM/TPM), inference tables, fallbacks, traffic splitting, observability via system tables
  • V2 coverage: coding agent governance (Cursor, Codex CLI, Gemini CLI, Claude Code), MCP server access control, agentic multi-step traceability, unified dashboard
  • Registers skill in install_skills.sh and adds cross-references to databricks-model-serving and databricks-agent-bricks

Files

File Purpose
SKILL.md Main entrypoint with V1/V2 decision tables, quick starts, SDK reference, common issues
1-guardrails.md PII BLOCK/MASK/NONE, safety, keywords, topics, LLM judge, custom guardrails
2-rate-limits-and-fallbacks.md QPM/TPM hierarchy, fallback config, traffic splitting patterns
3-inference-tables.md Payload logging to UC Delta tables
4-observability.md System tables, cost attribution, audit trail queries
5-coding-agents.md Cursor, Codex CLI, Gemini CLI, Claude Code gateway routing setup
6-agentic-governance.md MCP governance, multi-step traceability, unified control plane

Test plan

  • install_skills.sh --list shows the new skill
  • All internal links (6 sub-files) resolve correctly
  • All cross-reference links (5 related skills) resolve correctly
  • No sub-files have accidental frontmatter
  • All 10 SDK class names verified against actual SDK source
  • Rate limits + usage tracking tested on custom model endpoint (sn-lead-scoring-praneeth_paikray on e2-demo-west)
  • Full config (guardrails + rate limits + usage tracking) tested on external model endpoint (costco-meat-ai-gateway on e2-demo-west)
  • Guardrails correctly rejected on custom model endpoint (documented limitation)
  • Both endpoints cleaned up and restored to original state

This pull request was AI-assisted by Isaac.

…ng agent governance

AI Gateway is a major Databricks product with zero skill coverage today.
This skill covers both V1 (per-endpoint config: guardrails, rate limits,
inference tables, fallbacks) and V2 Beta (enterprise control plane: coding
agent governance, MCP server access control, agentic traceability).

Tested against e2-demo-west with real serving endpoints — rate limits,
usage tracking, and guardrails all verified working.

Co-authored-by: Isaac
@dustinvannoy-db
Copy link
Copy Markdown
Collaborator

I like the idea of a skill for this, would like to try and reduce size and also use references/ folder for the extra markdowns. @jacksandom do you have any time to guide Praneeth on setting up evals and trying to reduce size?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants