What problem does it solve? Getting a new model and endpoint into the local OpenRouter development environment requires correctly populating Postgres rows (models, endpoints, pricing_versions), refreshing the KV cache, and verifying routing — a process with many failure modes like hidden rows, stale caches, and missing pricing. This Skill provides the exact two-path workflow (prod-first seeding or manual SQL staging) plus debugging guidance to get an endpoint serving requests locally. ## Core Features & Use Cases - Prod-first seeding path: Refresh seed CSVs from the prod DB via the GitHub Action, run the seed script, and unhide rows locally for zero-drift staging. - Manual SQL staging fallback: Insert models, endpoints, and pricing_versions by hand with full field references for modalities, features JSON, provider overrides, and parameter arrays. - Cache refresh and curl testing: Ordered catalog refresh (restart api, run api-kv-cron, restart consumers) plus ready-made curl commands for chat, streaming, reasoning, tools, embeddings, and video. - Use Case: You are developing a new provider adapter and need to test requests end-to-end locally — stage the model and endpoint, refresh the KV cache, and debug routing via dev-fs-logs until curl returns a valid response. ## Quick Start Ask the agent to stage the model openai/gpt-4o with provider OpenAI in the local database and verify it with a test chat completion request.
