feat(core): honor per-model provider overrides in the models.dev gateway#18928
Contributor trust inconclusive
Contributor andrelandgraf (Andre Landgraf, andre@neon.tech) shows extensive cross-repository activity concentrated on the Neon ecosystem and related AI SDK integrations. Of the 33 assigned PRs, only 10 have any hydrated patch content and 23 are metadata-only. Of the 10 hydrated, 4 have full or near-full patches available and 6 are in preview/truncated mode. The visible patches in anomalyco/models.dev#3021, vercel/ai#15997, #18928, neon-solutions/add-mcp#56, and neondatabase/neon-pkgs#279 show legitimate feature work—TOML model configuration, AI SDK provider scaffolding, gateway routing logic, CLI tooling, and SDK unification—with no evidence of credential exfiltration, hidden network calls, obfuscated code, CI tampering, or dependency abuse. However, the majority of the shard is unreviewable due to omission or truncation. The non-dominant repository contributions (vercel/ai, mastra-ai/mastra) contain truncated implementation patches, preventing adversarial verification of the full provider/gateway code. Given that most candidate PRs are metadata-only, the contributor has broad cross-repo activity, and 70 of 100 total PRs were omitted entirely, this shard cannot support a confident safe verdict.