v1.100.0-dev.2

🐛 Bug Fixes

• Map /v1/messages provider errors before failure logging.
• Map upstream status codes for providers with no exception_type branch.
• Record ownership of streamed responses under managed IDs.
• Encrypt streamed responses IDs on /openai/v1/responses and /responses aliases.
• Fail open on response_format instead of dropping it for unregistered models.
• Honor DATABASE_DISABLE_PREPARED_STATEMENTS in componentized entrypoints.
• Preserve structured output strictness for Anthropic responses.
• Resolve hidden aliases for explicit lookup in the router.
• Keep focus in the add model public name input while typing.
• Price 1-hour cache writes on claude-3-haiku and claude-3-opus at 2x input.
• Keep the caller's Google token on credential-less Vertex passthrough under custom auth.
• Preserve provider access token lifetime in MCP.
• Raise bedrock_mantle gpt-5.5 and gpt-5.4 max_input_tokens to Mantle's enforced 1050000.
• Authenticate realtime websocket with Azure AD token when no api-key.
• Map reasoning_effort to reasoning.effort for OpenAI GPT-5.x on Converse.
• Support credential providers across Redis clients.
• Support `mode: image_edit` in health checks.
• Add US data residency uplift to claude-sonnet-4-6 and mythos entries.
• Raise missing-credential error on /v1/messages passthrough for Anthropic.
• Complete DCR bridge OAuth challenges in MCP.
• Tighten the pull-before-work rule in CLAUDE.md.
• Strip credential fields from GET /health output.
• Attach MINIMAX_API_KEY on Anthropic messages requests.
• Probe Azure GA realtime path for transcription-only models.
• Parse cacheDetails for Converse 1h/5m cache write cost split.
• Flush async cache writes cancelled at event loop shutdown.
• Resolve model_group_alias before pre-routing strategy dispatch.
• Stop cache eviction errors from failing /key/update.
• Honor global ssl_verify on the aiohttp_openai handler path.
• Rescue dequeued logging tasks lost at event loop close.
• Install the shadcn alert primitive.
• Re-pull label, textarea, separator and skeleton from the registry.
• Let the Together replayed-reasoning case survive a single provider miss.
• Pin the rate fallbacks inside a tiered-pricing tier.
• Cover the Bedrock provider-feature cells customers run.
• Pin the prices and period totals /cost/estimate returns.
• Let the mutation workflow find covered lines so it generates mutants.
• Cover caller-identity config failure cases.
• Pin the request-validation contracts in proxy/_types.py.
• Add pull-before-work rule to CLAUDE.md.
• Scale cache costs by fast mode and trust served speed for Anthropic.
• Add azure gpt-5.6 cache write rates and correct data zone priority.
• Remove stale "New" badges from the dashboard.
• Replace hand-picked z-index values with one named scale and lint it.
• Move every page header onto the shared PageHeader.