core/v1.7.11

## Core Release v1.7.11

- fix: retry after an unverifiable reasoning refusal on chat-shaped requests too - `/v1/chat/completions` and `/v1/messages` carry replayed reasoning on `reasoning_details`, but the fail-soft strip only handled Responses-shaped items, so a router that switched models mid-conversation returned "messages.N.content.0: Invalid `signature` in `thinking` block" straight to the client instead of retrying without the signature
- fix: strip thinking signatures off Responses content blocks, not just `encrypted_content` on the reasoning item - a message could need the strip with `encrypted_content` already absent, and only reasoning items are dropped when nothing survives so an ordinary message keeps its own content
- fix: stop sending `reasoning.content` to non-gpt-oss OpenAI/Azure reasoning models, which cap the array at zero entries and reject a populated one with "Invalid 'input[N].content': array too long. Expected an array with maximum length 0"; replayed Anthropic thinking blocks translate into `reasoning_text` blocks and were hitting this. `summary` + `encrypted_content` already carry everything OpenAI accepts
- fix: stop clearing `reasoning_effort` for current-generation Grok models - the rule substring-matched "grok-3-mini", so `grok-4.5`, `grok-4.6` and `grok-4.20-multi-agent` all silently lost the field and answered at the wrong reasoning depth, cost and latency. Replaced with an exact-match deny-list (`SupportsGrokReasoningEffort`) that normalizes routing prefixes, `-latest` and xAI's 4-digit date suffixes
- fix: keep `reasoning_effort: "xhigh"` for `grok-4.6` and `grok-4.20-multi-agent` - the shared OpenAI-dialect normalizer downgraded it to "high" before the xAI compat pass ran, so the value was lost even with the deny-list corrected. `grok-4.5` still downgrades on purpose, matching xAI's documented upstream coercion
- fix: emit `content_part.added`, `output_text.delta`, `output_text.done` and `content_part.done` when a tool-based structured-output call is reassembled into a message on the Responses streaming path - only `output_item.added`/`done` were emitted, so every consumer reading incremental events rather than the item snapshot saw a stream with no text at all. A schema-constrained `streamGenerateContent` to Bedrock Mantle returned `{"candidates":[{"content":{"role":"model"},"finishReason":"STOP"}]}` with tokens billed. Affects Vertex, Bedrock Mantle and Azure Claude, the three providers that emulate structured output with a forced tool call
- feat: inline URL-sourced images and documents for AWS-hosted Claude on the native-Anthropic path - Bedrock Mantle rejects `{"source":{"type":"url"}}` with "URL content sources are not yet supported for this model". Fetches go through the SSRF-safe dialer with a size cap, and a failed fetch aborts the request rather than silently dropping an attachment. Brings the native-Anthropic surface to parity with Bedrock's Converse path
- feat: bedrock vpc endpoints support (#6064)
- feat: add `use_idp_credentials` to token-exchange config so SSO login app credentials can be reused for providers like Microsoft Entra ID (#6068)
- feat: add w3c trace id to context (#5945)
- feat: persist and resync MCP tool discoveries uniformly across all client types via a hash-gated core callback
- feat: add per user oauth mcp support for config.json
- feat: add a context path for skipping auth resolution on trusted internal callers
- feat: cost accounting for prompt guardrails (#4931)
- fix: path normalization auth bypass (#5763)
- fix: preserve minimal reasoning effort for GPT-5-family OpenAI models (thanks [@jitokim](https://github.com/jitokim)!) (#6046)
- fix: map truncated Gemini responses to the MAX_TOKENS finish reason (thanks [@AdityaPainuli](https://github.com/AdityaPainuli)!) (#5979)
- fix: omit absent tool-call function name on streaming deltas instead of emitting null (thanks [@AdityaPainuli](https://github.com/AdityaPainuli)!) (#5966)
- fix: bedrock files handling in inference (#5947)
- fix: cost in usd ticks for xai usage (#5950)
- fix: add anthropic error branch when stripping encrypted reasoning content
- fix: discover tools synchronously for per-call MCP clients, fix shared-OAuth reconnect and verify errors
- fix: break lock-order inversion in ConnectionCheckerManager, close a data race in the performCheck test
- fix: rebuild ephemeral client fresh across the whole connect+init retry
- fix: preserve last-known tool maps across close-first reconnects
- fix: bind MCP connect attempts to entry identity and guard AddClient's discovery path
- fix: pin needs_session_stickiness across config.json reconciliation so an unrelated file edit cannot revert it to per-call
- fix: restrict Reauthorize to shared OAuth clients
- fix: reject inactive tokens in ValidateToken, document the shared vs per-identity oauth token lookup contract
- fix: don't silently drop stored oauth scopes on decode failure, skip rotation instead
- fix: gate SSE OnConnectionLost on connection identity
- fix: close the verify-headers double-submit race, preserve TLS, timeout and per-user-header fields on OAuth-completion updates
- fix: repair shared connections regardless of destructive hint, fail closed on missing tool annotations, dedupe background reconnect
- fix: configure bounded http.Server timeouts and a request-body limit
- fix: guard nil ConfigStore, propagate resource, surface pending-bootstrap cleanup failure
- chore: dependabot dependency updates (#6040)

### Installation

```bash
go get github.com/maximhq/bifrost/core@v1.7.11
```

---
_This release was automatically created from version file: `core/version`_