Goose

✨ Features

• Stable agent event message identity #10716
• Tab completion and switch provider for model command #10585
• Support latest Gemini models #10630
• Configurable GOOSE_DOCS_ROOT for air-gapped docs access #10294
• Structured summary output with template rendering for compaction #10471
• Allow disabling built-in skills #10600
• Gate tool-call label enrichment on ACP client capability #10644

🐛 Bug Fixes

• Restore dynamic model discovery and current Grok support #10756
• Update DeepSeek model names to v4 API #10729
• Forward Claude Code model picker selection to the adapter #10675
• Forward Pi model selection to the adapter #10707
• Preserve enabled flag when migrating platform extensions #10690
• Pass session working_dir to on_call_tool #10689
• Preserve adversary tool arguments #10529
• Keep XML snippets inside JSON as data for local inference #10606
• Sync generated chat title in header #10578
• Prompt for tool approval in telegram gateway sessions #10613
• Keep CLI provider prompts out of process arguments #10547
• Ignore stale inventory for static model lists #10591
• Use shell working directory on initial desktop launch #10655
• Enable inventory refresh for OpenRouter model picker #10641
• Preserve read lines and reload tool output in ACP #10662
• Surface tool call output in CLI and Desktop #10654
• Truncate overlong function names in provider requests #10659
• Preserve ACP tool call update fields in desktop #10653
• Use configured session manager for tool summaries #10628
• Omit disabled OpenRouter reasoning #10517
• Clear stale pending ACP connection after terminal recovery failure #10552
• Confirm pending steer message on queuedSteer notification #10532
• Apply hermit env directly in node shims so fish login shell doesn't break MCP startup #10028

🔧 Improvements

• Upgrade to rmcp 2.0 #10584
• Opus5 support including adaptive thinking #10686
• Refine dark theme message surfaces for consistent, compliant dark mode UX #10693
• Fix Korean translation errors, untranslated entries, and terminology consistency #10667
• Simplify ACP tool-call handling and fix chain summaries #10599
• Enhance the uniffi API layer #10427
• Disable scheduler by default for ACP #10781
• UI cleanup to remove unused components and functions #10761
• Bump nostr to 0.44.6 to clear RUSTSEC-2026-0216 #10733
• Create source archive as release artifact #10786

Security fixes

• Arbitrary command execution in goose CLI via goose review (via git core.fsmonitor config) - GHSA-r5pp-p5r8-466r (credit to @0xmagic0)

✨ Features

• Support for latest Gemini models #10630
• Pass working_dir to the Stop hook context #10296
• Declarative Sakana AI provider for OpenAI-compatible Fugu API #10357
• Model interactions viewer restored #10205
• Chunk command-classifier input with overlapping windows for security #10416
• Search filter for provider grid #10437
• Muse Spark 1.1 support via Meta Models API #10432
• GPT-5.6 support #10384
• Reconnect desktop ACP sessions after sleep and connection loss #10411
• Per-message usage stats UI (tokens, cost, TTFT, tok/s) #10210
• OllamaCloudProvider with dynamic model discovery and context limits #10264
• Delete support for custom apps from Apps UI #10127
• --edit session flag to edit conversation before forking #9799
• iFlytek Spark and Astron MaaS providers #9837
• Fireworks AI declarative provider #9990
• OpenRouter request parameters support #9276
• French, German, Italian, Portuguese, Indonesian, Malay, Vietnamese, zh-TW desktop locales #10072
• MiniMax-M3 and missing M2.7 variants #10431

🐛 Bug Fixes

• Preserve command arguments through extension forms #10527
• Enforce manual approval for code mode #10528
• Confine app file operations #10481
• Preserve featured model size on delete and backfill missing sizes #10422
• Preserve recursive schema types with pctx 0.4.1 #10572
• Create inventory tables atomically with schema version #10586
• Rewrite oneOf to anyOf in tool schemas for OpenAI-compatible backends #10571
• Allow custom model as default for non-local providers #10438
• Require absolute goose path roots #10454
• Scope smart approval by request #10457
• Preserve fixed subrecipe values #10452
• Bundle git in Flatpak so hermit can clone its package registry #10511
• Expose AGENT_SESSION_ID to shell commands #10428
• Forward image content chunks to client during live session #10485
• Dispatch Code Mode callbacks on main runtime to prevent hangs #10484
• Protect OAuth token caches #10516
• Preserve content audience boundaries #10493
• Bound scheduled recipe validation #10509
• Return JSON-RPC errors from dispatch handlers instead of killing connection #10499
• Isolate prepare tools test state #10504
• Avoid local inference Metal teardown crash #10508
• Enable indexmap serde support for musl builds #10507
• Scroll chat to latest reply on session resume #10501
• Secure large response spill files #10482
• Register LiteLLM with refresh_only inventory so model picker lists models #10489
• Limit provider selector height in CLI #10420
• Tighten existing secrets file permissions #10453
• Classify image access capabilities #10451
• Preserve one-time permission scope #10450
• Register goose:// scheme handler in Flatpak desktop entry #10461
• Reject duplicate recipe parameter keys #10448
• Synchronously kill orphaned goose serve backends on GUI close #10292
• Break self-referential $ref cycles before Code Mode schema #10469
• Stop multiline paste auto-submitting line-by-line on Windows #10326
• Hide nested config values in Configuration Editor #10390
• Stop resolving TUI scripts from cwd #10449
• Exclude MCP App tools from Code Mode callbacks #10290
• Apply default extension timeout to developer shell tool calls #10348
• Preserve reasoning_content for DeepSeek thinking mode on multi-turn tool calls #10366
• Make prompt timestamps timezone-explicit #10209
• Disable compaction during approval #10089
• Restore pnpm 11 project configuration #10395
• Treat empty-string finish_reason as non-terminal in streaming #10258
• Fix Gemini OAuth onboarding #10363
• Retry empty provider turns instead of stopping silently #10360
• Guard llama.cpp backend init against SIGILL on x86_64 CPUs without FMA/AVX2 #10105
• Sanitize responses function call names #10344
• Pick session mode from the ids the agent actually offers #10320
• Show manual-setup indicator for extensions without install link #10321
• Dedupe duplicate signed thinking blocks to fix Anthropic 400 #10083

🔧 Improvements

• Extract tool call handling from ACP server #10574
• Report cache-aware Harbor costs #10430
• Trim macOS desktop entitlements #10490
• Consolidate and harden release workflows #10447
• TLS backend matrix job for rustls-tls and native-tls #10148
• Bound code-mode execution with timeout and cancellation #10214
• Recover malformed tool calls from GLM/Minimax models instead of rejecting them #10230
• Group chat sessions by project in nav panel #10212
• Per-message usage/cost tracking with derived session totals #10172
• Return effective context limit from /model-info #10165
• Disable thinking preservation for Groq #10284
• Clarify terminal color semantics in CLI #10414

📚 Documentation

• Clean up stale documentation audit findings #10114
• Update environment and config references #10203
• Document hook deny contract, matcher regex, tool_input keys, and provider config #10467
• Remove stale gooseignore references #10359
• Add Code of Conduct #10424
• Clarify session --edit context retention #10257

✨ Features

• Reconnect desktop ACP sessions after sleep and connection loss #10411
• Per-message usage stats UI (tokens, cost, TTFT, tok/s) #10210
• Per-message usage/cost tracking with derived session totals #10172
• Group chat sessions by project in nav panel #10212
• Show project names in directory switcher dropdown #10207
• OllamaCloudProvider with dynamic model discovery and context limits #10264
• Delete support for custom apps from Apps UI #10127
• --edit session flag to edit conversation before forking #9799
• iFlytek Spark and Astron MaaS providers #9837
• Fireworks AI declarative provider #9990
• OpenRouter request parameters support #9276
• OVHcloud Qwen3.6-27B model (262K context) #10339
• GPT-5.6 family support #10349
• GLM-5.2 support #9967
• French, German, Italian, Portuguese, Indonesian, Malay, Vietnamese, and Traditional Chinese desktop locales #10072
• Provider bindings MVP to goose-sdk with Python wheel publishing #10208
• TLS support for ACP serve #10088
• Streamable HTTP support in deeplink generator #9954

🐛 Bug Fixes

• Exclude MCP App tools from Code Mode callbacks #10290
• Preserve reasoning_content for DeepSeek thinking mode on multi-turn tool calls #10366
• Compaction disabled during approval #10089
• Treat empty-string finish_reason as non-terminal in streaming #10258
• Gemini OAuth onboarding failure #10363
• Retry empty provider turns instead of stopping silently #10360
• Guard llama.cpp backend init against SIGILL on x86_64 CPUs without FMA/AVX2 #10105
• Sanitize responses function call names #10344
• Pick session mode from the IDs the agent actually offers #10320
• Desktop ACP shell PATH #10336
• Dedupe duplicate signed thinking blocks to fix Anthropic 400 #10083
• Resolve provider default model instead of leaking sentinel "current" #10305
• Return effective context limit from /model-info #10165
• Gate rcgen aws_lc_rs feature behind rustls-tls #10116
• Recover malformed tool calls from GLM/Minimax models instead of rejecting them #10230
• Restore cursor when configure exits #10222
• Skip scheduled sessions for implicit resume #10178
• Sync slash help with builtins #10173
• Treat unicode punctuation as image path terminators #10106
• Don't bleed provider-specific request_params into delegate() sessions #9906
• Purge empty phantom sessions on startup #9651
• Include structuredContent from MCP tool response #10128
• Propagate agent cleanup errors on session delete #10112
• Tolerate partial Responses output items #10108
• Always use OpenAI-compatible "parameters" for Databricks tools #10101
• Unescape shell-escaped image paths #10098
• Send inference config (max_tokens, temperature) on Bedrock Converse #9889
• Dedupe Nostr session deep link imports #9918
• Don't retry deterministically-permanent 400s (thinking-block immutability) #10005
• Use async reqwest client so OTLP export works in goose serve mode #10100
• Reject non-object tool-call arguments instead of panicking #9832
• Save /edit prompts to history #10011
• Buffer thinking across stream chunks for DeepSeek/Kimi tool calls #9704
• Surface actionable errors for truncated tool-call arguments #9946
• Show optimistic toggle state when disabling an extension #9882
• Resolve databricks_v2 pricing and surface cost in standard usage update #9925
• Activate hermit on macOS so STDIO extensions resolve node #9482
• Prompt timestamps are now timezone-explicit #10209

🔧 Improvements

• Apply the default extension timeout to developer shell tool calls #10348
• Disable thinking preservation for Groq #10284
• Bound code-mode execution with timeout and cancellation #10214
• Keep turn-context out of the Anthropic prefix cache #10030
• Disable extended thinking for fast-model operations #9815
• Switch the local inference provider MLX backend to use the safemlx crate #10304
• OpenAI Responses API store param configurable #10040
• Support streamed tool calls without indexes #10023
• Fail closed on unverifiable update provenance #10168
• Migrate diagnostics to JSON report #9964
• Route MCP elicitations through tool streams #9943
• ACP migration for desktop UI #10081
• Relabel settings 'Session' tab to 'External Backend' #10318
• Show manual-setup indicator for extensions without an install link #10321

📚 Documentation

• Clarify session --edit context retention #10257
• Name the message field in the hooks payload guide #9913
• Custom agents guide #9293
• Correct custom agents guide #9947

✨ Features

• OllamaCloudProvider with dynamic model discovery and context limits #10264
• Per-message usage stats UI (tokens, cost, TTFT, tok/s) #10210
• Per-message usage/cost tracking with derived session totals #10172
• Group chat sessions by project in nav panel #10212
• Show project names in directory switcher dropdown #10207
• Provider bindings MVP to goose-sdk with Python wheel publishing #10208
• AskAI bot hints for followups #10291

🐛 Bug Fixes

• Resolve provider default model instead of leaking sentinel "current" #10305
• Return effective context limit from /model-info #10165
• Recover malformed tool calls from GLM/Minimax models instead of rejecting them #10230
• Restore cursor when configure exits #10222
• Sync slash help with builtins #10173
• Treat unicode punctuation as image path terminators #10106
• Don't bleed provider-specific request_params into delegate() sessions #9906
• Purge empty phantom sessions on startup #9651
• Include structuredContent from MCP tool response #10128
• Gate rcgen aws_lc_rs feature behind rustls-tls #10116

🔧 Improvements

• Increase /mcp-app-guest body limit for larger single-file MCP apps #10324
• Relabel settings 'Session' tab to 'External Backend' #10318
• Switch local inference provider MLX backend to use the safemlx crate #10304
• Disable thinking preservation for Groq #10284
• Bound code-mode execution with timeout and cancellation #10214
• Skip scheduled sessions for implicit resume #10178
• Fail closed on unverifiable update provenance #10168
• Update Rust toolchain and raise recursion limit #10303
• Drop unsupported gpt-5.3-codex from chatgpt_codex provider #10271
• Move Google provider into goose-providers #10216
• Move Snowflake provider #10218
• Move Databricks providers into goose-providers crate #10162
• Move declarative definitions #10167

📚 Documentation

• Clarify session --edit context retention #10257
• Simplified the issue templates #10316
• Remove stale crates/goose-server and update docs #10224

✨ Features

• iFlytek Spark and Astron MaaS providers (#9837)
• --edit session flag to edit conversation before forking (#9799)
• Fireworks AI declarative provider (#9990)
• OpenRouter request parameters support (#9276)
• French, German, Italian, Portuguese, Indonesian, Malay, Vietnamese, and Traditional Chinese desktop locales (#10072)
• Spanish (es) locale support (#9833)
• Korean localization for desktop app (#9856)
• Japanese locale support (#9768)
• Hindi desktop locale (#9733)
• Language selection feature in desktop (#9405)
• /status slash command in CLI (#9845)
• Streamable HTTP support in deeplink generator (#9954)
• GLM-5.2 model support (#9967)
• Global hints loading from ~/.agents/AGENTS.md (#9736)
• Azure Entra ID bearer token auth via AZURE_OPENAI_AD_TOKEN (#9716)
• Together AI declarative provider (#9545)
• OrcaRouter as declarative OpenAI-compatible provider (#9515)
• EmpirioLabs AI declarative provider (#9771)
• MLX model support for local inference provider (#9154)
• Summon peek mode for async background tasks (#9519)
• Summon context parameter for delegate tasks (#9518)
• Summon working_dir override for delegate tasks (#9520)
• Structured metadata for summon task load results (#9521)
• TLS support for ACP serve (#10088)
• OpenAI Responses API store param configurability (#10040)
• Exposed available tools in ACP schema (#10097)
• ACP thinking effort config option (#10711)
• Last message snippets for ACP sessions (#9798)
• OTLP logging schema for cross-tool detection (#9713)
• Option to disable automatic update downloads (#9872)
• Cache token tracking for accurate cost reporting (#9752)
• Session import functionality (#9474)
• xAI SuperGrok OAuth subscription provider (#9420)
• Perplexity as declarative OpenAI-compatible provider (#9324)
• Alibaba (Qwen via DashScope) declarative provider (#9443)
• Configurable tool output size limit via GOOSE_MAX_TOOL_RESPONSE_SIZE (#9256)
• /model slash command for session model switching (#8747)
• Databricks AI Gateway provider (#9274)
• Russian language support (#9406)
• Turkish desktop locale (#9392)
• NEAR AI Cloud provider (#9352)
• Scaleway provider (#9254)
• Harbor eval runner (#9138)
• /goal command for agent self-evaluation before finishing (#9069)
• Summon subagent instructions (#9325)
• Open-plugins generalization + skills (#9112)
• PreToolUse denial in hooks (#9304)
• Optional api_key configuration for declarative openai-engine providers (#9202)
• goose review local code review command (#9114)
• TUI diff viewer (#9260)
• Paginated session list in ACP (#9199)
• Structured per-provider config block with non-destructive provider switching (#8977)
• tui command on goose-cli (#9385)
• Configurable MAX_CODE_BLOCK_LINES via env vars (#9301)
• goose://resume session deep link (#9343)
• GOOSE_FAST_MODEL env var support (#9296)
• Vercel AI Gateway provider (#9144)
• Quarterly option for scheduler (#9076)
• Hooks feature (#9093)
• Projects as backend sources with system prompt injection (#8739)
• Worktree-aware directory switcher (#8450)
• Chat history search feature (#8448)
• Optional --parameters for scheduled recipes (#8741)
• GitHub Copilot /responses API (#9043)
• Unified thinking effort control across all providers (#9242)
• Slash commands (built-in, skill, recipe) in ACP server (#9238)
• Nushell terminal and completion support (#8628)
• Simplified Chinese (zh-CN) translation (#8765)
• Atomic Chat as declarative OpenAI-compatible provider (#9210)
• GOOSE_OAUTH_CALLBACK_PORT for stable OAuth redirect_uri (#9209)
• Proactive OAuth token refresh to avoid re-auth on every session (#8386)
• goose://new-session deep link for fresh chat (#9196)
• Routstr as declarative provider (#9175)
• FuturMix provider (#8840)
• oMLX declarative provider (#9177)
• Declarative provider support in goose-providers crate (#9992)
• ACP elicitation improvements (#9797)
• Image read tool (#9607)
• gpt-5.5 model support for ChatGPT Codex provider (#9292)

🐛 Bug Fixes

• Canonical model output paths (#10196)
• ACP agent cleanup errors propagation on session delete (#10112)
• claude-sonnet-5 added to canonical model registry (#10183)
• MCP-app-guest request body limit increased (#10180)
• Tolerate partial Responses output items (#10108)
• Databricks tool parameters for OpenAI compatibility (#10101)
• Unescape shell-escaped image paths (#10098)
• Desktop Mac auto-update config seeding (#10132)
• /edit prompts saved to history (#10011)
• Schedule sessions listing using session.message_count (#10026)
• Session store sync after provider/model update and thinking effort update (#10060)
• Fallback REST API call removed in ACP mode for edit in place (#10034)
• Don't retry deterministically-permanent 400s (thinking-block immutability) (#10005)
• Async reqwest client for OTLP export in goose serve mode (#10100)
• Reject non-object tool-call arguments instead of panicking (#9832)
• Databricks pricing resolution and cost surfacing (#9925)
• Removed unused SubagentStart/SubagentStop hook events (#9960)
• Recipe modal closes when canceling parameter form (#9195)
• Tool request timestamp precedes tool response timestamp (#9462)
• Activate hermit on macOS so STDIO extensions resolve node (#9482)
• Auto-visualiser JSON-encoded string handling (#9838)
• User-Agent sent on URL image fetches (#9927)
• Do not pass temperature to ChatGPT Codex provider (#9931)
• Nostr session deep link import deduplication (#9918)
• Provider refresh when session working directory changes (#9883)
• GCP refreshed gcloud token after reauth (retry Vertex AI on 401/403) (#9849)
• Start a turn when /goal or /grind is set (#9801)
• Handle resume deep links (#9298)
• PATH repair for plugin hook commands (#9615)
• Cmd/Ctrl+T new chat shortcut restored (#9614)
• Mention menu icon opens sidebar (#9643)
• Dedupe React to prevent multiple instances in renderer (#9631)
• Hermit cleanup added to uvx and jbang scripts (#9616)
• Blocking OTLP HTTP exporter (#9599)
• Fall back to default provider when resuming session with unavailable provider (#9547)
• Tokio runtime provided for OTLP HTTP exporter batch thread (#9541)
• OpenAI meta.n_ctx read from /v1/models for local server context limit (#9530)
• Session history scrollbar aligned with window edge (#9601)
• Session name refresh in sidebar after rename (#9543)
• Resolved skill supporting file paths shown (#9584)
• Bedrock real streaming via ConverseStream API (#9579)
• OpenAI-native cached_tokens recorded in usage metering (#9829)
• Moonshot provider API URL override option (#9819)
• Image paths with spaces detected (#9387)
• New chat shortcut navigation restored (#9528)
• Stale completed tasks evicted in summon (#9514)
• OAuth callback timeout in WSL instead of hanging (#9536)
• Unmatched extension names warning in delegate (#9513)
• Elicitation decline and cancel actions propagated end-to-end (#9437)
• Bedrock inference config (max_tokens, temperature) sent on Converse (#9889)
• Optimistic toggle state shown when disabling an extension (#9882)
• Windows backslash paths preserved in custom extension command (#9741)
• ACP cancel race condition fixed (#9804)
• Removed window goosed when closing window (#9818)
• Context exceeded checker made more precise (#9831)
• Login-shell PATH inherited in spawned subagents (#9737)
• String values accepted for GOOSE_CONTEXT_LIMIT (#9738)
• Rejected OAuth credentials cleared after refresh (#9694)
• Extended thinking kept within Anthropic output cap (#9814)
• RPM build-id links disabled (#9671)
• Custom API path preserved in OpenAI base_url derivation (#9649)
• Bedrock ValidationException classified as ExecutionError (#9735)
• Unparseable extension entries preserved during config refresh (#9439)
• Session info ACP method (#9729)
• Anthropic stream refusals surfaced as visible errors (#9724)
• Ollama_cloud correctly mapped to canonical provider (#9639)
• Groq provider max tokens values corrected (#9790)
• Stream decode errors marked retryable (#9723)
• Databricks AI Gateway v2 endpoints pagination (#9753)
• Pattern-based prompt injection detection confidence scores adjusted (#9690)
• ACP compatibility in TUI for config extensions (#9683)
• Thinking content buffered across stream chunks for DeepSeek/Kimi tool calls (#9704)
• Actionable errors surfaced for truncated tool-call arguments (#9946)
• Deleting recently created session and creating new one no longer shows deleted content (#9351)
• Custom OpenAI provider port preserved when URL scheme omitted (#9730)
• ACP agentInfo included in initialize response (#9765)
• Local inference media marker parts used for vision prompts (#9452)
• Extensions icon visible after switching sessions (#9787)
• Non-recipe project config files skipped in summon (#9808)
• Duplicate tool-call ids deduped within a turn (#9792)
• Bundled extensions resolved from discovery (#9759)
• Thinking config passed to Bedrock Anthropic models (#9794)
• LiteLLM context limit from /model/info for custom models (#9303)
• Thinking content preserved for provider context (#9314)
• Configurable timeout env vars mentioned in Ollama stream stall error (#9246)
• Full recipe parameter details included in load/discovery output (#9233)
• Unused fetch-metadata IPC handler removed (SSRF) (#9340)
• Gateway respects GOOSE_MAX_TURNS (#9354)
• Anthropic empty object sent instead of null for tool_use input (#9355)
• Selected branch preserved across project chats (#9010)
• Tolerate missing responses output (#9449)
• Azure api-version query param made optional (#9221)
• Pending env vars preserved in Add Extension form (#9285)
• Linux x86_64 built inside manylinux_2_28 for glibc 2.28+ compat (#9415)
• Main window stops growing taller on every launch (#9409)
• Desktop chat search session limiting (#9366)
• Per-session agent creation serialized to stop duplicate MCP init (#9357)
• OTLP signals skipped when protocol=grpc to avoid background-thread panic (#9512)
• Stop hook decisions honored when blocking (#9468)
• ACP server context window size forwarded to clients (#9455)
• Provider list refresh in Switch Models picker (#9408)
• New chat started in current window from recipe param modal (#9422)
• Trace_output emitted as span attribute instead of event (#9255)
• File fallback checked when keyring has no entry (#9279)
• ScheduleModal error message styling (#9278)
• Sidebar hamburger aligned in macOS fullscreen (#9257)
• current_exe() used instead of PATH lookup when spawning goose (#9236)
• TCP_USER_TIMEOUT set on streamable HTTP clients (#9207)
• Custom provider activated after adding via configure (#9213)
• OTLP traces flushed reliably on exit with configurable timeout (#9228)
• Excessive MISSING_TRANSLATION warnings reduced for fallback locales (#9294)
• Cross-window deep link contamination eliminated (#9273)
• Telegram gateway error reporting and connection reliability improved (#9223)
• VT processing enabled on Windows Console Host (#9248)
• Cost zeroed for local providers (ollama, local) (#9222)
• Tool-use marker leakage prevented in toolshim output (#8310)
• Summon canonical limits re-applied when delegate overrides model (#9183)
• CLI cumulative total_tokens reported in stream-json/json output (#8910)
• GCP metadata server token refreshed on expiration (#8929)
• WSL2 OS detection corrected by removing PWD-based Windows override (#8869)
• Bedrock ReasoningContent blocks handled gracefully (#8843)
• Tool pair summaries inserted at chronological position in conversation (#9087)
• SQLite foreign key enforcement enabled in session pool (#9121)
• Deleted chat session still appearing in sidebar list (#8674)
• Extension_name required on read_resource (#8989)
• SACP notifies clients of generated session names (#8983)
• Quoted numeric config values converted to numbers if needed (#8844)
• SSE reconnect loop kept alive on long disconnects (#8717)
• OpenAI request_params applied to outgoing API payload (#9151)
• Gemini 3.x known_location Global routing and KNOWN_MODELS entries (#9142)
• Anthropic Retry-After honored on 429 responses (#9161)
• Streaming Thinking deltas coalesced + available tools listed on not-found (#9162)
• User-set session name preserved for recipe-based chats (#9079)
• OIDC-proxy exp enforcement independent of MAX_TOKEN_AGE_SECONDS (#8832)
• Inline code contrast improved in light theme (#9058)
• Max_tokens omitted for OpenAI-compatible requests when unset (#9123)
• MCP-app host capabilities set correctly (#9116)
• Nullable schemas normalized for Vertex Gemini compatibility (#8930)
• GitHub recipe temp paths isolated (#8878)
• X-Initiator header set on GitHub Copilot requests (#8809)
• Dynamic_models: false honored in declarative provider configs (#8795)
• Frontend extension session state and tool routes (#8464)
• Bedrock trailing message cached for stable prefix across agent turns (#8916)
• Case-insensitive model name lookup for context_limit (#8906)
• Cumulative token usage emitted in stream-json/json complete event (#8870)
• 400 returned instead of panicking on invalid CSP header value (#8810)
• ACP streaming chunks coalesced under one message id (#8788)
• Goose2 vite server reused on port conflict (#8722)
• Python3 used in developer extension instructions for macOS/Linux compatibility (#8784)
• ACP child synchronously reaped to avoid SIGCHLD race (#9023)
• Goose2 small-window chat and settings layouts (#9019)
• Elicitation fixes (#8999)
• OpenAI null tool_call arguments accepted in streaming chunks (#9035)
• Plain '> ' prompt used instead of goose emoji in CLI (#9305)
• Goosed not killed when a window closes (#9302)
• Databricks parallel tool image responses don't interleave tool results (#9241)
• Accumulated cost persisted in session DB to survive reload (#9191)
• Publish-npm binary built from current SHA + compat check (#9212)
• Preprompt excluded from session title generation (#8793)
• Login-shell PATH probe prevented from suspending goose on startup (#8804)
• Venice replaced with declarative config (#9234)
• MOIM allowlist widened for expected fix_conversation warnings (#9226)
• mkdir -p used for self-test workspace initialization (#9247)
• Tool name shown in approval prompt (#9216)
• ACP images replayed on session load (#9496)
• Azure CLI resolved on Windows using az.cmd (#9215)
• Non-interactive terminal handled in goose configure on Windows (#9214)
• ACP recipe state applied during session load and fork (#9998)
• Local inference GGUF requirements stricter, auto detection of tool calling support, thinking output parsing fixed (#9442)
• Goose alias preferred for Databricks v2 inventory (#9430)
• Streamed tool calls without indexes supported (#10023)
• Allow local file ACP origins (#10194)
• MCP app sandbox bridge lifecycle fixed (#10064)

🔧 Improvements

• Refactored local inference provider crates (#10169)
• ACP auth and origin defaults improved (#9886)
• UI connected to ACP directly instead of goosed (#10081)
• Turn-context kept out of Anthropic prefix cache (#10030)
• Extended thinking disabled for fast-model operations (#9815)
• Diagnostics migrated to JSON report (#9964)
• MCP elicitations routed through tool streams (#9943)
• Provider refactored: model config not required to create provider (#9953)
• Session name as source of truth for displayed session titles (#9841)
• Provider inventory split out of providers (#9826)
• MCP extensions supported in open plugins (#9471)
• Databricks Responses API support check (#9347)
• Initial prompt with goose://new-session (#9427)
• View model interactions (#9806)
• API Client moved to goose-providers (#9869)
• Request logger refactored out of providers (#9847)
• Provider trait moved into goose-providers (#9860)
• Global config access moved out of ModelConfig (#9769)
• Debug info dropped for dependencies in dev profile (#9843)
• Formats/openai.rs moved into goose-providers crate (#9633)
• Goose-providers crate created with canonical models, conversation and other types (#9588)
• Generic validator for declarative providers (#10010)
• UI types decoupled from generated OpenAPI types (#10149, #10109)
• ACP SDK upgraded to new HTTP/WS crate (#10082)
• MCP sampling support removed (#10087)
• Legacy MCP-UI proxy support removed (#10086)
• Session_id removed from provider streaming trait methods (#9984)
• Ollama provider moved into goose-providers (#9986)
• Anthropic provider moved into goose-providers (#9985)
• OpenAI provider moved into goose-providers (#9880)
• LRU cache used for token counting (#9586)
• Hugging Face OAuth support added with auth tab in settings (#9552)
• Review subprocess timeout replaced with turn limits (#9571)
• Benchmarking improved (#9465, #9637)
• Canonical thinking modes added (#9743)
• Skill instructions dynamically refreshed each turn (#9217)
• Pasted HTML parsed better as markdown (#9190)
• Dependency hygiene improved (#9360)
• Summon instructions built per turn (#9329)
• Protocol cleanup (#9147)
• Settings moved into app shell (#9047)
• Extension page redesigned (#8940)
• Provider catalog moved behind ACP layer in goose2 (#9030)
• Skills library redesigned (#8868)
• Sidebar navigation and project icons polished (#8896)
• Provider resolution parallelized and SQLite pool eagerly initialized (#8899)
• Skill categories removed (#9008)
• Consecutive tool calls grouped into summarized chain card (#8995)
• Artifact heuristics/regexes replaced with protocol messages (#8996)
• CLI help text updated for --session-id (#10077)
• Canonical model registry refreshed (#10709, #9551, #9331)
• EmpirioLabs provider display name updated (#10126)

📚 Documentation

• Custom agents guide (#9293)
• Custom agents guide correction (#9947)
• Message field named in hooks payload guide (#9913)
• Release testing instructions updated (#10122)
• ACP clients documentation updated (#9772)
• Session History navigation/delete docs fixed (#9727)
• Remote goosed server connection guide (#9275)
• SaladCloud AI Gateway documentation (#9253)
• Summon extension requirement documented for delegate and load tools (#9231)
• Hooks guide (#9288)
• Hooks feature blog post (#9227)
• Documentation reorganized (#9310)
• Scaleway provider documentation (#9423)
• Self-improving agents blog post (#9846)
• AAIF migration blog post (#9866)
• Orchestrating with goose blog post (#9104)
• Minimax and office QA benchmark findings blog post (#8984)
• Goose with peekaboo blog post (#8884)
• Built-in Local Inference blog post (#8808)
• DCO references removed (#9864)
• Internal documentation anchors fixed (#9094)
• Ophis MCP server added to extensions registry (#9861)
• Scholar Sidekick tutorial page (#9477)
• Installation whitespace fix (#9867)
• Stats update (#9410)
• MCP blog post typo fix (#9641)
• Analyze extension instructions updated (#9585)

✨ Features

• Language selection in desktop app #9405
• French, German, Italian, Portuguese, Indonesian, Malay, Vietnamese, Traditional Chinese desktop locales #10072
• Korean localization for desktop app #9856
• Japanese locale support #9768
• Spanish (es) locale support #9833
• Hindi desktop locale #9733
• /status slash command in CLI #9845
• /model slash command for session model switching #8747
• /goal command for agent self-evaluation before finishing #9069
• goose review local code review command #9114
• Peek mode for async background tasks in summon #9519
• Context parameter and working_dir override for delegate tasks #9518 #9520
• GLM-5.2 support #9967
• Together AI declarative provider #9545
• EmpirioLabs as declarative OpenAI-compatible provider #9771
• OrcaRouter as declarative OpenAI-compatible provider #9515
• NEAR AI Cloud provider #9352
• Perplexity as declarative OpenAI-compatible provider #9324
• Alibaba (Qwen via DashScope) declarative provider #9443
• Databricks AI Gateway provider #9274
• Scaleway provider #9254
• Vercel AI Gateway #9144
• xAI SuperGrok OAuth subscription provider #9420
• OpenAI GPT models via AWS Bedrock mantle endpoint #9707
• Azure Entra ID bearer token auth via AZURE_OPENAI_AD_TOKEN #9716
• MLX models support for local inference provider #9154
• Streamable HTTP support for deeplink generator #9954
• Global hints loading from ~/.agents/AGENTS.md #9736
• Unified thinking effort control across all providers #9242
• Canonical thinking modes #9743
• Hooks system for extensibility #9093
• PreToolUse denial in hooks #9304
• Honor blocking Stop hook decisions #9468
• Option to disable automatic update downloads #9872
• Track cache tokens for accurate cost reporting #9752
• Configurable tool output size limit via GOOSE_MAX_TOOL_RESPONSE_SIZE #9256
• Configurable MAX_CODE_BLOCK_LINES via env vars #9301
• GOOSE_FAST_MODEL env var support #9296
• Hugging Face OAuth support with auth tab in settings #9552
• Chat history search in navigation panel #8448
• Import sessions #9474
• View model interactions #9806
• View JSON content #9678
• Image read tool #9607
• goose://new-session deep link #9196
• goose://resume session deep link #9343
• Initial prompt with goose://new-session #9427
• MCP extensions support in open plugins #9471
• Nushell terminal and completion support #8628
• TUI command on goose-cli #9385
• TUI diff viewer #9260
• Unified OTLP logging schema for cross-tool detection #9713
• Prompt injection mitigation enabled by default for internal users #9612
• Encrypted Nostr session sharing #8922
• Quarterly option for scheduler #9076
• Optional --parameters for scheduled recipes #8741
• Worktree-aware directory switcher #8450
• Harbor eval runner #9138
• Linux musl CLI builds #9240

🐛 Bug Fixes

• Mac auto-update config seeding #10132
• Use async reqwest client so OTLP export works in goose serve mode #10100
• Reject non-object tool-call arguments instead of panicking #9832
• Save /edit prompts to history #10011
• Use session.message_count for schedule sessions listing #10026
• Sync session store after provider and model update #10060
• Buffer thinking across stream chunks for DeepSeek/Kimi tool calls #9704
• Surface actionable errors for truncated tool-call arguments #9946
• Show optimistic toggle state when disabling an extension #9882
• Do not pass temperature to ChatGPT Codex provider #9931
• Resolve databricks_v2 pricing and surface cost in standard usage update #9925
• Activate hermit on macOS so STDIO extensions resolve node #9482
• Autovisualiser: accept JSON-encoded string for the data parameter #9838
• Developer/read_image: send a User-Agent on URL image fetches #9927
• Start a turn when /goal or /grind is set #9801
• Pass thinking config to Bedrock Anthropic models #9794
• Preserve Windows backslash paths in custom extension command #9741
• Custom OpenAI provider dropping port when URL scheme is omitted #9730
• ACP cancel race condition #9804
• Removed window goosed when closing window #9818
• Context exceeded checker more precise #9831
• Inherit login-shell PATH in spawned subagents #9737
• Accept string values for GOOSE_CONTEXT_LIMIT #9738
• Clear rejected OAuth credentials after refresh #9694
• Keep extended thinking within the Anthropic output cap #9814
• RPM build-id links disabled #9671
• Preserve custom API path in base_url derivation for OpenAI #9649
• Handle resume deep links #9298
• Repair PATH for plugin hook commands #9615
• Restore Cmd/Ctrl+T new chat shortcut #9614
• Lead with extension description, shorten HTTP/SSE transport label #9653
• Mention menu icon to open the sidebar #9643
• Dedupe React to prevent multiple instances in renderer #9631
• Hermit cleanup in uvx and jbang scripts #9616
• Use blocking OTLP HTTP exporter #9599
• Fall back to default provider when resuming session with unavailable provider #9547
• Provide Tokio runtime for OTLP HTTP exporter batch thread #9541
• Read meta.n_ctx from /v1/models to fix context limit for local servers #9530
• Align session history scrollbar with window edge #9601
• Refresh session name in sidebar after rename #9543
• Show resolved skill supporting file paths #9584
• Implement real streaming via Bedrock ConverseStream API #9579
• Record OpenAI-native cached_tokens in usage metering #9829
• Override API URL option for Moonshot provider #9819
• Detect image paths with spaces #9387
• Restore new chat shortcut navigation #9528
• Evict stale completed tasks to prevent unbounded memory growth #9514
• OAuth callback timeout instead of hanging in WSL #9536
• Warn on unmatched extension names in delegate #9513
• Deleting recently created session and creating new shows deleted session's content #9351
• Close modal when canceling recipe parameter form #9195
• Ensure tool request timestamp precedes tool response timestamp #9462
• Correctly map ollama_cloud to canonical provider #9639
• Incorrect max tokens values for groq provider #9790
• Mark stream decode errors retryable #9723
• Classify Bedrock ValidationException as ExecutionError #9735
• Page through all Databricks AI Gateway v2 endpoints when listing models #9753
• Preserve unparseable extension entries during config refresh #9439
• Surface Anthropic stream refusals as visible errors #9724
• Skip non-recipe project config files in summon #9808
• Dedup duplicate tool-call ids within a turn #9792
• Resolve bundled extensions from discovery #9759
• Include agentInfo in ACP initialize response #9765
• Keep extensions icon visible after switching sessions #9787
• Use media marker parts for vision prompts in local-inference #9452
• Make azure api-version query param optional #9221
• Preserve pending env vars in Add Extension form #9285
• Tolerate missing responses output #9449
• Stop main window growing taller on every launch #9409
• Serialize per-session agent creation to stop duplicate MCP init #9357
• Respect GOOSE_MAX_TURNS in gateway sessions #9354
• Send empty object instead of null for Anthropic tool_use input #9355
• Use context limit from /model/info for LiteLLM custom models #9303
• Skip OTLP signals when protocol=grpc to avoid background-thread panic #9512
• Forward ACP server context window size to clients #9455
• Refresh provider list in Switch Models picker #9408
• Start new chat in current window from recipe param modal #9422
• Use refreshed gcloud token after reauth (retry Vertex AI on 401/403) #9849
• Set TCP_USER_TIMEOUT on streamable HTTP clients #9207
• Activate custom provider after adding via configure #9213
• Reduce excessive MISSING_TRANSLATION warnings for fallback locales #9294
• Eliminate cross-window deep link contamination #9273
• Improve Telegram gateway error reporting and connection reliability #9223
• Enable VT processing on Windows Console Host #9248
• Zero out cost for local providers (ollama, local) #9219
• Autovisualiser: use appInfo instead of clientInfo in MCP Apps init handshake #9249
• Use mkdir -p for self-test workspace initialization #9247
• Show tool name in approval prompt #9216
• Resolve Azure CLI on Windows by using az.cmd #9215
• Handle non-interactive terminal in goose configure on Windows #9214
• Persist accumulated cost in session DB to survive reload #9191
• Prevent tool-use marker leakage in toolshim output #8310
• Re-apply canonical limits when delegate overrides model #9183
• Remove hard-coded React alias that points to wrong location #9842

🔧 Improvements

• Keep turn-context out of the Anthropic prefix cache #10030
• Disable extended thinking for fast-model operations #9815
• Support streamed tool calls without indexes #10023
• Diagnostics migrated to JSON report #9964
• Route MCP elicitations through tool streams #9943
• Propagate elicitation decline and cancel actions end-to-end #9437
• Allow unlisted models in search #9933
• Sanitize extension environment maps #9884
• Drop debug info for dependencies in dev profile #9843
• Databricks responses API support check #9347
• Open Model Gym output dir configurable #9789
• Per-agent timeout configurable (GYM_AGENT_TIMEOUT) #9791
• Session name as source of truth for displayed session titles #9841
• Stricter GGUF requirements with auto detection of tool calling support #9442
• Prefer goose aliases for Databricks v2 inventory #9430
• Build summon instructions per turn #9329
• ACP session list pagination #9199
• Structured per-provider config block, non-destructive provider switching #8977
• Emit trace_output as span attribute instead of event #9255
• Flush OTLP traces reliably on exit with configurable timeout #9228
• Proactive OAuth token refresh to avoid re-auth on every session #8386
• Better parsing of pasted HTML as markdown #9190
• Preserve thinking content for providers that require it #8857
• Dynamically refresh skill instructions each turn #9217
• GOOSE_OAUTH_CALLBACK_PORT support for stable OAuth redirect_uri #9209
• Honor Retry-After on 429 responses #9161
• Coalesce streaming Thinking deltas #9162
• Strip chain-of-thought markers from custom provider output #8635
• Use LRU cache for token counting #9586
• Replace review subprocess timeout with turn limits #9571
• Opt-in ACP last message snippets #9798
• Expose raw provider supported models over ACP #9475
• CLI to list skills with token counts #9326
• Include request URL in provider error messages #9232
• Surface resolved Databricks model metadata #9206
• Ensure parallel tool image responses don't interleave for Databricks #9241
• Slash commands (built-in, skill, recipe) in ACP server #9238
• Open-plugins generalization + skills #9112
• Use plain '> ' prompt instead of goose emoji #9305

📚 Documentation

• Name the message field in the hooks payload guide #9913
• Custom agents guide #9293
• Correct custom agents guide #9947
• Add guide for connecting goose Desktop to a remote goosed server #9275
• Hooks guide #9288
• Blog post for hooks feature #9227
• Document summon extension requirement for delegate and load tools #9231
• ACP clients documentation update #9772
• AAIF migration blog post #9866
• Self-improving agents blog post #9846
• Orchestrating with goose blog post #9104
• Goose with peekaboo blog post #8884
• Built-in Local Inference blog post #8808
• MiniMax and office QA benchmark findings blog post #8984
• Scaleway provider documentation #9423
• SaladCloud AI Gateway documentation #9253
• Scholar Sidekick tutorial page #9477
• Documentation reorganization #9310

✨ Features

• ACP method for managing recipes and UI changes #9951
• Use ACP to manage global config and session extensions #9948
• Recipe handling in new_session #9935
• /status slash command in CLI #9845
• ACP new session on desktop (non-recipe sessions) #9914
• Typed request/response for add, get session extensions in ACP #9890
• Option to disable automatic update downloads #9872
• Change cwd using ACP and refactor extension in bottom menubar #9887
• Track cache tokens for accurate cost reporting #9752
• Load session using ACP #9875
• View model interactions #9806
• Support OpenAI GPT models via AWS Bedrock mantle endpoint #9707
• Peek mode for async background tasks in summon #9519
• Korean localization for desktop app #9856
• Chat reply UI with ACP #9857
• Durable ACP chat session state for desktop UI #9852
• ACP for desktop chat prompt (feature toggle off) #9802
• MCP extensions support in open plugins #9471
• Initial prompt with goose://new-session deep link #9427
• Configurable Open Model Gym output dir #9789
• Spanish (es) locale support #9833
• EmpirioLabs as a declarative OpenAI-compatible provider #9771
• MLX models support for local inference provider #9154
• Elicitation method for ACP with improvements #9797
• View JSON #9678
• Load global hints from ~/.agents/AGENTS.md #9736
• Configurable per-agent timeout (GYM_AGENT_TIMEOUT) #9791
• Azure Entra ID bearer token auth via AZURE_OPENAI_AD_TOKEN #9716
• _meta.conversationBefore for ACP forking session #9821
• Quarterly option for scheduler #9076
• Together AI declarative provider #9545
• Image read tool #9607
• Context parameter for summon delegate tasks #9518
• Structured metadata for summon task load results #9521
• Working_dir override for summon delegate tasks #9520
• OrcaRouter as a declarative OpenAI-compatible provider #9515
• Elicitation decline and cancel actions propagation #9437
• ACP opt-in last message snippets #9798
• Japanese locale support #9768
• Unified OTLP logging schema for cross-tool detection #9713
• Canonical thinking modes #9743
• Hindi Desktop locale #9733
• Custom ACP method to get session info #9729
• GOOSE_SERVER__SECRET_KEY support at goose serve ACP endpoint #9726
• Surface Anthropic stream refusals as visible errors #9724
• ACP search session in Desktop #9717
• Steering messages with ACP #9560
• ACP list session with keyword and type filter #9695
• ACP thinking effort config option #9711
• Prompt injection mitigation enabled by default for internal users #9612
• ACP list sessions and manage sessions in Desktop #9687
• ACP methods for config extensions #9581
• Custom notifications only when ACP client specifies capability #9596
• Minimal uniffi setup for cross-language SDK #9593
• Hugging Face OAuth support with auth tab in settings #9552
• Import sessions #9474
• xAI SuperGrok OAuth subscription provider #9420
• Replay ACP images on session load #9496
• Raw provider supported models over ACP #9475
• ACP session system prompt setter #9478
• Scholar Sidekick MCP extension #9433
• tui feature flag to gate the tui command #9428
• CLI to list skills with token counts #9326
• Perplexity as a declarative OpenAI-compatible provider #9324
• Alibaba (Qwen via DashScope) declarative provider #9443
• Configurable tool output size limit via GOOSE_MAX_TOOL_RESPONSE_SIZE #9256
• /model slash command to CLI for session model switching #8747
• Databricks AI Gateway provider #9274
• Russian language support #9406
• Turkish desktop locale #9392
• Language selection feature #9405
• Handle resume deep links #9298

🐛 Bug Fixes

• Publish mac updater metadata #9945
• Activate hermit on macOS so STDIO extensions resolve node #9482
• Accept JSON-encoded string for autovisualiser data parameter #9838
• Send User-Agent on URL image fetches in developer/read_image #9927
• Allow unlisted models in search #9933
• Do not pass temperature to ChatGPT Codex provider #9931
• Sanitize extension environment maps #9884
• Desktop updater to target aaif-goose releases #9868
• Refresh provider when session working directory changes #9883
• Use refreshed gcloud token after reauth (retry Vertex AI on 401/403) #9849
• Start a turn when /goal or /grind is set #9801
• Remove hard-coded React alias that breaks dev render #9842
• Pass thinking config to Bedrock Anthropic models #9794
• Check for responses API support in Databricks #9347
• Use media marker parts for vision prompts in local-inference #9452
• Keep extensions icon visible after switching sessions #9787
• Skip non-recipe project config files in summon #9808
• Dedupe duplicate tool-call ids within a turn #9792
• Resolve bundled extensions from discovery #9759
• Include agentInfo in ACP initialize response #9765
• Custom OpenAI provider dropping port when URL scheme is omitted #9730
• Preserve Windows backslash paths in custom extension command #9741
• ACP cancel race condition #9804
• Removed window goosed when closing window #9818
• Inherit login-shell PATH in spawned subagents #9737
• Accept string values for GOOSE_CONTEXT_LIMIT #9738
• Clear rejected OAuth credentials after refresh #9694
• Keep extended thinking within the Anthropic output cap #9814
• Disable RPM build-id links #9671
• Preserve custom API path in OpenAI base_url derivation #9649
• Repair PATH for plugin hook commands #9615
• Restore Cmd/Ctrl+T new chat shortcut #9614
• Lead with extension description, shorten HTTP/SSE transport label #9653
• Mention menu icon to open the sidebar #9643
• Dedupe React to prevent multiple instances in renderer #9631
• Add hermit cleanup to uvx and jbang scripts #9616
• Use blocking OTLP HTTP exporter #9599
• Fall back to default provider when resuming session with unavailable provider #9547
• Provide Tokio runtime for OTLP HTTP exporter batch thread #9541
• Read meta.n_ctx from /v1/models to fix context limit for local OpenAI servers #9530
• Align session history scrollbar with window edge #9601
• Refresh session name in sidebar after rename #9543
• Show resolved skill supporting file paths #9584
• Implement real streaming via ConverseStream API for Bedrock #9579
• Record OpenAI-native cached_tokens in usage metering #9829
• Add option for overriding API URL for Moonshot provider #9819
• Detect image paths with spaces #9387
• Restore new chat shortcut navigation #9528
• Evict stale completed tasks to prevent unbounded memory growth in summon #9514
• OAuth timeout instead of hanging in WSL #9536
• Warn on unmatched extension names in summon delegate #9513
• Deleting a recently created session then creating new session shows deleted session's content #9351
• Close modal when canceling recipe parameter form #9195
• Ensure tool request timestamp precedes tool response timestamp #9462
• Correctly map ollama_cloud to canonical provider and update max tokens #9639
• Incorrect max tokens values for groq provider #9790
• Mark stream decode errors retryable #9723
• Classify Bedrock ValidationException as ExecutionError #9735
• Page through all Databricks AI Gateway v2 endpoints when listing models #9753
• Preserve unparseable extension entries during config refresh #9439
• Goose-sdk release compat check with new schema #9697
• Compatibility of config extension ACP call in TUI #9683

🔧 Improvements

• Move OpenAI provider into goose-providers #9880
• Move ApiClient to goose-providers, lifting out TlsConfig #9869
• Refactor request logger out of providers #9847
• Delete embeddings support #9865
• Move provider trait into goose-providers #9860
• Move global config access out of ModelConfig, move ModelConfig into goose_providers #9769
• Add Ophis MCP server to the extensions registry #9861
• Make session.name the source of truth for displayed session titles #9841
• Split provider inventory out of providers #9826
• Drop debug info for dependencies in dev profile #9843
• Make context exceeded checker more precise #9831
• Update analyze extension instructions #9585
• Validate desktop i18n catalogs #9776
• Refresh canonical model registry #9709
• Move formats/openai.rs into goose-providers crate #9633
• Create goose-providers crate with canonical models, conversation and other types #9588
• Remove dead component and useNavigationSessions cleanup #9603
• ACP session setup refactor #9488
• Use LRU cache for token counting #9586
• Replace review subprocess timeout with turn limits #9571
• Improve benchmarking #9637
• Pattern-based detection adjustments for prompt injection confidence scores #9690
• Removed create recipe from session modal and REST API #9916

📚 Documentation

• Correct custom agents guide #9947
• Custom agents documentation #9293
• AAIF migration blog post #9866
• Remove DCO references #9864
• Self-improving agents blog post #9846
• Correct whitespace in installation fenced code block #9867
• Update docs for ACP clients #9772
• Fix stale session navigation/delete docs #9727
• Fix typo in MCP blog post #9641
• Add Scaleway provider docs #9423
• Scholar Sidekick tutorial page #9477

✨ Features

• Opt-in ACP last message snippets for sessions #9798
• Unified OTLP logging schema for cross-tool detection #9713
• Canonical thinking modes across providers #9743
• Custom ACP method to get session info #9729
• ACP secret key support at goose serve endpoint #9726
• Surface Anthropic stream refusals as visible errors #9724
• ACP search session support in Desktop #9717
• Steering messages with ACP #9560
• ACP list sessions with keyword and type filter #9695
• ACP thinking effort config option #9711
• ACP session management in Desktop #9687
• Image read tool #9607
• ACP methods for config extensions #9581
• Custom notifications only when ACP client specifies capability #9596
• Minimal uniffi setup for cross-language SDK #9593
• Hugging Face OAuth support with auth tab in settings #9552
• Session import functionality #9474
• Replay ACP images on session load #9496
• Raw provider supported models over ACP #9475
• ACP session system prompt setter #9478
• Scholar Sidekick MCP extension #9433
• TUI feature flag to gate tui command #9428
• CLI to list skills with token counts #9326
• xAI SuperGrok OAuth subscription provider #9420
• Perplexity as declarative OpenAI-compatible provider #9324
• Alibaba (Qwen via DashScope) declarative provider #9443
• Configurable tool output size limit via GOOSE_MAX_TOOL_RESPONSE_SIZE #9256
• /model slash command in CLI for session model switching #8747
• Databricks AI Gateway provider #9274
• TUI command on goose-cli #9385
• Configurable MAX_CODE_BLOCK_LINES via env vars #9301
• NEAR AI Cloud provider #9352
• goose://resume session deep link #9343
• Scaleway provider #9254
• Honor GOOSE_FAST_MODEL env var in ModelConfig::with_fast #9296
• goose review local code review command #9114
• TUI diff viewer #9260
• ACP paginate session list #9199
• Structured per-provider config block, non-destructive provider switching #8977
• Pass session cwd param to ACP providers #9229
• gpt-5.5 added to known models #9292
• Opt-in to Vercel AI Gateway leaderboard #9259
• Harbor eval runner #9138
• Slash commands (built-in, skill, recipe) in ACP server #9238
• /goal command for agent self-evaluation before finishing #9069
• Summon subagent instructions #9325
• Open-plugins generalization with skills #9112
• PreToolUse denial hook #9304
• Optional api_key configuration for declarative openai-engine providers #9202
• Login shell PATH flag #9313
• Linux musl CLI builds #9240
• goose://new-session deep link to open fresh chat #9196
• GOOSE_DISABLE_TOOL_CALL_SUMMARY to opt out of per-tool-call summaries #8947
• Recipe discovery and execution via ACP server #8925
• GOOSE_OAUTH_CALLBACK_PORT for stable OAuth redirect_uri #9209
• Proactive OAuth token refresh to avoid re-auth on every session #8386
• Encrypted Nostr session sharing #8922
• Simplified Chinese (zh-CN) translation #8765
• Nushell terminal and completion support #8628
• Optional --parameters for scheduled recipes #8741
• GitHub Copilot /responses API #9043
• Vercel AI Gateway provider #9144
• Routstr declarative provider #9175
• FuturMix provider #8840
• oMLX declarative provider #9177
• Strip chain-of-thought markers from custom provider output #8635
• Quarterly option for scheduler #9076
• Hooks system #9093
• Atomic Chat as declarative OpenAI-compatible provider #9210
• Projects as backend sources with system prompt injection #8739
• Worktree-aware directory switcher #8450
• Chat history search feature in navigation panel #8448
• Expose built-in skills through sources list ACP calls #9045
• ACP streamable HTTP spec compliance #9034
• Move goose2 provider catalog behind ACP layer #9030
• Group consecutive tool calls into one summarized chain card #8995
• Replace raw config and secret methods in ACP #9000
• Support Google model inventory refresh #8913
• MCP Apps inline rendering in goose2 #8877
• Goose2 custom providers support in UI and ACP #8924
• Ollama host configurable in goose2 #8912
• Skills added to chat composer #8881
• MCP Apps hydrate and replay payloads in Goose2 #8632
• Windows CUDA release artifacts #8750

🐛 Bug Fixes

• Desktop new chat shortcut #9659
• Correctly map ollama_cloud to canonical provider and update max tokens #9639
• Incorrect max tokens values for Groq provider #9790
• Classify Bedrock ValidationException as ExecutionError #9735
• Page through all Databricks AI Gateway v2 endpoints when listing models #9753
• Preserve unparseable extension entries during config refresh #9439
• goose-sdk release compat check with new schema #9697
• Compatibility of config extension ACP call in TUI #9683
• Intermittent missing extension override on UI #9575
• Scheduled recipe session params #9553
• Skip OTLP signals when protocol=grpc to avoid background-thread panic #9512
• Forward ACP server context window size to clients #9455
• Quote release PR search phrase in pre-release.sh #9573
• Refresh provider list in Switch Models picker #9408
• Start new chat in current window from recipe param modal #9422
• Make Azure api-version query param optional #9221
• Preserve pending env vars in Add Extension form #9285
• Tolerate missing responses output #9449
• Stricter GGUF requirements, auto detection of tool calling support, fixed thinking output parsing #9442
• Build linux x86_64 standard inside manylinux_2_28 for glibc 2.28+ compat #9415
• Stop desktop main window growing taller on every launch #9409
• Desktop chat search session limiting #9366
• Serialize per-session agent creation to stop duplicate MCP init #9357
• Respect GOOSE_MAX_TURNS in gateway sessions #9354
• Send empty object instead of null for Anthropic tool_use input #9355
• Use context limit from /model/info for LiteLLM custom models #9303
• Preserve thinking content for provider context #9314
• Mention configurable timeout env vars in Ollama stream stall error #9246
• Include full recipe parameter details in load/discovery output #9233
• Remove unused fetch-metadata IPC handler (SSRF) #9340
• Emit trace_output as span attribute instead of event #9255
• Check file fallback when keyring has no entry #9279
• Desktop ScheduleModal error message styling #9278
• Align sidebar hamburger in macOS fullscreen #9257
• Use current_exe() instead of PATH lookup when spawning goose #9236
• Set TCP_USER_TIMEOUT on streamable HTTP clients #9207
• Activate custom provider after adding via configure #9213
• Reduce excessive MISSING_TRANSLATION warnings for fallback locales #9294
• Eliminate cross-window deep link contamination #9273
• Improve Telegram gateway error reporting and connection reliability #9223
• Enable VT processing on Windows Console Host #9248
• Zero out cost for local providers (ollama, local) #9219
• Replace Venice custom provider with declarative config #9234
• Widen MOIM allowlist to suppress expected fix_conversation warnings #9226
• Use mkdir -p for self-test workspace initialization #9247
• Re-apply canonical limits when delegate overrides model #9183
• Report cumulative total_tokens in stream-json/json output #8910
• Refresh GCP metadata server token on expiration #8929
• Convert quoted numeric config values to numbers if needed #8844
• Keep SSE reconnect loop alive on long disconnects #8846
• Apply request_params to outgoing OpenAI API payload #9151
• Add Gemini 3.x known_location Global routing and KNOWN_MODELS entries #9142
• Coalesce streaming Thinking deltas + list available tools on not-found #9162
• Honor Retry-After on 429 responses #9161
• Preserve user-set session name for recipe-based chats #9079
• Enforce exp independently of MAX_TOKEN_AGE_SECONDS in OIDC proxy #8839
• Improve inline code contrast in light theme #9058
• Omit max_tokens for OpenAI-compatible requests when unset #9123
• Set correct mcp-app host capabilities #9116
• Normalize nullable schemas for Vertex Gemini compatibility #8930
• Isolate GitHub recipe temp paths #8878
• Set X-Initiator header on GitHub Copilot requests #8809
• Honor dynamic_models: false in declarative provider configs #8795
• Cache trailing message for stable prefix across Bedrock agent turns #8916
• Case-insensitive model name lookup for context_limit #8906
• Emit cumulative token usage in stream-json/json complete event #8870
• Return 400 instead of panicking on invalid CSP header value #8810
• Coalesce streaming chunks under one message id in ACP #8788
• Reuse goose2 vite server on port conflict #8722
• Correct WSL2 OS detection by removing PWD-based Windows override #8869
• Handle Bedrock ReasoningContent blocks gracefully #8843
• Insert tool pair summaries at chronological position in conversation #9087
• Enable SQLite foreign key enforcement in session pool #9121
• Deleted chat session still appears in sidebar list #8674
• Require extension_name on read_resource #8989
• Queued message lost when clicking sending now #9074
• Keep model picker and session config in sync in goose2 #9048
• Read prompted file parameters in recipes #8504
• Accept null tool_call arguments in streaming chunks #9035
• Use python3 in developer extension instructions for macOS/Linux compatibility #8784
• Synchronously reap ACP child to avoid SIGCHLD race #9023
• Elicitation fixes #8999
• SACP notifies clients of generated session names #8983
• CRT linkage in Windows CUDA build #8987
• Unscheduling a recipe should not delete them #8978
• Seed provider handoff history in ACP #8941
• Respect goose2 interface density settings #8953
• Make goose2 respect accent color #8952
• Preserve replay message timestamps #8942
• Goose2 window minimum sizing #8946
• Avoid transform-rasterized dialog text in goose2 #8937
• Keep renamed skills open in detail view #8935
• Gemini OAuth onboarding failure #8905
• Model picker stays usable during provider loading #8900
• Handle ACP requests concurrently #8781
• Prevent tool-use marker leakage in toolshim output #8310
• Exclude preprompt from session title generation #8793
• Prevent login-shell PATH probe from suspending goose on startup #8804

🔧 Improvements

• Mark stream decode errors retryable #9723
• Validate desktop i18n catalogs #9776
• Move formats/openai.rs into goose-providers crate #9633
• Create goose-providers crate with canonical models and types #9588
• ACP session setup refactor #9488
• LRU cache for token counting #9586
• Replace review subprocess timeout with turn limits #9571
• Benchmarking infrastructure #9465
• Honor blocking Stop hook decisions #9468
• Forward custom headers through OAuth connect path #9388
• Prefer goose aliases for Databricks v2 inventory #9430
• Improve dependency hygiene #9360
• Simplify UI customization #9353
• Build summon instructions per turn #9329
• Preserve selected branch across project chats #9010
• Unified thinking effort control across all providers #9242
• Linux desktop Vulkan packages #9323
• Flush OTLP traces reliably on exit with configurable timeout #9228
• Mount ACP in goosed server for iterative migration #9097
• Use plain '> ' prompt instead of goose emoji in CLI #9305
• Remove popular chat topics from new chat screen #9307
• Stop killing goosed when a window closes #9302
• Remove vendored Windows binaries #9318
• Include request URL in provider error messages #9232
• Ensure parallel tool image responses don't interleave for Databricks #9241
• Surface resolved Databricks model metadata #9206
• Better parsing of pasted HTML as markdown #9190
• Persist accumulated cost in session DB to survive reload #9191
• Prompt injection mitigation pattern updates #9198
• TUI spacing/layout improvements #9243
• Preserve thinking content for providers that require it #8857
• Dynamically refresh skill instructions each turn #9217
• Build non-vulkan Linux variants using ubuntu 22.04 #9211
• Show tool name in approval prompt #9216
• Localize hardcoded strings in provider settings UI #8931
• Move settings into app shell #9047
• Add Location column to CLI skills table #8785
• Consolidate logging setup into shared helper #8817
• Protocol cleanup #9147
• Refresh onboarding when provider catalog loads #9051
• Frontend extension session state and tool routes #8464
• Remove threads layer, use sessions directly for ACP #9078
• Install plugins to ~/.agents/plugins #9088
• Use native arm64 runners for Linux artifact builds #9075
• Linux Vulkan support for local inference #9038
• Switch to official new rust-sdk for ACP #9062
• Provider-first onboarding #9039
• Replace artifact heuristics/regexes with protocol messages #8996
• Fold UI refactor review into code review skill #8982
• LM Studio declarative provider improvements #8973
• Remove artifacts dir handling #8945
• Install plugins support #8949
• Redesign Extensions page, treat Extension Manager as core MCP enabler #8940
• Goose2 settings headers cleanup #8951
• Goose2 compact settings modal headers #8950
• Break up acp/server.rs #8932
• Goose2 skill refactor #8897
• Polish sidebar navigation and project icons #8896
• Update provider row after saving credentials #8914
• Parallelize provider resolution and eagerly init SQLite pool #8899
• Update goose2 credential management behind provider-scoped ACP/core API #8887
• Deduplicate _goose/providers/list RPC call at startup #8873
• Mergeable configs #8378
• Agent provider to use explicit type states #8879
• Translate ACP host capabilities into MCP initialization #8623
• Redesign skills library #8868
• Prompt injection mitigation enabled by default for internal users #9612
• Re-adjust pattern-based detection confidence scores #9690
• Pattern-based detection updates to reduce false positives #9350
• Add directionality to egress logging #9546

📚 Documentation

• Update ACP clients documentation #9772
• Fix stale session navigation/delete docs #9727
• Fix typo in MCP blog post #9641
• Update analyze extension instructions #9585
• Add Scaleway provider documentation #9423
• Add Repology badge to README #9245
• Document summon extension requirement for delegate and load tools #9231
• Guide for connecting goose Desktop to remote goosed server #9275
• SaladCloud AI Gateway documentation #9253
• Hooks guide #9288
• Blog post for hooks feature #9227
• Fix internal documentation anchors #9094
• Blog: orchestrating with goose #9104
• Updated tunnel remote access documentation #9077
• Blog: goose with peekaboo #8884
• Built-in Local Inference blogpost #8808
• Blog: findings with minimax and office QA benchmark #8984
• Reorganize documentation #9310
• Stats update #9410

🌐 Localization

• Japanese locale support #9768
• Hindi Desktop locale #9733
• Russian language support #9406
• Turkish desktop locale #9392

✨ Features

• xAI SuperGrok OAuth subscription provider #9420
• Replay ACP images on session load #9496
• Expose raw provider supported models over ACP #9475
• ACP session system prompt setter #9478
• Scholar Sidekick MCP extension #9433
• tui feature flag to gate the TUI command #9428
• CLI to list skills with token counts #9326
• Perplexity as a declarative OpenAI-compatible provider #9324
• Alibaba (Qwen via DashScope) declarative provider #9443
• Configurable tool output size limit via GOOSE_MAX_TOOL_RESPONSE_SIZE #9256
• /model slash command to CLI for session model switching #8747
• Databricks AI Gateway provider #9274
• Russian language support #9406
• Turkish desktop locale #9392
• TUI command on goose-cli #9385
• Configurable MAX_CODE_BLOCK_LINES via env vars #9301
• NEAR AI Cloud provider #9352
• goose://resume session deep link #9343
• Scaleway provider #9254
• GOOSE_FAST_MODEL env var support in ModelConfig::with_fast #9296
• Harbor eval runner #9138
• Slash commands (built-in, skill, recipe) in ACP server #9238
• /goal command for agent self-evaluation before finishing #9069
• Summon subagent instructions #9325
• Open-plugins generalization + skills #9112
• PreToolUse denial hooks #9304
• ACP session list pagination #9199
• goose review local code review command #9114
• TUI diff viewer #9260
• goose://new-session deep link to open fresh chat #9196
• Atomic Chat as declarative OpenAI-compatible provider #9210
• GOOSE_OAUTH_CALLBACK_PORT for stable OAuth redirect_uri #9209
• Proactive OAuth token refresh to avoid re-auth on every session #8386
• GPT-5.5 added to ChatGPT Codex known models #9292
• Opt-in to Vercel AI gateway leaderboard #9259
• ACP pass session cwd param to providers #9229
• Routstr as a declarative provider #9175
• FuturMix provider #8840
• oMLX declarative provider #9177
• Strip chain-of-thought markers from custom provider output #8635
• Encrypted Nostr session sharing #8922
• Simplified Chinese (zh-CN) translation #8765
• Nushell terminal and completion support #8628
• Optional --parameters to scheduled recipe #8741
• GitHub Copilot /responses API #9043
• Vercel AI Gateway provider #9144
• Hooks system #9093
• Projects as backend sources with system prompt injection #8739
• Linux Vulkan support for local inference #9038
• ACP streamable HTTP spec compliance #9034
• Group consecutive tool calls into one summarized chain card #8995
• Quarterly option for scheduler #9076

🐛 Bug Fixes

• Forward custom headers through OAuth connect path #9388
• Scheduled recipe session params #9553
• Skip OTLP signals when protocol=grpc to avoid background-thread panic #9512
• Refresh provider list in Switch Models picker #9408
• Start new chat in current window from recipe param modal #9422
• Make Azure api-version query param optional #9221
• Preserve pending env vars in Add Extension form #9285
• Tolerate missing responses output #9449
• Build linux x86_64 standard inside manylinux_2_28 for glibc 2.28+ compat #9415
• Stop the main window growing taller on every launch #9409
• Desktop chat search session limiting #9366
• Serialize per-session agent creation to stop duplicate MCP init #9357
• Forward ACP server context window size to clients #9455
• Use context limit from /model/info for custom LiteLLM models #9303
• Preserve thinking content for provider context #9314
• Mention configurable timeout env vars in Ollama stream stall error #9246
• Include full recipe parameter details in load/discovery output #9233
• Remove unused fetch-metadata IPC handler (SSRF) #9340
• Respect GOOSE_MAX_TURNS in gateway sessions #9354
• Send empty object instead of null for Anthropic tool_use input #9355
• Preserve selected branch across project chats #9010
• Use current_exe() instead of PATH lookup when spawning goose #9236
• Set TCP_USER_TIMEOUT on streamable HTTP clients #9207
• Activate custom provider after adding via configure #9213
• Reduce excessive MISSING_TRANSLATION warnings for fallback locales #9294
• Eliminate cross-window deep link contamination #9273
• Improve Telegram gateway error reporting and connection reliability #9223
• Enable VT processing on Windows Console Host #9248
• Zero out cost for local providers (ollama, local) #9219
• Prevent tool-use marker leakage in toolshim output #8310
• Re-apply canonical limits when delegate overrides model #9183
• Report cumulative total_tokens in stream-json/json output #8910
• Refresh GCP metadata server token on expiration #8929
• Convert quoted numeric config values to numbers if needed #8844
• Keep SSE reconnect loop alive on long disconnects #8846
• Apply request_params to outgoing OpenAI API payload #9151
• Gemini 3.x known_location Global routing and KNOWN_MODELS entries #9142
• Honor Retry-After on 429 responses #9161
• Coalesce streaming Thinking deltas + list available tools on not-found #9162
• Preserve user-set session name for recipe-based chats #9079
• Enforce exp independently of MAX_TOKEN_AGE_SECONDS in OIDC proxy #8839
• Improve inline code contrast in light theme #9058
• Omit max_tokens for OpenAI-compatible requests when unset #9123
• Set correct mcp-app host capabilities #9116
• Normalize nullable schemas for Vertex Gemini compatibility #8930
• Isolate GitHub recipe temp paths #8878
• Set X-Initiator header on GitHub Copilot requests #8809
• Honor dynamic_models: false in declarative provider configs #8795
• Cache trailing message for stable Bedrock prefix across agent turns #8916
• Case-insensitive model name lookup for context_limit #8906
• Emit cumulative token usage in stream-json/json complete event #8870
• Return 400 instead of panicking on invalid CSP header value #8810
• Coalesce streaming chunks under one ACP message id #8788
• Reuse goose2 vite server on port conflict #8722
• Correct WSL2 OS detection by removing PWD-based Windows override #8869
• Handle Bedrock ReasoningContent blocks gracefully #8843
• Insert tool pair summaries at chronological position in conversation #9087
• Enable SQLite foreign key enforcement in session pool #9121
• Deleted chat session still appears in sidebar list #8674
• Require extension_name on read_resource #8989
• Queued message lost when clicking sending now #9074
• Keep model picker and session config in sync #9048
• Read prompted file parameters in recipes #8504
• Accept null tool_call arguments in OpenAI streaming chunks #9035
• Synchronously reap ACP child to avoid SIGCHLD race #9023
• Elicitation fixes #8999
• Persist accumulated cost in session DB to survive reload #9191
• Check file fallback when keyring has no entry #9279
• ScheduleModal error message styling #9278
• Align sidebar hamburger in macOS fullscreen #9257
• Emit trace_output as span attribute instead of event #9255
• Replace Venice custom provider with declarative config #9234
• Use mkdir -p for self-test workspace initialization #9247
• Show tool name in approval prompt #9216
• Resolve Azure CLI on Windows by using az.cmd #9215
• Handle non-interactive terminal in goose configure on Windows #9214
• Ensure parallel tool image responses don't interleave tool results for Databricks #9241
• Use plain '> ' prompt instead of goose emoji #9305
• Stop killing goosed when a window closes #9302
• Use appInfo instead of clientInfo in MCP Apps init handshake #9249

🔧 Improvements

• Honor blocking Stop hook decisions #9468
• Local inference: stricter GGUF requirements, auto detection of tool calling support, fixed thinking output parsing #9442
• Prefer goose aliases for Databricks v2 inventory #9430
• Simplify UI customization #9353
• Build summon instructions per turn #9329
• Optional api_key configuration for declarative openai-engine providers #9202
• Flag for login shell PATH #9313
• Remove popular chat topics from new chat screen #9307
• Linux musl CLI builds #9240
• Structured per-provider config block, non-destructive provider switching #8977
• Flush OTLP traces reliably on exit with configurable timeout #9228
• Better parsing of pasted HTML as markdown so agents understand #9190
• Prompt injection mitigation: update pattern-based detection #9198
• Prompt injection mitigation: update pattern-based detection to reduce FPs #9350
• Surface resolved Databricks model metadata #9206
• Unified thinking effort control across all providers #9242
• Linux desktop Vulkan packages #9323
• Preserve thinking content for providers that require it #8857
• Dynamically refresh skill instructions each turn #9217
• Build non-Vulkan Linux variants using Ubuntu 22.04 #9211
• Include request URL in provider error messages #9232
• Worktree-aware directory switcher #8450
• Chat history search feature to navigation panel #8448
• Provider-first onboarding #9039
• Improvements to LM Studio declarative provider #8973
• Location column in CLI skills table #8785
• GOOSE_DISABLE_TOOL_CALL_SUMMARY to opt out of per-tool-call summaries #8947
• Localize hardcoded strings in provider settings UI #8931
• Move settings into app shell #9047
• Parallelize provider resolution and eagerly init SQLite pool #8899
• Deduplicate _goose/providers/list RPC call at startup #8873
• Skills library redesign #8868
• TUI spacing/layout improvements #9243
• Polish sidebar and context panel #9059
• Polish inline code snippet styling #9011
• Polish sidebar navigation and project icons #8896
• Extensions page redesign #8940
• Mergeable configs + cleanup #8378
• Consolidate logging setup into shared helper in goose crate #8817
• Replace lazy_static with std::sync::LazyLock #8815
• DEFAULT_PROVIDER_TIMEOUT_SECS constant for providers #8816
• Remove vendored Windows binaries #9318
• Use native arm64 runners for Linux artifact builds #9075

📚 Documentation

• Scaleway provider documentation #9423
• Guide for connecting goose Desktop to a remote goosed server #9275
• SaladCloud AI Gateway provider documentation #9253
• Hooks guide #9288
• Document summon extension requirement for delegate and load tools #9231
• Blog post for hooks feature #9227
• Documentation reorganization #9310
• Tunnel remote access documentation update #9077
• Orchestrating with goose blog post #9104
• Blog post on findings with minimax and office QA benchmark #8984
• Blog post on goose with peekaboo #8884
• Built-in Local Inference blog post #8808

✨ Features

• TUI command on goose-cli #9385
• TUI diff viewer #9260
• goose review local code review command #9114
• /goal command for agent self-evaluation before finishing #9069
• Hooks system for PreToolUse denial and extensibility #9093, #9304
• Slash commands (built-in, skill, recipe) in ACP server #9238
• Open-plugins generalization + skills #9112
• Summon subagent instructions #9325
• Unified thinking effort control across all providers #9242
• NEAR AI Cloud provider #9352
• Scaleway provider #9254
• Vercel AI Gateway provider #9144
• Atomic Chat declarative OpenAI-compatible provider #9210
• Routstr declarative provider #9175
• FuturMix provider #8840
• oMLX declarative provider #9177
• Harbor eval runner #9138
• goose://resume and goose://new-session deep links #9343, #9196
• MAX_CODE_BLOCK_LINES configurable via env vars #9301
• Honor GOOSE_FAST_MODEL env var in ModelConfig #9296
• GPT-5.5 in known models #9292
• Paginate ACP session list #9199
• Pass session cwd param to ACP providers #9229
• Opt-in to Vercel AI gateway leaderboard #9259
• GOOSE_OAUTH_CALLBACK_PORT for stable OAuth redirect_uri #9209
• Proactive OAuth token refresh to avoid re-auth on every session #8386
• Strip chain-of-thought markers from custom provider output #8635
• Encrypted Nostr session sharing #8922
• Simplified Chinese (zh-CN) translation #8765
• Russian language support #9406
• Turkish desktop locale #9392
• Nushell terminal and completion support #8628
• Optional --parameters flag for scheduled recipes #8741
• Quarterly option for scheduler #9076
• GitHub Copilot /responses API support #9043
• Worktree-aware directory switcher #8450
• Chat history search feature #8448
• Projects as backend sources with system prompt injection #8739
• Linux Vulkan support for local inference #9038
• Linux musl CLI builds #9240
• Windows CUDA release artifacts #8750
• GOOSE_DISABLE_TOOL_CALL_SUMMARY env var #8947

🐛 Bug Fixes

• Tolerate missing responses output #9449
• Stop main window growing taller on every launch #9409
• Desktop chat search session limiting #9366
• Serialize per-session agent creation to stop duplicate MCP init #9357
• Respect GOOSE_MAX_TURNS in gateway sessions #9354
• Send empty object instead of null for Anthropic tool_use input #9355
• Preserve thinking content for provider context #9314
• Mention configurable timeout env vars in Ollama stream stall error #9246
• Include full recipe parameter details in load/discovery output #9233
• Remove unused fetch-metadata IPC handler (SSRF) #9340
• Use context limit from /model/info for LiteLLM custom models #9303
• Plain > prompt instead of goose emoji in CLI #9305
• Stop killing goosed when a window closes #9302
• Emit trace_output as span attribute instead of event #9255
• Check file fallback when keyring has no entry #9279
• Align sidebar hamburger in macOS fullscreen #9257
• Use current_exe() instead of PATH lookup when spawning goose #9236
• Set TCP_USER_TIMEOUT on streamable HTTP clients #9207
• Activate custom provider after adding via configure #9213
• Flush OTLP traces reliably on exit with configurable timeout #9228
• Reduce excessive MISSING_TRANSLATION warnings for fallback locales #9294
• Eliminate cross-window deep link contamination #9273
• Improve Telegram gateway error reporting and connection reliability #9223
• Enable VT processing on Windows Console Host #9248
• Zero out cost for local providers (ollama, local) #9219
• Resolve Azure CLI on Windows by using az.cmd #9215
• Handle non-interactive terminal in goose configure on Windows #9214
• Persist accumulated cost in session DB to survive reload #9191
• Prevent tool-use marker leakage in toolshim output #8310
• Re-apply canonical limits when delegate overrides model #9183
• Report cumulative total_tokens in stream-json/json output #8910
• Refresh GCP metadata server token on expiration #8929
• Apply request_params to outgoing OpenAI API payload #9151
• Gemini 3.x known_location Global routing entries #9142
• Coalesce streaming Thinking deltas + list available tools on not-found #9162
• Honor Retry-After on 429 responses #9161
• Preserve user-set session name for recipe-based chats #9079
• Enforce exp independently of MAX_TOKEN_AGE_SECONDS in OIDC proxy #8839
• Improve inline code contrast in light theme #9058
• Omit max_tokens for OpenAI-compatible requests when unset #9123
• Set correct MCP-app host capabilities #9116
• Normalize nullable schemas for Vertex Gemini compatibility #8930
• Isolate GitHub recipe temp paths #8878
• Set X-Initiator header on GitHub Copilot requests #8809
• Honor dynamic_models: false in declarative provider configs #8795
• Cache trailing message for stable prefix across Bedrock agent turns #8916
• Case-insensitive model name lookup for context_limit #8906
• Handle Bedrock ReasoningContent blocks gracefully #8843
• Ensure parallel tool image responses don't interleave on Databricks #9241
• Insert tool pair summaries at chronological position in conversation #9087
• Enable SQLite foreign key enforcement in session pool #9121
• Deleted chat session still appears in sidebar list #8674
• Correct WSL2 OS detection by removing PWD-based Windows override #8869
• Accept null tool_call arguments in OpenAI streaming chunks #9035
• Use python3 in developer extension instructions for macOS/Linux #8784
• Synchronously reap ACP child to avoid SIGCHLD race #9023
• Elicitation fixes #8999
• Return 400 instead of panicking on invalid CSP header value #8810
• Keep SSE reconnect loop alive on long disconnects #8846
• Convert quoted numeric config values to numbers if needed #8844
• Use appInfo instead of clientInfo in MCP Apps init handshake #9249
• Use mkdir -p for self-test workspace initialization #9247
• Exclude preprompt from session title generation #8793
• Prevent login-shell PATH probe from suspending goose on startup #8804

🔧 Improvements

• Local inference: stricter GGUF requirements, auto detection of tool calling support, fixed thinking output parsing #9442
• Simplify UI customization #9353
• Build summon instructions per turn #9329
• Protocol cleanup #9147
• Dependency hygiene #9360
• Preserve selected branch across project chats #9010
• Prompt injection mitigation: update pattern-based detection to reduce FPs #9350
• Surface resolved Databricks model metadata #9206
• Include request URL in provider error messages #9232
• Flag for login shell PATH #9313
• Remove popular chat topics from new chat screen #9307
• Structured per-provider config block, non-destructive provider switching #8977
• TUI spacing/layout improvements #9243
• Preserve thinking content for providers that require it #8857
• Dynamically refresh skill instructions each turn #9217
• Better parsing of pasted HTML as markdown #9190
• Show tool name in approval prompt #9216
• Localize hardcoded strings in provider settings UI #8931
• Move settings into app shell #9047
• Location column in CLI skills table #8785
• Consolidate logging setup into shared helper #8817
• Support optional api_key configuration for declarative openai-engine providers #9202
• Remove vendored Windows binaries #9318
• Polish sidebar and context panel #9059
• Polish inline code snippet styling #9011
• Refresh onboarding when provider catalog loads #9051
• Recipe discovery/execution in ACP server #8925
• Group consecutive tool calls into one summarized chain card #8995
• Replace artifact heuristics/regexes with protocol messages #8996
• Improvements to LM Studio declarative provider #8973
• Parallelize provider resolution and eagerly init SQLite pool #8899
• Deduplicate _goose/providers/list RPC call at startup #8873
• Native arm64 runners for Linux artifact builds #9075
• ACP streamable HTTP spec compliance #9034
• Provider-first onboarding #9039
• Redesign extensions page #8940
• Redesign skills library #8868
• Skills in chat composer #8881
• Mergeable configs cleanup #8378

📚 Documentation

• Stats update #9410
• Hooks guide #9288
• Hooks feature blog post #9227
• Document summon extension requirement for delegate and load tools #9231
• Guide for connecting goose Desktop to a remote goosed server #9275
• SaladCloud AI Gateway provider documentation #9253
• Reorganize documentation #9310
• Fix internal documentation anchors #9094
• Tunnel remote access documentation update #9077
• Linux desktop Vulkan packages #9323
• Orchestrating with goose blog #9104
• Goose with Peekaboo blog #8884
• Built-in local inference blog post #8808
• MiniMax and office QA benchmark blog post #8984
• Repology badge in README #9245