Releases · vybestack/llxprt-code

Release list

Release v0.10.0-nightly.260724.a20e30f14 Latest

Latest

github-actions released this 24 Jul 02:45

v0.10.0-nightly.260724.a20e30f14

a20e30f

Release v0.10.0-nightly.260724.a20e30f14

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

Added Kimi K3, GPT-5.6 Sol/Terra/Luna, Claude Sonnet 5, Claude Opus 4.8 (now default), claude-fable-5, and GLM catalog support across providers
Added LiteLLM provider support for expanded model access
Added configurable reasoning field name support for Ollama streaming
Added /introspect slash command for runtime introspection
Added support for V2 settings format migration
Added semantic response outcome metadata for improved result handling
Exposed tool listing and discovery via the Agent API
Ported Zed integration to the Agent API with improved chat loop handling
Bun-native launcher and runtime improvements: CLI now runs directly from TypeScript via Bun with no-compile launcher, self-exit guards, and Windows PTY/shell hardening
Explicit provider selection is now required at startup to prevent accidental silent fallback
Manual version selection is honored for nightly releases
Subagent streaming text now accumulates correctly instead of replacing tokens
Restored chat rendering with --continue now displays output properly instead of one token per line
Windows fixes: launcher process resolution, shell guidance for quoted paths, CJK input preservation, ripgrep results, named-pipe credential proxy, and terminal hiding during OAuth
OAuth and auth hardening: gated by base URL to prevent wrong-provider prompts, refreshed auth status after toggle, invalidated caches on 401/403, preserved Linux fallback paths, and eliminated per-keystroke keychain prompts in /key autocomplete
Load-balancer and provider resilience: context accounting fixed, failover rotations reset, quota errors preserved across retries, provider identity kept coherent during startup, and context overflow recovery hardened
MCP discovery is now non-fatal and no longer drops input during startup
Compression and streaming stability: masked empty-summary errors no longer abort turns, Kimi K2 special buffering removed, idle timeouts increased and made configurable, stream watchdogs are provider-liveness aware, and whitespace-only subagent stream deltas are preserved
OCR, eval, and notification hardening: nightly eval response extraction fixed, reporting hardened, provider budgeting and redaction improved, and notification writes made idempotent
Security: home paths redacted in assignment-style tokens, ambiguous canonical registry matches fail closed, and Bun install scripts classified by trust

All Changes

Fix nightly eval response extraction (Fixes #2651) (#2652) (a20e30f)
Fix ACP/Zed: task tool crash + todo_pause loop-break (Closes #2653) (#2655) (a5be47a)
Complete OS-standard path migration for OAuth and global state (Fixes #2606) (#2646) (a259657)
Launch installed CLI directly with bundled Bun (Fixes #2603) (#2610) (5868dac)
test(providers): tighten profile fallback regression coverage (#2636) (395efed)
Prevent OAuth browser launch from hiding Windows terminals (#2632) (826a244)
Harden OCR provider budgeting, redaction, and notification-write idempotency (Fixes #2576) (#2631) (83bd764)
Stop per-keystroke keychain prompts in /key autocomplete (Fixes #2620) (#2622) (c28806b)
Add Kimi K3 provider defaults and docs (Fixes #2613) (#2621) (7252612)
Fix Windows launcher process resolution (#2609) (be8f36c)
Make stream watchdogs provider-liveness aware (Fixes #2607) (#2612) (4819f04)
Harden nightly eval reporting (Fixes #2605) (#2611) (2556249)
Harden provider context overflow recovery (Fixes #2588) (#2598) (d58ee53)
Add Windows Bun native-module nightly smoke (Fixes #2301) (#2596) (947c7d3)
Unify session telemetry across provider attempts (Fixes #2585) (#2600) (abd16ea)
Consolidate release guard regression coverage (Fixes #2597) (#2599) (d69bda6)
Preserve whitespace-only subagent stream deltas (Fixes #2555) (#2591) (a1dd642)
Use Bun-native development launcher (Fixes #2342) (#2594) (e3004a8)
Keep guard tests responsive during release checks (Fixes #2584) (#2593) (2b418c1)
Honor manual versions for nightly releases (Fixes #2589) (#2592) (cf296c4)
Require explicit provider selection at startup (Fixes #2481) (#2583) (0d82e78)
Harden subagent review follow-ups (Fixes #2472) (#2582) (8390626)
Add Bun/Vitest compatibility shim for bun test runner (Fixes #2475) (#2484) (91e7462)
Enforce the GenAI enclave boundary (Fixes #2352) (#2546) (2496e89)
Preserve provider quota errors across retry timeouts (Fixes #2559) (#2568) (11328eb)
Finish provider-neutral Gemini cleanup (#2573) (#2574) (30de44a)
Merge pull request #2570 from vybestack/issue2537 (2056a55)
Remove inherited Gemini names from neutral APIs (#2537) (f10865f)
Raise unconfigured subagent turn fallback to 1000 (Fixes #2541) (#2566) (c0c27e8)
Lower DEFAULT_TOKEN_LIMIT to 200K, add GLM catalog, consolidate resolution (Fixes #2527) (#2565) (62a89dc)
good memory (c7b1b78)
test: isolate all package tests from real user storage (fixes #2528) (#2545) (d7e6cb0)
Keep Codex provider identity coherent during startup (Fixes #2544) (#2561) (bf4beab)
Fix restored chat rendering one token per line with --continue (Fixes #2549) (#2558) (6a34605)
Merge pull request #2543 from vybestack/issue5228 (a521510)
Merge remote-tracking branch 'origin/main' into issue5228 (ac29484)
Keep PowerShell parameter checks field-specific (7f38bc7)
Make PowerShell schema tests wording-independent (d165445)
Fix Windows shell guidance for quoted paths (#2538) (d3eefc8)
fix: preserve CJK input and Windows ripgrep results (#2539) (0820185)
Fix Windows shell guidance for quoted paths (fdd9502)
Remove GenAI SDK coupling from leaf packages (Fixes #2351) (#2520) (646bac9)
Make MCP discovery non-fatal and stop dropping input during startup (Fixes #2516) (#2521) (4cfe82b)
Thin the CLI entry chain: runtime assembly behind createAgent, kill deep imports, shrink core import surface (Fixes #2378) (#2493) (c3ec9d3)
refactor(providers): unify OpenAI schema converter and preserve dropped JSON-schema keywords (Fixes #2508) (#2514) (b475247)
Make startup migration and profile loading resilient (Fixes #2476, Fixes #2477) (#2504) (07c96ca)
Merge pull request #2515 from vybestack/issue2509 (e1a8c29)
Merge pull request #2513 from vybestack/issue2505 (21cb698)
Normalize core memory test paths (#2509) (049f779)
Merge pull request #2440 from vybestack/issue2349 (d4d8837)
Fix cross-platform test expectations (#2509) (b3576c8)
fix: preserve valid Responses reasoning IDs (#2349) (056a7ae)
fix(providers): trim reasoning field name and make resolveReasoningValue return undefined when unusable (f7a1043)
Fix Windows test portability (#2509) (5b0a3ea)
test(providers): use fixed timestamp in streaming test chunks (5cd89e2)
fix(providers): add reasoning.fieldName support to classic openai provider (Fixes #2505) (6c68c98)
Merge remote-tracking branch 'origin/main' into issue2349 (d74b84f)
Merge remote-tracking branch 'origin/main' into issue2349 (1a2e0c1)
Merge pull request #2507 from vybestack/issue2501 (e47e14a)
Merge pull request #2503 from vybestack/issue2500 (f70c49d)
test: align retry assertions with neutral content (#2349) (38a0028)
fix: preserve neutral retry and AFC history semantics (#2349) (e000e61)
test(agents): assert exact restored turn in ChatSessionFactory #2500 fold test (5e7ff49)
fix(providers): preserve currentProfile across provider switches (Fixes #2501) (b8192a7)
fix(core): preserve carried history across rebuilds when client not yet initialized (#2500) (dc1adcf)
fix(agents): fold restored history into reused empty HistoryService (Fixes #2500) (b36150d)
fix(ui): show effective model in profile-qualified response header (Fixes #2501) (de9f5f0)
Merge remote-tracking branch 'origin/main' into issue2349 (27cbc7b)
fix: harden neutral message boundaries after review (#2349) (d3d63da)
Reset exhausted load-balancer failover rotations (Fixes #2492) (#2495) (0c99622)
Add GPT-5.6 Sol, Terra, and Luna support (Fixes #2483) (#2497) (63b8855)
feat(providers): support configurable reasoning field name for Ollama streaming (Fixes #2488) (#2491) (ddff241)
fix(ui): accumulate subagent streaming text instead of replacing each token (Fixes #2008) (#2486) (ec14820)
Eliminate inverted core→tools dependency and barrel over-import chain (Fixes #2417) (#2473) (a4e7175)
feat(test): add Bun-backed root test orchestration entry point (Fixes #2463) (#2470) (ae7633d)
fix(launcher): add self-exit guards to prevent orphaned shim processes on terminal/PTY death (Fixes #2468) (#2474) (4fca1a4)
Fail loudly on unavailable profile provider; never save LB runtime as a standard profile (Fixes #2479, Fixes #2478) (#2480) (2d97afd)
fix: exclude prototype keys from executor template property test (#2349) (f1f5eab)
fix: expose placeholder model through core root (#2349) (2fbd53a)
Isolate tmux harness from nested sessions (#2471) (87e133d)
merge main into issue2349 and resolve neutral conflicts (cdf3c33)
fix(agents): activate subagent providers via isolated runtime, restore LB failover, guard Null message (Fixes #2410) (#2415) (e0a0ee8)
fix: keep empty merged content as human speaker (#2349) (12eb3ab)
fix: preserve hook telemetry metadata (#2349) (4ca3fcf)
fix: harden stream telemetry and validation paths (#2349) (a9deb44)
fix: address additional OCR migration findings (#2349) (3bd1558)
fix: address CodeRabbit migration comments (#2349) (6395d50)
fix: address neutral migration review findings (#2349) (1322381)
fix(agents): convert legacy Part-shaped arrays in toPartListUnion (#2...

Assets 3

Release v0.10.0-nightly.260723.a25965779

github-actions released this 23 Jul 06:15

v0.10.0-nightly.260723.a25965779

a259657

Release v0.10.0-nightly.260723.a25965779

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

Add Kimi K3, GPT-5.6 variants, Claude Sonnet 5, Claude Fable 5, GLM models, LiteLLM provider, and Ollama Cloud/Makora aliases
Update built-in provider model versions and defaults, including setting Anthropic's default to Claude Opus 4.8, lowering the default token limit to 200K, adding reasoning field name support for the classic OpenAI provider, and fixing token limits for dated snapshots
Launch installed CLI directly with bundled Bun and introduce no-compile npm runtime with Bun relaunch support
Migrate configuration and OAuth state to OS-standard platform paths (XDG on Linux, etc.)
Make MCP server discovery non-blocking and non-fatal during startup
Expose tool listing and discovery via the Agent API
Add /introspect slash command
Introduce Agent Skills: autonomous activation, /skills reload, CLI management, status bar integration, and tiered discovery
Add extensions install/uninstall commands, settings info, and update notifications
Introduce policy engine with granular shell allowlisting and unified shell security
Add hooks core capabilities (tool input modification, STOP_EXECUTION) and hooks UI (visual indicators, security warnings)
Add environment variable sanitization service and default folder trust to untrusted
Stop per-keystroke keychain prompts in /key autocomplete
Prevent OAuth browser launch from hiding Windows terminals
Fix Windows launcher process resolution, shell guidance for quoted paths, CJK input preservation, and credential proxy transport
Resolve Windows storage crashes and PTY resize crashes
Preserve whitespace-only subagent stream deltas, fix subagent streaming text accumulation, and preserve streaming newlines in assistant output
Make stream watchdogs provider-liveness aware, harden provider context overflow recovery, and make stream idle timeout configurable with a 10-minute default
Retry transient connection errors instead of breaking the stream
Stop masked compression errors from aborting turns and preserve the last user prompt during middle-out compression
Activate subagent providers via isolated runtime; forward systemInstruction to all providers and gate cache_control for z.ai
Inherit parent max turns for subagents and prevent crashes and hangs during subagent launch
Reset exhausted load-balancer failover rotations and fix model identity and context accounting
Fail loudly on unavailable profile providers and never save load-balancer runtime as a standard profile
Gate Anthropic OAuth eligibility by base URL and preserve keyfile profile auth
Harden secure-store and sibling encrypted file stores with machine-secret KDF
Remove discontinued Qwen OAuth provider (Qwen is now API-key-only) and remove google_web_search/google_web_fetch tools
Support JSON Schema draft 2020-12 for MCP tool schemas and make search_file_content respect gitignore and llxprtignore rules
Fix restored chat rendering one token per line with --continue and settings search focus
Validate compression thresholds and fix context-limit application on first profile load
Detect actual image format from magic bytes for Anthropic
Show effective model in profile-qualified response header
Sanitize ANSI escape codes in skill consent prompts and deep merge hooks settings across scopes
Cache CLI version to display the running version instead of the installed version
Surface Claude Fable 5 classifier refusals as a user-visible notice
Remove deprecated Gemini Code Assist OAuth login; route Gemini auth through API key + Vertex
Classify Bun install script trust for improved security
Harden OCR provider budgeting, redaction, and notification-write idempotency
Preserve provider quota errors across retry timeouts

All Changes

Complete OS-standard path migration for OAuth and global state (Fixes #2606) (#2646) (a259657)
Launch installed CLI directly with bundled Bun (Fixes #2603) (#2610) (5868dac)
test(providers): tighten profile fallback regression coverage (#2636) (395efed)
Prevent OAuth browser launch from hiding Windows terminals (#2632) (826a244)
Harden OCR provider budgeting, redaction, and notification-write idempotency (Fixes #2576) (#2631) (83bd764)
Stop per-keystroke keychain prompts in /key autocomplete (Fixes #2620) (#2622) (c28806b)
Add Kimi K3 provider defaults and docs (Fixes #2613) (#2621) (7252612)
Fix Windows launcher process resolution (#2609) (be8f36c)
Make stream watchdogs provider-liveness aware (Fixes #2607) (#2612) (4819f04)
Harden nightly eval reporting (Fixes #2605) (#2611) (2556249)
Harden provider context overflow recovery (Fixes #2588) (#2598) (d58ee53)
Add Windows Bun native-module nightly smoke (Fixes #2301) (#2596) (947c7d3)
Unify session telemetry across provider attempts (Fixes #2585) (#2600) (abd16ea)
Consolidate release guard regression coverage (Fixes #2597) (#2599) (d69bda6)
Preserve whitespace-only subagent stream deltas (Fixes #2555) (#2591) (a1dd642)
Use Bun-native development launcher (Fixes #2342) (#2594) (e3004a8)
Keep guard tests responsive during release checks (Fixes #2584) (#2593) (2b418c1)
Honor manual versions for nightly releases (Fixes #2589) (#2592) (cf296c4)
Require explicit provider selection at startup (Fixes #2481) (#2583) (0d82e78)
Harden subagent review follow-ups (Fixes #2472) (#2582) (8390626)
Add Bun/Vitest compatibility shim for bun test runner (Fixes #2475) (#2484) (91e7462)
Enforce the GenAI enclave boundary (Fixes #2352) (#2546) (2496e89)
Preserve provider quota errors across retry timeouts (Fixes #2559) (#2568) (11328eb)
Finish provider-neutral Gemini cleanup (#2573) (#2574) (30de44a)
Merge pull request #2570 from vybestack/issue2537 (2056a55)
Remove inherited Gemini names from neutral APIs (#2537) (f10865f)
Raise unconfigured subagent turn fallback to 1000 (Fixes #2541) (#2566) (c0c27e8)
Lower DEFAULT_TOKEN_LIMIT to 200K, add GLM catalog, consolidate resolution (Fixes #2527) (#2565) (62a89dc)
good memory (c7b1b78)
test: isolate all package tests from real user storage (fixes #2528) (#2545) (d7e6cb0)
Keep Codex provider identity coherent during startup (Fixes #2544) (#2561) (bf4beab)
Fix restored chat rendering one token per line with --continue (Fixes #2549) (#2558) (6a34605)
Merge pull request #2543 from vybestack/issue5228 (a521510)
Merge remote-tracking branch 'origin/main' into issue5228 (ac29484)
Keep PowerShell parameter checks field-specific (7f38bc7)
Make PowerShell schema tests wording-independent (d165445)
Fix Windows shell guidance for quoted paths (#2538) (d3eefc8)
fix: preserve CJK input and Windows ripgrep results (#2539) (0820185)
Fix Windows shell guidance for quoted paths (fdd9502)
Remove GenAI SDK coupling from leaf packages (Fixes #2351) (#2520) (646bac9)
Make MCP discovery non-fatal and stop dropping input during startup (Fixes #2516) (#2521) (4cfe82b)
Thin the CLI entry chain: runtime assembly behind createAgent, kill deep imports, shrink core import surface (Fixes #2378) (#2493) (c3ec9d3)
refactor(providers): unify OpenAI schema converter and preserve dropped JSON-schema keywords (Fixes #2508) (#2514) (b475247)
Make startup migration and profile loading resilient (Fixes #2476, Fixes #2477) (#2504) (07c96ca)
Merge pull request #2515 from vybestack/issue2509 (e1a8c29)
Merge pull request #2513 from vybestack/issue2505 (21cb698)
Normalize core memory test paths (#2509) (049f779)
Merge pull request #2440 from vybestack/issue2349 (d4d8837)
Fix cross-platform test expectations (#2509) (b3576c8)
fix: preserve valid Responses reasoning IDs (#2349) (056a7ae)
fix(providers): trim reasoning field name and make resolveReasoningValue return undefined when unusable (f7a1043)
Fix Windows test portability (#2509) (5b0a3ea)
test(providers): use fixed timestamp in streaming test chunks (5cd89e2)
fix(providers): add reasoning.fieldName support to classic openai provider (Fixes #2505) (6c68c98)
Merge remote-tracking branch 'origin/main' into issue2349 (d74b84f)
Merge remote-tracking branch 'origin/main' into issue2349 (1a2e0c1)
Merge pull request #2507 from vybestack/issue2501 (e47e14a)
Merge pull request #2503 from vybestack/issue2500 (f70c49d)
test: align retry assertions with neutral content (#2349) (38a0028)
fix: preserve neutral retry and AFC history semantics (#2349) (e000e61)
test(agents): assert exact restored turn in ChatSessionFactory #2500 fold test (5e7ff49)
fix(providers): preserve currentProfile across provider switches (Fixes #2501) (b8192a7)
fix(core): preserve carried history across rebuilds when client not yet initialized (#2500) (dc1adcf)
fix(agents): fold restored history into reused empty HistoryService (Fixes #2500) (b36150d)
fix(ui): show effective model in profile-qualified response header (Fixes #2501) (de9f5f0)
Merge remote-tracking branch 'origin/main' into issue2349 (27cbc7b)
fix: harden neutral message boundaries after review (#2349) (d3d63da)
Reset exhausted load-balancer failover rotations (Fixes #2492) (#2495) (0c99622)
Add GPT-5.6 Sol, Terra, and Luna support (Fixes #2483) (#2497) (63b8855)
feat(providers): support configurable reasoning field name for Ollama streaming (Fixes #2488) (#2491) (ddff241)
fix(ui): accumulate subagent streaming text instead of replacing each token (Fixes #2008) (#2486) (ec14820)
Eliminate inverted core→tools dependency and barrel over-import chain (Fixes #2417) (#2473) (a4e7175)
feat(test): add Bun-backed root test orchestration entry point (Fixes #2463) (#2470) (ae7633d)
fix(launcher): add self-exit ...

Assets 3

Release v0.10.0-nightly.260723.5868dac61

github-actions released this 23 Jul 02:53

v0.10.0-nightly.260723.5868dac61

5868dac

Release v0.10.0-nightly.260723.5868dac61

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

Bundled Bun runtime — CLI now launches directly with its bundled Bun runtime, eliminating the Node.js installation requirement (Fixes #2603, #2243)
New models and providers — Added GPT-5.6 Sol/Terra/Luna (Fixes #2483), Claude Sonnet 5 (Fixes #2289), Claude Fable 5 (Fixes #2328), Kimi K3 (Fixes #2613), LiteLLM provider (#2394), Ollama Cloud and Makora aliases (Fixes #2056), and the GLM model catalog (Fixes #2527)
Windows reliability — Fixed OAuth browser launch hiding Windows terminals (#2632), Windows launcher process resolution (#2609), shell guidance for quoted paths (#2538), CJK input and ripgrep result preservation (#2539), /copy crash prevention (#15657), and PTY resize crash (#15757)
Streaming and response fixes — Preserved whitespace-only subagent stream deltas (Fixes #2555), fixed subagent streaming text accumulation (Fixes #2008), resolved one-token-per-line rendering with --continue (Fixes #2549), preserved Anthropic stopReason deltas (Fixes #1856), and fixed Kimi tool call ID collisions and buffering hangs (Fixes #1872, #1890)
Auth and OAuth — Stopped per-keystroke keychain prompts in /key autocomplete (Fixes #2620), gated OAuth eligibility by base URL to prevent unwanted prompts for third-party gateways (Fixes #2411), preserved Linux auth fallback paths (Fixes #1895), fixed Anthropic OAuth 401 retry loops (Fixes #2035), and added disk-read on lock timeout to prevent cross-process token replay (Fixes #1781)
Subagent improvements — Subagents now inherit parent max turns (Fixes #2048), launch without crashes or hangs (Fixes #1838), filter disabled tools instead of throwing runtime errors (Fixes #1747), preserve tool defaults (Fixes #2069), and activate via isolated runtime with restored load-balancer failover (Fixes #2410)
Load balancer and provider robustness — Reset exhausted failover rotations (Fixes #2492), made failover retryable so all-429 responses do not fatally drop to prompt (Fixes #2450), preserved currentProfile across provider switches (Fixes #2501), and fixed context accounting and profile auth isolation (Fixes #2207, #2492)
Compression and context handling — Stopped masked empty-summary errors from aborting turns (Fixes #2333), retried compression on provider overload and rate-limit errors (Fixes #2045), preserved the last user prompt during middle-out compression (fixes #1650), converted media blocks to text placeholders during compression (Fixes #1875), and enforced hard-limit compression fallback with diagnostics (Fixes #1791)
Configuration and startup — Require explicit provider selection at startup (Fixes #2481), made startup migration and profile loading resilient (Fixes #2476, Fixes #2477), preserved keyfile profile auth in interactive startup (Fixes #2364), migrated configuration to platform-standard paths (Fixes #1465), and added configurable max heap setting (Fixes #1624)
Agent API and extensibility — Published the public Core Agent API surface (Fixes #1594), exposed tool listing and discovery via the Agent API (Fixes #2376), added native support for agents in settings.json, enabled subagents as extensions, and added hooks support for STOP_EXECUTION, tool input modification, and mcp_context in BeforeTool/AfterTool events
Agent Skills — Built-in agent skills are now enabled by default, with /skills reload, status bar skill counts, autonomous activation, conflict detection, security consent for installation, and extension support
CLI commands and UX — Added /introspect slash command (#17048), /chat debug command, extension install/uninstall commands with update notifications, settings descriptions, Esc-Esc to clear prompt, unified shell confirmation dialogs, improved keybindings, simplified paste handling, and undeprecated the --prompt flag
Performance and stability — Optimized UI text buffer and highlighting for large inputs (#16782), added OOM prevention in file search, converted conversation storage to async I/O (Fixes #1988), and serialized concurrent ConversationFileWriter writes to prevent data corruption (Fixes #1988)

All Changes

Launch installed CLI directly with bundled Bun (Fixes #2603) (#2610) (5868dac)
test(providers): tighten profile fallback regression coverage (#2636) (395efed)
Prevent OAuth browser launch from hiding Windows terminals (#2632) (826a244)
Harden OCR provider budgeting, redaction, and notification-write idempotency (Fixes #2576) (#2631) (83bd764)
Stop per-keystroke keychain prompts in /key autocomplete (Fixes #2620) (#2622) (c28806b)
Add Kimi K3 provider defaults and docs (Fixes #2613) (#2621) (7252612)
Fix Windows launcher process resolution (#2609) (be8f36c)
Make stream watchdogs provider-liveness aware (Fixes #2607) (#2612) (4819f04)
Harden nightly eval reporting (Fixes #2605) (#2611) (2556249)
Harden provider context overflow recovery (Fixes #2588) (#2598) (d58ee53)
Add Windows Bun native-module nightly smoke (Fixes #2301) (#2596) (947c7d3)
Unify session telemetry across provider attempts (Fixes #2585) (#2600) (abd16ea)
Consolidate release guard regression coverage (Fixes #2597) (#2599) (d69bda6)
Preserve whitespace-only subagent stream deltas (Fixes #2555) (#2591) (a1dd642)
Use Bun-native development launcher (Fixes #2342) (#2594) (e3004a8)
Keep guard tests responsive during release checks (Fixes #2584) (#2593) (2b418c1)
Honor manual versions for nightly releases (Fixes #2589) (#2592) (cf296c4)
Require explicit provider selection at startup (Fixes #2481) (#2583) (0d82e78)
Harden subagent review follow-ups (Fixes #2472) (#2582) (8390626)
Add Bun/Vitest compatibility shim for bun test runner (Fixes #2475) (#2484) (91e7462)
Enforce the GenAI enclave boundary (Fixes #2352) (#2546) (2496e89)
Preserve provider quota errors across retry timeouts (Fixes #2559) (#2568) (11328eb)
Finish provider-neutral Gemini cleanup (#2573) (#2574) (30de44a)
Merge pull request #2570 from vybestack/issue2537 (2056a55)
Remove inherited Gemini names from neutral APIs (#2537) (f10865f)
Raise unconfigured subagent turn fallback to 1000 (Fixes #2541) (#2566) (c0c27e8)
Lower DEFAULT_TOKEN_LIMIT to 200K, add GLM catalog, consolidate resolution (Fixes #2527) (#2565) (62a89dc)
good memory (c7b1b78)
test: isolate all package tests from real user storage (fixes #2528) (#2545) (d7e6cb0)
Keep Codex provider identity coherent during startup (Fixes #2544) (#2561) (bf4beab)
Fix restored chat rendering one token per line with --continue (Fixes #2549) (#2558) (6a34605)
Merge pull request #2543 from vybestack/issue5228 (a521510)
Merge remote-tracking branch 'origin/main' into issue5228 (ac29484)
Keep PowerShell parameter checks field-specific (7f38bc7)
Make PowerShell schema tests wording-independent (d165445)
Fix Windows shell guidance for quoted paths (#2538) (d3eefc8)
fix: preserve CJK input and Windows ripgrep results (#2539) (0820185)
Fix Windows shell guidance for quoted paths (fdd9502)
Remove GenAI SDK coupling from leaf packages (Fixes #2351) (#2520) (646bac9)
Make MCP discovery non-fatal and stop dropping input during startup (Fixes #2516) (#2521) (4cfe82b)
Thin the CLI entry chain: runtime assembly behind createAgent, kill deep imports, shrink core import surface (Fixes #2378) (#2493) (c3ec9d3)
refactor(providers): unify OpenAI schema converter and preserve dropped JSON-schema keywords (Fixes #2508) (#2514) (b475247)
Make startup migration and profile loading resilient (Fixes #2476, Fixes #2477) (#2504) (07c96ca)
Merge pull request #2515 from vybestack/issue2509 (e1a8c29)
Merge pull request #2513 from vybestack/issue2505 (21cb698)
Normalize core memory test paths (#2509) (049f779)
Merge pull request #2440 from vybestack/issue2349 (d4d8837)
Fix cross-platform test expectations (#2509) (b3576c8)
fix: preserve valid Responses reasoning IDs (#2349) (056a7ae)
fix(providers): trim reasoning field name and make resolveReasoningValue return undefined when unusable (f7a1043)
Fix Windows test portability (#2509) (5b0a3ea)
test(providers): use fixed timestamp in streaming test chunks (5cd89e2)
fix(providers): add reasoning.fieldName support to classic openai provider (Fixes #2505) (6c68c98)
Merge remote-tracking branch 'origin/main' into issue2349 (d74b84f)
Merge remote-tracking branch 'origin/main' into issue2349 (1a2e0c1)
Merge pull request #2507 from vybestack/issue2501 (e47e14a)
Merge pull request #2503 from vybestack/issue2500 (f70c49d)
test: align retry assertions with neutral content (#2349) (38a0028)
fix: preserve neutral retry and AFC history semantics (#2349) (e000e61)
test(agents): assert exact restored turn in ChatSessionFactory #2500 fold test (5e7ff49)
fix(providers): preserve currentProfile across provider switches (Fixes #2501) (b8192a7)
fix(core): preserve carried history across rebuilds when client not yet initialized (#2500) (dc1adcf)
fix(agents): fold restored history into reused empty HistoryService (Fixes #2500) (b36150d)
fix(ui): show effective model in profile-qualified response header (Fixes #2501) (de9f5f0)
Merge remote-tracking branch 'origin/main' into issue2349 (27cbc7b)
fix: harden neutral message boundaries after review (#2349) (d3d63da)
Reset exhausted load-balancer failover rotations (Fixes #2492) (#2495) (0c99622)
Add GPT-5.6 Sol, Terra, and Luna support (Fixes #2483) (#2497) (63b8855)
feat(providers): support configurable reasoning field name for Ollama streaming (Fixes #2488) (#2491) (ddff241)
fix(ui): accumulate subagent streaming text instead of replacing eac...

Assets 3

Release v0.10.0-nightly.260722.be8f36c6e

github-actions released this 22 Jul 02:37

v0.10.0-nightly.260722.be8f36c6e

be8f36c

Release v0.10.0-nightly.260722.be8f36c6e

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

Added GPT-5.6 Sol, Terra, and Luna model support
Added Claude Sonnet 5 to the model menu and set Claude Opus 4.8 as the Anthropic default
Added Claude Fable 5 classifier support and surfaced refusals as a user-visible notice
Added GLM model catalog and consolidated model resolution
Added Ollama Cloud and Makora provider aliases
Added LiteLLM provider support
Removed discontinued Qwen OAuth provider; Qwen is now API-key-only
Removed deprecated Gemini Code Assist OAuth login; Gemini auth now routes through API key and Vertex
Require explicit provider selection at startup to avoid ambiguous defaults
Migrated interactive CLI to consume agent.stream() directly instead of assembling the agentic loop
Introduced Agent at the interactive composition root and migrated CLI command surfaces to Agent APIs
Ported Zed integration onto the Agent API and removed the hand-rolled chat loop
Exposed tool listing and discovery via the Agent API
Restored correct chat rendering when using --continue, fixing one-token-per-line output
Fixed subagent streaming text accumulation so assistant output no longer replaces each token
Preserved whitespace-only subagent stream deltas
Fixed provider quota errors so they survive retry timeouts instead of being lost
Hardened provider context overflow recovery
Reset exhausted load-balancer failover rotations automatically
Fixed load-balancer model identity and bounded first model responses
Fixed cross-provider ephemeral leak in mixed load balancers
Retry transient mid-stream connection errors instead of breaking the loop
Retry Anthropic api_error internal server errors instead of breaking the loop
Handle Anthropic rate_limit_error in JSON body like a 429
Fail closed on ambiguous canonical registry matches
Harden stream watchdogs to be provider-liveness aware
Preserve provider terminal metadata and tool schemas through streaming
Drop cross-model thinking blocks on model switch to avoid 400 invalid signature errors
Unify session telemetry across provider attempts
Harden subagent review follow-ups
Keep Codex provider identity coherent during startup
Enforce the GenAI enclave boundary
Remove GenAI SDK coupling from leaf packages and migrate core/CLI/agents to neutral contract types
Made MCP server discovery non-fatal and stopped dropping input during startup
Made MCP discovery non-blocking at startup
Project real MCP OAuth quad-state through the Agent surface
Improved Windows support with a Bun-native development launcher, native-module nightly smoke, named-pipe credential proxy transport, shell guidance for quoted paths, and CJK/ripgrep preservation
Fixed Windows launcher process resolution
Fixed Windows shell guidance for quoted paths
Fixed Windows storage /dev/tty and EINVAL crashes
Preserved keyfile profile auth in interactive startup
Filtered disabled tools instead of throwing runtime errors
Fixed subagent wizard invocation crash
Inherited parent max turns for subagents
Added subagent recovery turns
Added configurable max heap setting
Added /introspect slash command
Added /skills reload and gemini skills CLI management command
Added Agent Skills with tiered discovery, autonomous activation, context injection, and status bar integration
Added environment variable sanitization service
Added security warning to HooksList for project-level hooks
Added explicit stop and block execution control in model hooks
Added mcp_context to BeforeTool and AfterTool hook inputs
Added hook event name validation
Added skill conflict detection and warnings
Added security consent for skill installation
Unified shell confirmation dialogs
Renamed disable settings to enable settings for clarity
Centralized subagent excluded tool governance
Tightened qualified subagent tool whitelist parsing
Added profile-qualified model identity in status bar, diagnostics, and response prefix
Show effective model in profile-qualified response header
Made stream idle timeout configurable and raised default to 10 minutes
Disabled idle timeout watchdog by default
Fixed stream idle timeout races with a shared helper
Fixed idle timeout AbortError suppression and sendPromise deadlock
Fixed Kimi tool call ID collisions and silent scheduler drops
Fixed Kimi K2 special buffering that caused Fireworks hangs
Fixed stream retry after content has already been yielded to consumer
Stopped invalid stream fallthrough retries
Increased idle timeouts and fixed AbortError suppression
Close iterator on abort to prevent sendPromise hang
Ensure first chunk consumption happens inside retry boundary
Re-estimated tokens and fixed cache double-counting on provider switch
Fixed stale context limit after profile load
Fixed context overflow preflight after model switch
Fixed compression threshold handling and preflight guard
Attempt compression when extraction is unsafe instead of throwing
Masked empty-summary compression errors from aborting turns
Preserve last user prompt during middle-out compression
Distinguish no-op, recent, and failed compression outcomes
Allow failover after compression callback failure
Harden hard-limit compression fallback
Harden compression config assertions
Consult provider context limit in preflight guard
Use load-balancer backend context limit for compaction
Fixed load-balancer context accounting
Fixed shared-reference token fallback accounting
Fixed post-merge gaps after config refactor and migration
Migrated configuration to platform-standard paths
Made startup migration and profile loading resilient

All Changes

Fix Windows launcher process resolution (#2609) (be8f36c)
Make stream watchdogs provider-liveness aware (Fixes #2607) (#2612) (4819f04)
Harden nightly eval reporting (Fixes #2605) (#2611) (2556249)
Harden provider context overflow recovery (Fixes #2588) (#2598) (d58ee53)
Add Windows Bun native-module nightly smoke (Fixes #2301) (#2596) (947c7d3)
Unify session telemetry across provider attempts (Fixes #2585) (#2600) (abd16ea)
Consolidate release guard regression coverage (Fixes #2597) (#2599) (d69bda6)
Preserve whitespace-only subagent stream deltas (Fixes #2555) (#2591) (a1dd642)
Use Bun-native development launcher (Fixes #2342) (#2594) (e3004a8)
Keep guard tests responsive during release checks (Fixes #2584) (#2593) (2b418c1)
Honor manual versions for nightly releases (Fixes #2589) (#2592) (cf296c4)
Require explicit provider selection at startup (Fixes #2481) (#2583) (0d82e78)
Harden subagent review follow-ups (Fixes #2472) (#2582) (8390626)
Add Bun/Vitest compatibility shim for bun test runner (Fixes #2475) (#2484) (91e7462)
Enforce the GenAI enclave boundary (Fixes #2352) (#2546) (2496e89)
Preserve provider quota errors across retry timeouts (Fixes #2559) (#2568) (11328eb)
Finish provider-neutral Gemini cleanup (#2573) (#2574) (30de44a)
Merge pull request #2570 from vybestack/issue2537 (2056a55)
Remove inherited Gemini names from neutral APIs (#2537) (f10865f)
Raise unconfigured subagent turn fallback to 1000 (Fixes #2541) (#2566) (c0c27e8)
Lower DEFAULT_TOKEN_LIMIT to 200K, add GLM catalog, consolidate resolution (Fixes #2527) (#2565) (62a89dc)
good memory (c7b1b78)
test: isolate all package tests from real user storage (fixes #2528) (#2545) (d7e6cb0)
Keep Codex provider identity coherent during startup (Fixes #2544) (#2561) (bf4beab)
Fix restored chat rendering one token per line with --continue (Fixes #2549) (#2558) (6a34605)
Merge pull request #2543 from vybestack/issue5228 (a521510)
Merge remote-tracking branch 'origin/main' into issue5228 (ac29484)
Keep PowerShell parameter checks field-specific (7f38bc7)
Make PowerShell schema tests wording-independent (d165445)
Fix Windows shell guidance for quoted paths (#2538) (d3eefc8)
fix: preserve CJK input and Windows ripgrep results (#2539) (0820185)
Fix Windows shell guidance for quoted paths (fdd9502)
Remove GenAI SDK coupling from leaf packages (Fixes #2351) (#2520) (646bac9)
Make MCP discovery non-fatal and stop dropping input during startup (Fixes #2516) (#2521) (4cfe82b)
Thin the CLI entry chain: runtime assembly behind createAgent, kill deep imports, shrink core import surface (Fixes #2378) (#2493) (c3ec9d3)
refactor(providers): unify OpenAI schema converter and preserve dropped JSON-schema keywords (Fixes #2508) (#2514) (b475247)
Make startup migration and profile loading resilient (Fixes #2476, Fixes #2477) (#2504) (07c96ca)
Merge pull request #2515 from vybestack/issue2509 (e1a8c29)
Merge pull request #2513 from vybestack/issue2505 (21cb698)
Normalize core memory test paths (#2509) (049f779)
Merge pull request #2440 from vybestack/issue2349 (d4d8837)
Fix cross-platform test expectations (#2509) (b3576c8)
fix: preserve valid Responses reasoning IDs (#2349) (056a7ae)
fix(providers): trim reasoning field name and make resolveReasoningValue return undefined when unusable (f7a1043)
Fix Windows test portability (#2509) (5b0a3ea)
test(providers): use fixed timestamp in streaming test chunks (5cd89e2)
fix(providers): add reasoning.fieldName support to classic openai provider (Fixes #2505) (6c68c98)
Merge remote-tracking branch 'origin/main' into issue2349 (d74b84f)
Merge remote-tracking branch 'origin/main' into issue2349 (1a2e0c1)
Merge pull request #2507 from vybestack/issue2501 (e47e14a)
Merge pull request #2503 from vybestack/issue2500 (f70c49d)
test: align retry assertions with neutral content (#2349) (38a0028)
fix: preserve...

Assets 3

Release v0.10.0-nightly.260721.d58ee53aa

github-actions released this 21 Jul 02:47

v0.10.0-nightly.260721.d58ee53aa

d58ee53

Release v0.10.0-nightly.260721.d58ee53aa

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

Add support for GPT-5.6 Sol, Terra, and Luna models
Add LiteLLM provider support for broader model compatibility
Add Claude Sonnet 5 and Claude Fable 5 to supported model lists
Add Ollama Cloud and Makora provider aliases
Add GLM model catalog
Set Anthropic default model to Claude Opus 4.8
Remove discontinued Qwen OAuth provider; Qwen now requires API key
Remove deprecated Gemini Code Assist OAuth login; route Gemini auth through API key and Vertex
Require explicit provider selection at startup for clearer setup
Make MCP discovery non-fatal and preserve input during startup
Make stream idle timeout configurable and raise default to 10 minutes
Add configurable max heap setting for performance tuning
Expose tool listing and discovery via the Agent API
Add public Core Agent API surface
Add /introspect slash command
Add extensions install/uninstall commands with update notifications
Add policy engine granular shell allowlisting
Add hooks UI with context injection, visual indicators, and security warnings
Add environment variable sanitization service
Add subagent recovery turns
Bound LSP requests with timeouts
Deprecate and remove OpenTUI experimental UI
Fix restored chat rendering one token per line with --continue
Fix subagent streaming text now accumulates instead of replacing each token
Fix CJK input preservation and Windows ripgrep results
Fix Windows shell guidance for quoted paths
Fix Windows storage 'EINVAL' and '/dev/tty' crashes
Fix Windows crash: "Cannot resize a pty that has already exited"
Fix /copy crash on Windows by skipping /dev/tty
Fix home/end keybinding conflict
Fix PTY focus routing for live shells
Fix noninteractive Ctrl-C hard abort
Fix error-on-cancel and ESC not cancelling subagents
Fix #1877: Subagent armTimeout fires against defunct instance
Fix #1801: Sanitize ANSI escape codes in skill consent prompts
Fix #1802: Deep merge hooksConfig and hooks settings across scopes
Fix #2002: todopause properly pauses the system
Fix #1776: Gate initial prompt on startup guards
Fix #1748: BeforeToolSelectionHook empty array treated as no restriction
Fix Project-level memories not loaded when workspace is subdirectory
Fix #1803: Reset geminiMessageBuffer on contextCleared event during streaming
Fix event-driven model profile display
Fix settings search focus and typing flow
Restore checkmark symbol in CLI UI indicators
Fix --version flag prints version to stdout instead of debug log
Fix context-limit not applied on first profile load
Fix Anthropic rate_limit_error handling in JSON body
Fix client-side 413 errors with retry and tool name guidance
Fix OAuth token cache invalidation on 401/403 revocation
Fix bucket failover during compression 401 errors
Fix profile switch stalls after failed turns
Fix stale context limit after profile load
Fix load-balancer context accounting
Fix cross-provider ephemeral leak in mixed load balancers
Fix load-balancer subagent runtime profiles
Fix subagent: run subagents non-interactively in -p mode and fix dispose crash
Fix subagent: prevent crashes and hangs during subagent launch
Fix subagent wizard invocation crash
Fix subagent disabled tools filtering instead of throwing runtime error
Fix set-command: remove model_param alias, keep only modelparam
Fix set-command: clear ephemeral shadow and accept model_param alias
Fix stats: detect API-key providers with kebab-case base-url config
Fix preserve currentProfile across provider switches
Fix carried history across rebuilds when client not yet initialized
Fix show effective model in profile-qualified response header
Fix reset exhausted load-balancer failover rotations
Fix #2099: Cancelling a tool call from a subagent during parallel tool calls can cause a hang
Fix #2133: apply_patch fails for directory-qualified paths
Fix apply_patch validation of preview replacements
Fix redact home paths in assignment-style tokens
Fix confirm arguments key before truncating partial JSON
Fix compression: use load-balancer backend context limit for compaction
Fix compression: consult provider context limit in preflight guard
Fix compression: attempt compression when extraction is unsafe instead of throwing
Fix compression: distinguish no-op, recent, and failed outcomes
Fix compression: stop masked empty-summary errors from aborting turns
Fix compression: preflight guard attempts compression before bailing
Fix context overflow preflight after model switch
Fix context overflow recovery
Fix stream idle timeout races and sendPromise deadlock
Fix stream: prevent sendPromise deadlock on idle timeout abort
Fix stream: ensure first chunk consumption happens inside retry boundary
Fix stream: close iterator on abort to prevent sendPromise hang
Fix stream: preserve Anthropic stopReason-only deltas and add terminal diagnostics
Fix stream: disable idle timeout watchdog by default
Fix stream: increase idle timeouts and fix AbortError suppression
Fix stream: prevent Kimi tool call ID collisions and silent scheduler drops
Fix subagent streaming newlines in assistant output
Fix Kimi special buffering that caused Fireworks hangs
Fix OAuth bridge lifecycle and cross-process refresh safety
Fix OAuth provider registration on all creation paths and normalize case
Fix Anthropic OAuth 401 retry loop on empty failed token
Fix auth bucket failover during compression 401 errors
Fix load-balancer model identity
Fix load-balancer profile handling
Fix provider-aware rate limit messages
Fix subagent PromptConfig initialMessages contract
Fix OpenAI tool format overrides
Fix Codex device-flow messages through bridge when UI is live
Fix Codex device-code fallback
Fix Codex alias config as model source
Fix local workspace source and guard dual-lockfile parity
Fix workspace build order
Fix Bun relaunch support
Fix Bun workspace-install smoke job
Fix Bun JSON output startup path
Fix Node 20 E2E bundle parse
Fix E2E bundle freshness
Fix git-commit info resiliently so a fresh checkout can't crash startup
Fix Codex provider identity coherent during startup
Fix provider context overflow recovery
Fix empty tool schemas for subagents and streaming pipeline hangs
Fix preserve provider quota errors across retry timeouts
Fix Anthropic api_error retry instead of breaking the loop
Fix load-balancer failover aggregate retryable so all-429 doesn't fatally drop to prompt
Fix shared-reference token fallback accounting
Fix stale turn from clobbering newer turn's isResponding state
Fix settings: validate compression thresholds
Fix settings: gate stream idle profile settings
Fix settings search focus and typing flow
Fix settings: harden parse boundaries
Fix settings: harden system settings path override

All Changes

Harden provider context overflow recovery (Fixes #2588) (#2598) (d58ee53)
Add Windows Bun native-module nightly smoke (Fixes #2301) (#2596) (947c7d3)
Unify session telemetry across provider attempts (Fixes #2585) (#2600) (abd16ea)
Consolidate release guard regression coverage (Fixes #2597) (#2599) (d69bda6)
Preserve whitespace-only subagent stream deltas (Fixes #2555) (#2591) (a1dd642)
Use Bun-native development launcher (Fixes #2342) (#2594) (e3004a8)
Keep guard tests responsive during release checks (Fixes #2584) (#2593) (2b418c1)
Honor manual versions for nightly releases (Fixes #2589) (#2592) (cf296c4)
Require explicit provider selection at startup (Fixes #2481) (#2583) (0d82e78)
Harden subagent review follow-ups (Fixes #2472) (#2582) (8390626)
Add Bun/Vitest compatibility shim for bun test runner (Fixes #2475) (#2484) (91e7462)
Enforce the GenAI enclave boundary (Fixes #2352) (#2546) (2496e89)
Preserve provider quota errors across retry timeouts (Fixes #2559) (#2568) (11328eb)
Finish provider-neutral Gemini cleanup (#2573) (#2574) (30de44a)
Merge pull request #2570 from vybestack/issue2537 (2056a55)
Remove inherited Gemini names from neutral APIs (#2537) (f10865f)
Raise unconfigured subagent turn fallback to 1000 (Fixes #2541) (#2566) (c0c27e8)
Lower DEFAULT_TOKEN_LIMIT to 200K, add GLM catalog, consolidate resolution (Fixes #2527) (#2565) (62a89dc)
good memory (c7b1b78)
test: isolate all package tests from real user storage (fixes #2528) (#2545) (d7e6cb0)
Keep Codex provider identity coherent during startup (Fixes #2544) (#2561) (bf4beab)
Fix restored chat rendering one token per line with --continue (Fixes #2549) (#2558) (6a34605)
Merge pull request #2543 from vybestack/issue5228 (a521510)
Merge remote-tracking branch 'origin/main' into issue5228 (ac29484)
Keep PowerShell parameter checks field-specific (7f38bc7)
Make PowerShell schema tests wording-independent (d165445)
Fix Windows shell guidance for quoted paths (#2538) (d3eefc8)
fix: preserve CJK input and Windows ripgrep results (#2539) (0820185)
Fix Windows shell guidance for quoted paths (fdd9502)
Remove GenAI SDK coupling from leaf packages (Fixes #2351) (#2520) (646bac9)
Make MCP discovery non-fatal and stop dropping input during startup (Fixes #2516) (#2521) (4cfe82b)
Thin the CLI entry chain: runtime assembly behind createAgent, kill deep imports, shrink core import surface (Fixes #2378) (#2493) (c3ec9d3)
refactor(providers): unify OpenAI schema converter and preserve dropped JSON-schema keywords (Fixes #2508) (#2514) (b475247)
Make startup migration and profile loading resilient (Fixes #2476, Fixes #2477) (#2504) (07c96ca)
Merge pull request #2515 from vybestack/issue2509 (e1a8c29)
Merge pull...

Assets 3

Release v0.10.0-nightly.260720.d69bda66a

github-actions released this 20 Jul 03:04

v0.10.0-nightly.260720.d69bda66a

d69bda6

Release v0.10.0-nightly.260720.d69bda66a

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

Add LiteLLM provider support for additional AI model backends
Add GPT-5.6 Sol, Terra, and Luna model support
Add Claude Sonnet 5 and Claude Fable 5 model support
Set Anthropic default model to Claude Opus 4.8
Add Ollama Cloud and Makora provider aliases
Lower default token limit to 200K and add GLM model catalog
Built-in Agent Skills with autonomous activation and context injection
Extensions management: install/uninstall commands, update notifications, and settings info
Policy engine with granular shell allowlisting and unified shell security
Hooks enhancements: stop/block execution control, MCP context for tool hooks, and enable-all/disable-all commands
New slash commands: /introspect and /bug for chat history export
Configurable max heap setting
Profile-qualified model identity in status bar and diagnostics
Require explicit provider selection at startup
Fix --continue flag causing degraded chat rendering (one token per line)
Fix Windows shell guidance for quoted paths and resolve '/dev/tty' crashes
Preserve CJK input and Windows ripgrep results
Fix Windows crash when resizing an already-exited PTY
Fix ESC and error-on-cancel not cancelling subagents
Fix load-balancer failover, context accounting, and profile handling
Fix Anthropic OAuth 401 retry loop on empty failed token
Fix subagent wizard invocation crash
Fix PTY focus routing for live shells
Fix noninteractive Ctrl-C hard abort
Sanitize ANSI escape codes in skill consent prompts
Fix todo pause not pausing todo updates
Surface Claude Fable 5 classifier refusals as a user-visible notice
Preserve whitespace-only subagent stream deltas
Fix subagent armTimeout firing against defunct instance
Filter emojis in todo writes

All Changes

Consolidate release guard regression coverage (Fixes #2597) (#2599) (d69bda6)
Preserve whitespace-only subagent stream deltas (Fixes #2555) (#2591) (a1dd642)
Use Bun-native development launcher (Fixes #2342) (#2594) (e3004a8)
Keep guard tests responsive during release checks (Fixes #2584) (#2593) (2b418c1)
Honor manual versions for nightly releases (Fixes #2589) (#2592) (cf296c4)
Require explicit provider selection at startup (Fixes #2481) (#2583) (0d82e78)
Harden subagent review follow-ups (Fixes #2472) (#2582) (8390626)
Add Bun/Vitest compatibility shim for bun test runner (Fixes #2475) (#2484) (91e7462)
Enforce the GenAI enclave boundary (Fixes #2352) (#2546) (2496e89)
Preserve provider quota errors across retry timeouts (Fixes #2559) (#2568) (11328eb)
Finish provider-neutral Gemini cleanup (#2573) (#2574) (30de44a)
Merge pull request #2570 from vybestack/issue2537 (2056a55)
Remove inherited Gemini names from neutral APIs (#2537) (f10865f)
Raise unconfigured subagent turn fallback to 1000 (Fixes #2541) (#2566) (c0c27e8)
Lower DEFAULT_TOKEN_LIMIT to 200K, add GLM catalog, consolidate resolution (Fixes #2527) (#2565) (62a89dc)
good memory (c7b1b78)
test: isolate all package tests from real user storage (fixes #2528) (#2545) (d7e6cb0)
Keep Codex provider identity coherent during startup (Fixes #2544) (#2561) (bf4beab)
Fix restored chat rendering one token per line with --continue (Fixes #2549) (#2558) (6a34605)
Merge pull request #2543 from vybestack/issue5228 (a521510)
Merge remote-tracking branch 'origin/main' into issue5228 (ac29484)
Keep PowerShell parameter checks field-specific (7f38bc7)
Make PowerShell schema tests wording-independent (d165445)
Fix Windows shell guidance for quoted paths (#2538) (d3eefc8)
fix: preserve CJK input and Windows ripgrep results (#2539) (0820185)
Fix Windows shell guidance for quoted paths (fdd9502)
Remove GenAI SDK coupling from leaf packages (Fixes #2351) (#2520) (646bac9)
Make MCP discovery non-fatal and stop dropping input during startup (Fixes #2516) (#2521) (4cfe82b)
Thin the CLI entry chain: runtime assembly behind createAgent, kill deep imports, shrink core import surface (Fixes #2378) (#2493) (c3ec9d3)
refactor(providers): unify OpenAI schema converter and preserve dropped JSON-schema keywords (Fixes #2508) (#2514) (b475247)
Make startup migration and profile loading resilient (Fixes #2476, Fixes #2477) (#2504) (07c96ca)
Merge pull request #2515 from vybestack/issue2509 (e1a8c29)
Merge pull request #2513 from vybestack/issue2505 (21cb698)
Normalize core memory test paths (#2509) (049f779)
Merge pull request #2440 from vybestack/issue2349 (d4d8837)
Fix cross-platform test expectations (#2509) (b3576c8)
fix: preserve valid Responses reasoning IDs (#2349) (056a7ae)
fix(providers): trim reasoning field name and make resolveReasoningValue return undefined when unusable (f7a1043)
Fix Windows test portability (#2509) (5b0a3ea)
test(providers): use fixed timestamp in streaming test chunks (5cd89e2)
fix(providers): add reasoning.fieldName support to classic openai provider (Fixes #2505) (6c68c98)
Merge remote-tracking branch 'origin/main' into issue2349 (d74b84f)
Merge remote-tracking branch 'origin/main' into issue2349 (1a2e0c1)
Merge pull request #2507 from vybestack/issue2501 (e47e14a)
Merge pull request #2503 from vybestack/issue2500 (f70c49d)
test: align retry assertions with neutral content (#2349) (38a0028)
fix: preserve neutral retry and AFC history semantics (#2349) (e000e61)
test(agents): assert exact restored turn in ChatSessionFactory #2500 fold test (5e7ff49)
fix(providers): preserve currentProfile across provider switches (Fixes #2501) (b8192a7)
fix(core): preserve carried history across rebuilds when client not yet initialized (#2500) (dc1adcf)
fix(agents): fold restored history into reused empty HistoryService (Fixes #2500) (b36150d)
fix(ui): show effective model in profile-qualified response header (Fixes #2501) (de9f5f0)
Merge remote-tracking branch 'origin/main' into issue2349 (27cbc7b)
fix: harden neutral message boundaries after review (#2349) (d3d63da)
Reset exhausted load-balancer failover rotations (Fixes #2492) (#2495) (0c99622)
Add GPT-5.6 Sol, Terra, and Luna support (Fixes #2483) (#2497) (63b8855)
feat(providers): support configurable reasoning field name for Ollama streaming (Fixes #2488) (#2491) (ddff241)
fix(ui): accumulate subagent streaming text instead of replacing each token (Fixes #2008) (#2486) (ec14820)
Eliminate inverted core→tools dependency and barrel over-import chain (Fixes #2417) (#2473) (a4e7175)
feat(test): add Bun-backed root test orchestration entry point (Fixes #2463) (#2470) (ae7633d)
fix(launcher): add self-exit guards to prevent orphaned shim processes on terminal/PTY death (Fixes #2468) (#2474) (4fca1a4)
Fail loudly on unavailable profile provider; never save LB runtime as a standard profile (Fixes #2479, Fixes #2478) (#2480) (2d97afd)
fix: exclude prototype keys from executor template property test (#2349) (f1f5eab)
fix: expose placeholder model through core root (#2349) (2fbd53a)
Isolate tmux harness from nested sessions (#2471) (87e133d)
merge main into issue2349 and resolve neutral conflicts (cdf3c33)
fix(agents): activate subagent providers via isolated runtime, restore LB failover, guard Null message (Fixes #2410) (#2415) (e0a0ee8)
fix: keep empty merged content as human speaker (#2349) (12eb3ab)
fix: preserve hook telemetry metadata (#2349) (4ca3fcf)
fix: harden stream telemetry and validation paths (#2349) (a9deb44)
fix: address additional OCR migration findings (#2349) (3bd1558)
fix: address CodeRabbit migration comments (#2349) (6395d50)
fix: address neutral migration review findings (#2349) (1322381)
fix(agents): convert legacy Part-shaped arrays in toPartListUnion (#2349) (ec3e3ee)
fix: remove unnecessary type assertion in boundaryRecovery test (#2349) (913de60)
fix: remove @google/genai import from issue2410 test (#2349) (32cef38)
fix: normalize additionalContext injection to ContentBlock[] (#2349) (facc294)
fix(hooks): convert hook-modified contents back to IContent via toIContents (38ea315)
fix(format): prettier fix on auto-fixed test file (894ac54)
fix(post-rebase): resolve type errors from main merge (ba7a7d5)
fix(lockfiles): remove @google/genai from agents workspace in lockfiles (cc9f7a4)
fix(core): update tests for neutral IContent/ContentBlock shapes (49b2a3e)
fix(tests): update assertions for neutral ContentBlock shapes (22157f0)
fix(cli): extract isIContentLike to reduce expression complexity (67c504f)
fix(cli): null-check req in loopTestHelpers for strict tsc (5c6d238)
fix(cli): merge duplicate imports to fix lint:ci --max-warnings 0 (4ad1022)
fix(cli): extract loopTestHelpers to fix max-lines lint (e704101)
fix(cli): accept unknown in addPartToResult for AgentRequestInput (19f9097)
fix(agents): fix formatting + extract modelInfoHelpers for max-lines (4ad131b)
fix(agents): address OCR findings + CI lint failures (7ad898b)
fix(agents): correct ToolGroupArray typing for ChatSessionConfig.tools (03e0edc)
fix(cli): use public root import for CheckpointContent (fixes CI lint boundary) (d4f7c23)
fix(agents): P33 fix tool group wrapping for provider compatibility (bd3f10f)
feat(agents): P32 remove @google/genai dependency from agents package.json (d343631)
feat(agents): P29-P31 enforcement gate — anti-#2424 CI wiring + test gate (fe79d9e)
feat(agents): P28 test migration — zero @google/genai imports in agents tests (5875e95)
feat(agents): P27 remaining group retype — ZERO prod @google/genai imports (9507c47)
feat(agents): P26 remaining grou...

Assets 3

Release v0.10.0-nightly.260719.d69bda66a

github-actions released this 19 Jul 02:45

v0.10.0-nightly.260719.d69bda66a

d69bda6

Release v0.10.0-nightly.260719.d69bda66a

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

Added GPT-5.6 Sol, Terra, and Luna model support
Added Claude Sonnet 5 and Claude Fable 5 to the supported model list
Set Claude Opus 4.8 as the default Anthropic model
Added Ollama Cloud and Makora provider aliases
Added GLM model catalog and consolidated model resolution
Added LiteLLM provider support
Added Agent Skills with status bar integration, /skills reload, and extension-based skill loading
Added /introspect slash command for debugging and diagnostics
Added --continue flag support with corrected restored chat rendering
Added MCP server context (mcp_context) to BeforeTool and AfterTool hook events for server-specific policies and audit logging
Added hooks support for explicit stop and block execution control
Added extensions install/uninstall commands, settings info, and update notifications
Added configurable max heap setting
Added configurable stream idle timeout with default raised to 10 minutes
Added compression threshold configuration and validation
Added tool listing and discovery through the Agent API
Added context limit resolver through the Agent API
Made MCP server discovery non-blocking at startup to reduce perceived load time
Made agent skills enabled by default
Made provider activation and auth driven through agent config rather than CLI orchestration
Made profile-qualified model identity visible in status bar, response prefix, and diagnostics
Made subagent providers activate via isolated runtime with restored load-balancer failover
Made the first model response bounded and fixed load-balancer model identity
Switched to Bun-native development launcher for faster startup
Preserved whitespace-only subagent stream deltas for accurate streaming output
Fixed restored chat rendering to display correctly with --continue
Fixed Windows shell guidance for quoted paths
Fixed Windows PTY crashes including /dev/tty and terminal resize errors
Fixed CJK input preservation and Windows ripgrep results
Fixed PowerShell parameter checks to be field-specific
Fixed non-interactive Ctrl-C hard abort behavior
Fixed error-on-cancel and ESC cancellation for subagents
Fixed todo pause filtering to prevent emoji leakage
Fixed subagent wizard invocation crash
Fixed load-balancer profile handling and context accounting
Fixed cross-provider ephemeral leak in mixed load balancers
Fixed profile auth isolation for load balancers
Fixed Anthropic OAuth 401 retry loop on empty failed token
Fixed OAuth token refresh and bucket failover during compression 401 errors
Fixed context overflow preflight after model switch
Fixed stale context limit after profile load
Fixed shared-reference token fallback accounting
Fixed compressed summary masking errors from aborting turns
Fixed streaming newline preservation in assistant output
Fixed shell security parsing backtick detection bug
Fixed tool-name and async-guidance prompt token leaks
Fixed apply_patch failures for multiline replacements and directory-qualified headers
Fixed subagent armTimeout firing against defunct instances
Fixed LSP requests bounded with timeouts to prevent hangs
Fixed /copy crash on Windows by skipping /dev/tty
Fixed list_subagents tool failing with synchronous availability error
Fixed tool call cancellation hang during parallel tool calls from subagents
Fixed stale turn state clobbering newer turn's isResponding state
Fixed provider quota error preservation across retry timeouts
Fixed load-balancer failover rotation reset on exhaustion
Fixed Codex provider identity coherence during startup
Fixed neutral message boundaries after review for safer hook execution
Fixed provider-aware rate limit messages
Fixed subagent system instruction forwarding to all providers
Fixed empty content block handling in subagent pipeline
Fixed Kimi tool call ID collisions and silent scheduler drops
Fixed Kimi K2 special buffering that caused Fireworks hangs
Fixed first chunk consumption happening inside retry boundary
Fixed stream idle timeout races with shared helper
Fixed sendPromise deadlock on idle timeout abort
Fixed sendPromise hang by closing iterator on abort
Fixed compression callback failure allowing failover
Fixed provider compression threshold clamping
Fixed streamIdleTimeoutMs leaking into provider request body
Fixed empty merged content speaker attribution
Fixed Anthropic rate_limit_error handling in JSON body like a 429
Fixed 413 errors with retry and tool name guidance
Fixed maxOutputTokens setting respect during Anthropic OAuth provider switch
Fixed --version flag printing to stdout instead of debug log
Fixed event-driven model profile display
Fixed dumpcontext request and response dumps
Fixed settings search focus and typing flow
Fixed BeforeToolSelectionHook treating empty allowedFunctionNames as no restriction
Fixed geminiMessageBuffer reset on contextCleared during streaming
Fixed subagent filtering of disabled tools instead of throwing runtime error
Fixed tool display-state flag after engine-owned continuation
Fixed cross-model thinking block drops causing 400 invalid signature on model switch
Fixed compression media sanitization
Fixed OAuth dialog guarding against wrong-provider auth prompts
Fixed OAuth reauth and relogin flow bounds
Fixed Linux auth fallback paths preservation
Fixed auth cache invalidation for unbucketed profiles
Fixed PKCE length issue and stabilized OAuth redirect port

All Changes

Consolidate release guard regression coverage (Fixes #2597) (#2599) (d69bda6)
Preserve whitespace-only subagent stream deltas (Fixes #2555) (#2591) (a1dd642)
Use Bun-native development launcher (Fixes #2342) (#2594) (e3004a8)
Keep guard tests responsive during release checks (Fixes #2584) (#2593) (2b418c1)
Honor manual versions for nightly releases (Fixes #2589) (#2592) (cf296c4)
Require explicit provider selection at startup (Fixes #2481) (#2583) (0d82e78)
Harden subagent review follow-ups (Fixes #2472) (#2582) (8390626)
Add Bun/Vitest compatibility shim for bun test runner (Fixes #2475) (#2484) (91e7462)
Enforce the GenAI enclave boundary (Fixes #2352) (#2546) (2496e89)
Preserve provider quota errors across retry timeouts (Fixes #2559) (#2568) (11328eb)
Finish provider-neutral Gemini cleanup (#2573) (#2574) (30de44a)
Merge pull request #2570 from vybestack/issue2537 (2056a55)
Remove inherited Gemini names from neutral APIs (#2537) (f10865f)
Raise unconfigured subagent turn fallback to 1000 (Fixes #2541) (#2566) (c0c27e8)
Lower DEFAULT_TOKEN_LIMIT to 200K, add GLM catalog, consolidate resolution (Fixes #2527) (#2565) (62a89dc)
good memory (c7b1b78)
test: isolate all package tests from real user storage (fixes #2528) (#2545) (d7e6cb0)
Keep Codex provider identity coherent during startup (Fixes #2544) (#2561) (bf4beab)
Fix restored chat rendering one token per line with --continue (Fixes #2549) (#2558) (6a34605)
Merge pull request #2543 from vybestack/issue5228 (a521510)
Merge remote-tracking branch 'origin/main' into issue5228 (ac29484)
Keep PowerShell parameter checks field-specific (7f38bc7)
Make PowerShell schema tests wording-independent (d165445)
Fix Windows shell guidance for quoted paths (#2538) (d3eefc8)
fix: preserve CJK input and Windows ripgrep results (#2539) (0820185)
Fix Windows shell guidance for quoted paths (fdd9502)
Remove GenAI SDK coupling from leaf packages (Fixes #2351) (#2520) (646bac9)
Make MCP discovery non-fatal and stop dropping input during startup (Fixes #2516) (#2521) (4cfe82b)
Thin the CLI entry chain: runtime assembly behind createAgent, kill deep imports, shrink core import surface (Fixes #2378) (#2493) (c3ec9d3)
refactor(providers): unify OpenAI schema converter and preserve dropped JSON-schema keywords (Fixes #2508) (#2514) (b475247)
Make startup migration and profile loading resilient (Fixes #2476, Fixes #2477) (#2504) (07c96ca)
Merge pull request #2515 from vybestack/issue2509 (e1a8c29)
Merge pull request #2513 from vybestack/issue2505 (21cb698)
Normalize core memory test paths (#2509) (049f779)
Merge pull request #2440 from vybestack/issue2349 (d4d8837)
Fix cross-platform test expectations (#2509) (b3576c8)
fix: preserve valid Responses reasoning IDs (#2349) (056a7ae)
fix(providers): trim reasoning field name and make resolveReasoningValue return undefined when unusable (f7a1043)
Fix Windows test portability (#2509) (5b0a3ea)
test(providers): use fixed timestamp in streaming test chunks (5cd89e2)
fix(providers): add reasoning.fieldName support to classic openai provider (Fixes #2505) (6c68c98)
Merge remote-tracking branch 'origin/main' into issue2349 (d74b84f)
Merge remote-tracking branch 'origin/main' into issue2349 (1a2e0c1)
Merge pull request #2507 from vybestack/issue2501 (e47e14a)
Merge pull request #2503 from vybestack/issue2500 (f70c49d)
test: align retry assertions with neutral content (#2349) (38a0028)
fix: preserve neutral retry and AFC history semantics (#2349) (e000e61)
test(agents): assert exact restored turn in ChatSessionFactory #2500 fold test (5e7ff49)
fix(providers): preserve currentProfile across provider switches (Fixes #2501) (b8192a7)
fix(core): preserve carried history across rebuilds when client not yet initialized (#2500) (dc1adcf)
fix(agents): fold restored history into reused empty HistoryService (Fixes #2500) (b36150d)
fix(ui): show effective model in profile-qualified response header (Fixes #2501) (de9f5f0)
Merge remote-tracking branch...

Assets 3

Release v0.10.0-nightly.260719.c2a2768a9

github-actions released this 19 Jul 06:59

v0.10.0-nightly.260719.c2a2768a9

c2a2768

Release v0.10.0-nightly.260719.c2a2768a9

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

Bun runtime migration: The CLI now runs directly on TypeScript source via Bun, removing the esbuild bundling step for faster startup and simpler development workflows.
New model support: Added GPT-5.6 Sol, Terra, and Luna; Claude Sonnet 5; Claude Fable 5 with user-visible refusal notices; Claude Opus 4.8 as the new Anthropic default; and expanded GLM catalog with a 200K token limit.
New providers: Added LiteLLM provider and Ollama Cloud/Makora aliases.
Provider authentication changes: Removed discontinued Qwen OAuth (now API-key-only) and deprecated Gemini Code Assist OAuth login; Gemini auth now routes through API key or Vertex. Require explicit provider selection at startup to prevent accidental usage.
Configuration on XDG paths: Migrated configuration and storage to platform-standard XDG directories.
Skills system: Introduced comprehensive agent skills with /skills reload, CLI management commands, extension support, status bar integration, and autonomous activation.
Hooks system: Added tool input modification, STOP_EXECUTION/block control in model hooks, mcp_context for BeforeTool/AfterTool events, and enable-all/disable-all commands.
Policy engine: Added granular shell allowlisting with legacy deprecation and unified shell security.
Extensions: Added install/uninstall commands, update notifications, and settings info.
MCP improvements: Made MCP server discovery non-blocking at startup and non-fatal; fixed MCP discovery dropping input during startup.
Tool changes: Removed google_web_search and google_web_fetch tools. Made search_file_content respect gitignore and llxprtignore configuration.
UI improvements: Profile-qualified model identity now appears in status bar, diagnostics, and response headers. Accumulated subagent streaming text instead of replacing per-token. Added security warnings for project-level hooks. Optimized text buffer for large inputs.
Keyboard and terminal: Added OSC-52 clipboard support for Windows Terminal, fixed home/end keybinding conflict, unified shell confirmation dialogs, added Esc-Esc to clear prompt, and restored checkmark indicators.
New commands: Added /introspect slash command, /chat debug command for nightly builds, and export chat history via /bug with GitHub issue prefill.
Removed experimental UI: Deprecated and removed OpenTUI experimental UI.
Chat and streaming fixes: Fixed --continue rendering one token per line. Fixed subagent streaming text display. Fixed whitespace-only stream deltas. Fixed stream idle timeout races, deadlocks, and Kimi tool call ID collisions. Removed Kimi K2 special buffering that caused Fireworks hangs. Stopped invalid stream fallthrough retries. Fixed Anthropic stopReason-only delta preservation.
Windows support: Fixed shell guidance for quoted paths, CJK input preservation, Windows ripgrep results, PTY crash on terminal exit, storage EINVAL errors, and /dev/tty access crashes.
Subagent fixes: Fixed parallel tool call cancellation causing hangs. Fixed ESC and error-on-cancel not cancelling subagents. Fixed Ctrl-C hard abort in non-interactive mode. Fixed subagent wizard invocation crash. Raised unconfigured subagent turn fallback to 1000.
Compression fixes: Fixed masked empty-summary errors aborting turns. Fixed compression preflight guard attempting compression before bailing. Fixed extraction safety instead of throwing. Fixed hard-limit compression fallback with diagnostics. Added 0.5% context-limit fudge factor. Preserved last user prompt during middle-out compression.
Context and history: Fixed context overflow preflight after model switch. Fixed load-balancer context accounting. Fixed context-limit not applied on first profile load. Fixed stale context limit after profile load. Fixed carried history loss across rebuilds. Fixed restored history folding into reused HistoryService.
Provider fixes: Fixed OpenAI tool format overrides. Fixed provider-aware rate limit messages. Fixed reasoning IDs and field name support for classic OpenAI provider. Fixed currentProfile preservation across provider switches. Fixed OAuth eligibility gating by base URL for third-party gateways. Restored OAuth system prompt in Anthropic provider. Fixed token cache invalidation on 401/403 revocation. Enabled bucket failover during compression 401 errors. Fixed auth cache invalidation for unbucketed profiles. Preserved Linux auth fallback paths.
Load balancer fixes: Fixed exhausted failover rotation reset. Fixed 429 aggregate retry preventing fatal drops to prompt. Fixed profile auth isolation. Fixed cross-provider ephemeral leak in mixed load balancers. Fixed shared-reference token fallback accounting. Fixed active provider restoration after fromConfig(). Fixed LB model identity.
Hooks and settings: Fixed ANSI escape code leakage in skill consent prompts. Fixed deep merge of hooksConfig/hooks settings across scopes. Fixed OAuth dialog showing prompts for wrong provider. Fixed initial prompt gating on startup guards. Fixed empty allowedFunctionNames in BeforeToolSelectionHook. Fixed systemMessage threading through hook stop/block events with action-required indicator.
Tool fixes: Fixed MCP tool lookup in tool registry. Fixed disabled tools filtering instead of throwing runtime error. Fixed sanitized tool descriptions for failures. Enforced BeforeToolSelection hook tool restrictions at runtime.
Security fixes: Enforced GenAI enclave boundary. Hardened subagent review follow-ups. Hardened sibling encrypted file stores with machine-secret KDF. Preserved fail-open-on-fault contract in envelope decrypt. Added disk-read on lock timeout to prevent cross-process token replay. Hardened secure-store fallback encryption. Removed flash fallback message from quota error responses.
Other fixes: Fixed todopause not actually pausing todo reminders. Fixed prompt-interactive scheduler readiness. Fixed profile switch stalls after failed turns. Added retry for Anthropic internal server errors instead of breaking. Disabled google_web_search by default. Fixed cross-model thinking block signature errors on model switch. Surface Claude Fable 5 classifier refusals as notices. Preserved keyfile profile auth in interactive startup. Preserved provider quota errors across retry timeouts. Fixed git-commit info loading on fresh checkout. Converted media blocks to text placeholders during compression. Fixed URL-backed PDF handling. Fixed apply_patch directory-qualified header validation. Fixed home path redaction in assignment-style tokens. Fixed JSON truncation and tab handling in export.

All Changes

Consolidate release guard regression coverage (Fixes #2597) (c2a2768)
Preserve whitespace-only subagent stream deltas (Fixes #2555) (#2591) (a1dd642)
Use Bun-native development launcher (Fixes #2342) (#2594) (e3004a8)
Keep guard tests responsive during release checks (Fixes #2584) (#2593) (2b418c1)
Honor manual versions for nightly releases (Fixes #2589) (#2592) (cf296c4)
Require explicit provider selection at startup (Fixes #2481) (#2583) (0d82e78)
Harden subagent review follow-ups (Fixes #2472) (#2582) (8390626)
Add Bun/Vitest compatibility shim for bun test runner (Fixes #2475) (#2484) (91e7462)
Enforce the GenAI enclave boundary (Fixes #2352) (#2546) (2496e89)
Preserve provider quota errors across retry timeouts (Fixes #2559) (#2568) (11328eb)
Finish provider-neutral Gemini cleanup (#2573) (#2574) (30de44a)
Merge pull request #2570 from vybestack/issue2537 (2056a55)
Remove inherited Gemini names from neutral APIs (#2537) (f10865f)
Raise unconfigured subagent turn fallback to 1000 (Fixes #2541) (#2566) (c0c27e8)
Lower DEFAULT_TOKEN_LIMIT to 200K, add GLM catalog, consolidate resolution (Fixes #2527) (#2565) (62a89dc)
good memory (c7b1b78)
test: isolate all package tests from real user storage (fixes #2528) (#2545) (d7e6cb0)
Keep Codex provider identity coherent during startup (Fixes #2544) (#2561) (bf4beab)
Fix restored chat rendering one token per line with --continue (Fixes #2549) (#2558) (6a34605)
Merge pull request #2543 from vybestack/issue5228 (a521510)
Merge remote-tracking branch 'origin/main' into issue5228 (ac29484)
Keep PowerShell parameter checks field-specific (7f38bc7)
Make PowerShell schema tests wording-independent (d165445)
Fix Windows shell guidance for quoted paths (#2538) (d3eefc8)
fix: preserve CJK input and Windows ripgrep results (#2539) (0820185)
Fix Windows shell guidance for quoted paths (fdd9502)
Remove GenAI SDK coupling from leaf packages (Fixes #2351) (#2520) (646bac9)
Make MCP discovery non-fatal and stop dropping input during startup (Fixes #2516) (#2521) (4cfe82b)
Thin the CLI entry chain: runtime assembly behind createAgent, kill deep imports, shrink core import surface (Fixes #2378) (#2493) (c3ec9d3)
refactor(providers): unify OpenAI schema converter and preserve dropped JSON-schema keywords (Fixes #2508) (#2514) (b475247)
Make startup migration and profile loading resilient (Fixes #2476, Fixes #2477) (#2504) (07c96ca)
Merge pull request #2515 from vybestack/issue2509 (e1a8c29)
Merge pull request #2513 from vybestack/issue2505 (21cb698)
Normalize core memory test paths (#2509) (049f779)
Merge pull request #2440 from vybestack/issue2349 (d4d8837)
Fix cross-platform test expectations (#2509) (b3576c8)
fix: preserve valid Responses reasoning IDs (#2349) (056a7ae)
fix(providers): trim reasoning field name and make resolveReasoningValue return undefined when unusable (f7a1043)
Fix Windows test portability (#2509) (5b0a3ea...

Assets 3

Release v0.10.0-nightly.260718.d69bda66a

github-actions released this 18 Jul 02:39

v0.10.0-nightly.260718.d69bda66a

d69bda6

Release v0.10.0-nightly.260718.d69bda66a

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

Add support for GPT-5.6 Sol, Terra, and Luna models
Add Claude Sonnet 5 (claude-sonnet-5) to the model menu
Add claude-fable-5 to the Anthropic OAuth (Claude Code) model list
Set Anthropic default model to Claude Opus 4.8
Add deprecated getLatestClaude4Model alias for backward compatibility
Add LiteLLM provider support
Add configurable reasoning field name for Ollama streaming
Introduce Agent at the interactive composition root
Make MCP server discovery non-blocking at startup for faster initialization
Remove deprecated google_web_search and google_web_fetch tools
Migrate configuration to platform-standard paths
Lower DEFAULT_TOKEN_LIMIT to 200K and add GLM model catalog
Require explicit provider selection at startup to avoid ambiguous defaults
Raise unconfigured subagent turn fallback to 1000 for improved reliability
Preserve whitespace-only subagent stream deltas for correct streaming output
Preserve provider quota errors across retry timeouts for clearer error messages
Keep Codex provider identity coherent during startup
Fix restored chat rendering one token per line with --continue
Fix Windows shell guidance for quoted paths
Preserve CJK input and Windows ripgrep results
Make MCP discovery non-fatal and stop dropping input during startup
Preserve valid Responses reasoning IDs
Add reasoning.fieldName support to classic OpenAI provider
Fix load-balancer context accounting to prevent context window overruns
Fix cross-provider ephemeral leak in mixed load balancers
Preserve streaming newlines in assistant output
Detect actual image format from magic bytes for Anthropic
Fix cache double-counting on provider switch
Retry transient mid-stream connection errors instead of breaking the loop
Harden secure-store fallback encryption
Fix active session bucket resolution during 401 force-refresh
Fix subagent crashes and hangs during launch
Handle 413 errors with retry and tool name guidance
Fix /set command to remove model_param alias, keeping only modelparam
Detect API-key providers with kebab-case base-url config
Restore checkmark symbol in CLI UI indicators
Fix apply_patch failures and directory-qualified header validation
Redact home paths in assignment-style tokens for improved security
Confirm arguments key before truncating partial JSON; handle tab in export
Display .LLXPRT_SYSTEM files separately from context files in status

All Changes

Consolidate release guard regression coverage (Fixes #2597) (#2599) (d69bda6)
Preserve whitespace-only subagent stream deltas (Fixes #2555) (#2591) (a1dd642)
Use Bun-native development launcher (Fixes #2342) (#2594) (e3004a8)
Keep guard tests responsive during release checks (Fixes #2584) (#2593) (2b418c1)
Honor manual versions for nightly releases (Fixes #2589) (#2592) (cf296c4)
Require explicit provider selection at startup (Fixes #2481) (#2583) (0d82e78)
Harden subagent review follow-ups (Fixes #2472) (#2582) (8390626)
Add Bun/Vitest compatibility shim for bun test runner (Fixes #2475) (#2484) (91e7462)
Enforce the GenAI enclave boundary (Fixes #2352) (#2546) (2496e89)
Preserve provider quota errors across retry timeouts (Fixes #2559) (#2568) (11328eb)
Finish provider-neutral Gemini cleanup (#2573) (#2574) (30de44a)
Merge pull request #2570 from vybestack/issue2537 (2056a55)
Remove inherited Gemini names from neutral APIs (#2537) (f10865f)
Raise unconfigured subagent turn fallback to 1000 (Fixes #2541) (#2566) (c0c27e8)
Lower DEFAULT_TOKEN_LIMIT to 200K, add GLM catalog, consolidate resolution (Fixes #2527) (#2565) (62a89dc)
good memory (c7b1b78)
test: isolate all package tests from real user storage (fixes #2528) (#2545) (d7e6cb0)
Keep Codex provider identity coherent during startup (Fixes #2544) (#2561) (bf4beab)
Fix restored chat rendering one token per line with --continue (Fixes #2549) (#2558) (6a34605)
Merge pull request #2543 from vybestack/issue5228 (a521510)
Merge remote-tracking branch 'origin/main' into issue5228 (ac29484)
Keep PowerShell parameter checks field-specific (7f38bc7)
Make PowerShell schema tests wording-independent (d165445)
Fix Windows shell guidance for quoted paths (#2538) (d3eefc8)
fix: preserve CJK input and Windows ripgrep results (#2539) (0820185)
Fix Windows shell guidance for quoted paths (fdd9502)
Remove GenAI SDK coupling from leaf packages (Fixes #2351) (#2520) (646bac9)
Make MCP discovery non-fatal and stop dropping input during startup (Fixes #2516) (#2521) (4cfe82b)
Thin the CLI entry chain: runtime assembly behind createAgent, kill deep imports, shrink core import surface (Fixes #2378) (#2493) (c3ec9d3)
refactor(providers): unify OpenAI schema converter and preserve dropped JSON-schema keywords (Fixes #2508) (#2514) (b475247)
Make startup migration and profile loading resilient (Fixes #2476, Fixes #2477) (#2504) (07c96ca)
Merge pull request #2515 from vybestack/issue2509 (e1a8c29)
Merge pull request #2513 from vybestack/issue2505 (21cb698)
Normalize core memory test paths (#2509) (049f779)
Merge pull request #2440 from vybestack/issue2349 (d4d8837)
Fix cross-platform test expectations (#2509) (b3576c8)
fix: preserve valid Responses reasoning IDs (#2349) (056a7ae)
fix(providers): trim reasoning field name and make resolveReasoningValue return undefined when unusable (f7a1043)
Fix Windows test portability (#2509) (5b0a3ea)
test(providers): use fixed timestamp in streaming test chunks (5cd89e2)
fix(providers): add reasoning.fieldName support to classic openai provider (Fixes #2505) (6c68c98)
Merge remote-tracking branch 'origin/main' into issue2349 (d74b84f)
Merge remote-tracking branch 'origin/main' into issue2349 (1a2e0c1)
Merge pull request #2507 from vybestack/issue2501 (e47e14a)
Merge pull request #2503 from vybestack/issue2500 (f70c49d)
test: align retry assertions with neutral content (#2349) (38a0028)
fix: preserve neutral retry and AFC history semantics (#2349) (e000e61)
test(agents): assert exact restored turn in ChatSessionFactory #2500 fold test (5e7ff49)
fix(providers): preserve currentProfile across provider switches (Fixes #2501) (b8192a7)
fix(core): preserve carried history across rebuilds when client not yet initialized (#2500) (dc1adcf)
fix(agents): fold restored history into reused empty HistoryService (Fixes #2500) (b36150d)
fix(ui): show effective model in profile-qualified response header (Fixes #2501) (de9f5f0)
Merge remote-tracking branch 'origin/main' into issue2349 (27cbc7b)
fix: harden neutral message boundaries after review (#2349) (d3d63da)
Reset exhausted load-balancer failover rotations (Fixes #2492) (#2495) (0c99622)
Add GPT-5.6 Sol, Terra, and Luna support (Fixes #2483) (#2497) (63b8855)
feat(providers): support configurable reasoning field name for Ollama streaming (Fixes #2488) (#2491) (ddff241)
fix(ui): accumulate subagent streaming text instead of replacing each token (Fixes #2008) (#2486) (ec14820)
Eliminate inverted core→tools dependency and barrel over-import chain (Fixes #2417) (#2473) (a4e7175)
feat(test): add Bun-backed root test orchestration entry point (Fixes #2463) (#2470) (ae7633d)
fix(launcher): add self-exit guards to prevent orphaned shim processes on terminal/PTY death (Fixes #2468) (#2474) (4fca1a4)
Fail loudly on unavailable profile provider; never save LB runtime as a standard profile (Fixes #2479, Fixes #2478) (#2480) (2d97afd)
fix: exclude prototype keys from executor template property test (#2349) (f1f5eab)
fix: expose placeholder model through core root (#2349) (2fbd53a)
Isolate tmux harness from nested sessions (#2471) (87e133d)
merge main into issue2349 and resolve neutral conflicts (cdf3c33)
fix(agents): activate subagent providers via isolated runtime, restore LB failover, guard Null message (Fixes #2410) (#2415) (e0a0ee8)
fix: keep empty merged content as human speaker (#2349) (12eb3ab)
fix: preserve hook telemetry metadata (#2349) (4ca3fcf)
fix: harden stream telemetry and validation paths (#2349) (a9deb44)
fix: address additional OCR migration findings (#2349) (3bd1558)
fix: address CodeRabbit migration comments (#2349) (6395d50)
fix: address neutral migration review findings (#2349) (1322381)
fix(agents): convert legacy Part-shaped arrays in toPartListUnion (#2349) (ec3e3ee)
fix: remove unnecessary type assertion in boundaryRecovery test (#2349) (913de60)
fix: remove @google/genai import from issue2410 test (#2349) (32cef38)
fix: normalize additionalContext injection to ContentBlock[] (#2349) (facc294)
fix(hooks): convert hook-modified contents back to IContent via toIContents (38ea315)
fix(format): prettier fix on auto-fixed test file (894ac54)
fix(post-rebase): resolve type errors from main merge (ba7a7d5)
fix(lockfiles): remove @google/genai from agents workspace in lockfiles (cc9f7a4)
fix(core): update tests for neutral IContent/ContentBlock shapes (49b2a3e)
fix(tests): update assertions for neutral ContentBlock shapes (22157f0)
fix(cli): extract isIContentLike to reduce expression complexity (67c504f)
fix(cli): null-check req in loopTestHelpers for strict tsc (5c6d238)
fix(cli): merge duplicate imports to fix lint:ci --max-warnings 0 (4ad1022)
fix(cli): extract loopTestHelpers to fix max-lines lint (e704101)
fix(cli): accept unknown in addPartToResult for AgentRequestInput (19f9097)
fix(agents): fix formatting + extract modelInfoHelpers for max-lines (4ad131b)
fix(agents): address OCR findings +...

Assets 3

Release v0.10.0-nightly.260717.d69bda66a

github-actions released this 17 Jul 19:18

v0.10.0-nightly.260717.d69bda66a

d69bda6

Release v0.10.0-nightly.260717.d69bda66a

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

Added GPT-5.6 Sol, Terra, and Luna model support
Added Claude Sonnet 5 to the model menu
Added Claude Fable 5 classifier refusal notices for Anthropic OAuth users
Set Anthropic default model to Claude Opus 4.8
Added GLM model catalog and lowered default token limit to 200K
Added Ollama Cloud and Makora provider aliases
Added /introspect slash command
Exposed tool listing and discovery via the Agent API
Introduced Agent at the interactive composition root with unified stream handling
Extended profile-qualified model identity into response headers and status display
Added settings descriptions and remote admin settings support in the /settings dialog
Added extensions install/uninstall commands, settings info, and update notifications
Added hooks UI with context injection, visual indicators, and security warnings
Added environment variable sanitization service
Added configurable stream idle timeout with a higher default
Required explicit provider selection at startup to prevent silent misconfiguration
Removed deprecated Gemini Code Assist OAuth login; Gemini auth now routes through API key or Vertex
Disabled google_web_search by default
Removed experimental OpenTUI UI
Fixed Windows shell guidance for quoted paths
Fixed Windows storage crashes and /dev/tty handling
Fixed CJK input preservation and Windows ripgrep results
Fixed restored chat rendering when using --continue
Preserved whitespace-only subagent stream deltas
Preserved provider quota errors across retry timeouts
Made MCP server discovery non-blocking at startup and stopped dropping input
Made startup migration and profile loading resilient to corrupt or missing data
Kept Codex provider identity coherent during startup
Reset exhausted load-balancer failover rotations
Fixed load-balancer model identity and bounded first model responses
Fixed subagent wizard invocation crash and list_subagents tool
Fixed error-on-cancel and ESC cancellation for subagents
Inherited parent max turns for subagents
Fixed Anthropic OAuth 401 retry loops and rate_limit_error handling
Fixed stale context limit after profile load
Fixed #1801: sanitized ANSI escape codes in skill consent prompts
Fixed #1802: deep-merged hooksConfig/hooks settings across scopes
Fixed #1877: prevented subagent armTimeout from firing against defunct instances
Fixed #2002: restored interactive todo updates and paused system messaging during todo pauses
Fixed #2133: apply_patch validation and export handling
Fixed #1776: gated initial prompt on startup guards
Fixed #1878: guarded OAuth dialog against wrong-provider auth prompts
Fixed #1748: treated empty allowedFunctionNames as no restriction in BeforeToolSelectionHook
Filtered emojis in todo writes and made emoji filter mode readonly
Added 0.5% context-limit fudge factor before hard-limit truncation
Retried compression on provider overload and rate-limit errors

All Changes

Consolidate release guard regression coverage (Fixes #2597) (#2599) (d69bda6)
Preserve whitespace-only subagent stream deltas (Fixes #2555) (#2591) (a1dd642)
Use Bun-native development launcher (Fixes #2342) (#2594) (e3004a8)
Keep guard tests responsive during release checks (Fixes #2584) (#2593) (2b418c1)
Honor manual versions for nightly releases (Fixes #2589) (#2592) (cf296c4)
Require explicit provider selection at startup (Fixes #2481) (#2583) (0d82e78)
Harden subagent review follow-ups (Fixes #2472) (#2582) (8390626)
Add Bun/Vitest compatibility shim for bun test runner (Fixes #2475) (#2484) (91e7462)
Enforce the GenAI enclave boundary (Fixes #2352) (#2546) (2496e89)
Preserve provider quota errors across retry timeouts (Fixes #2559) (#2568) (11328eb)
Finish provider-neutral Gemini cleanup (#2573) (#2574) (30de44a)
Merge pull request #2570 from vybestack/issue2537 (2056a55)
Remove inherited Gemini names from neutral APIs (#2537) (f10865f)
Raise unconfigured subagent turn fallback to 1000 (Fixes #2541) (#2566) (c0c27e8)
Lower DEFAULT_TOKEN_LIMIT to 200K, add GLM catalog, consolidate resolution (Fixes #2527) (#2565) (62a89dc)
good memory (c7b1b78)
test: isolate all package tests from real user storage (fixes #2528) (#2545) (d7e6cb0)
Keep Codex provider identity coherent during startup (Fixes #2544) (#2561) (bf4beab)
Fix restored chat rendering one token per line with --continue (Fixes #2549) (#2558) (6a34605)
Merge pull request #2543 from vybestack/issue5228 (a521510)
Merge remote-tracking branch 'origin/main' into issue5228 (ac29484)
Keep PowerShell parameter checks field-specific (7f38bc7)
Make PowerShell schema tests wording-independent (d165445)
Fix Windows shell guidance for quoted paths (#2538) (d3eefc8)
fix: preserve CJK input and Windows ripgrep results (#2539) (0820185)
Fix Windows shell guidance for quoted paths (fdd9502)
Remove GenAI SDK coupling from leaf packages (Fixes #2351) (#2520) (646bac9)
Make MCP discovery non-fatal and stop dropping input during startup (Fixes #2516) (#2521) (4cfe82b)
Thin the CLI entry chain: runtime assembly behind createAgent, kill deep imports, shrink core import surface (Fixes #2378) (#2493) (c3ec9d3)
refactor(providers): unify OpenAI schema converter and preserve dropped JSON-schema keywords (Fixes #2508) (#2514) (b475247)
Make startup migration and profile loading resilient (Fixes #2476, Fixes #2477) (#2504) (07c96ca)
Merge pull request #2515 from vybestack/issue2509 (e1a8c29)
Merge pull request #2513 from vybestack/issue2505 (21cb698)
Normalize core memory test paths (#2509) (049f779)
Merge pull request #2440 from vybestack/issue2349 (d4d8837)
Fix cross-platform test expectations (#2509) (b3576c8)
fix: preserve valid Responses reasoning IDs (#2349) (056a7ae)
fix(providers): trim reasoning field name and make resolveReasoningValue return undefined when unusable (f7a1043)
Fix Windows test portability (#2509) (5b0a3ea)
test(providers): use fixed timestamp in streaming test chunks (5cd89e2)
fix(providers): add reasoning.fieldName support to classic openai provider (Fixes #2505) (6c68c98)
Merge remote-tracking branch 'origin/main' into issue2349 (d74b84f)
Merge remote-tracking branch 'origin/main' into issue2349 (1a2e0c1)
Merge pull request #2507 from vybestack/issue2501 (e47e14a)
Merge pull request #2503 from vybestack/issue2500 (f70c49d)
test: align retry assertions with neutral content (#2349) (38a0028)
fix: preserve neutral retry and AFC history semantics (#2349) (e000e61)
test(agents): assert exact restored turn in ChatSessionFactory #2500 fold test (5e7ff49)
fix(providers): preserve currentProfile across provider switches (Fixes #2501) (b8192a7)
fix(core): preserve carried history across rebuilds when client not yet initialized (#2500) (dc1adcf)
fix(agents): fold restored history into reused empty HistoryService (Fixes #2500) (b36150d)
fix(ui): show effective model in profile-qualified response header (Fixes #2501) (de9f5f0)
Merge remote-tracking branch 'origin/main' into issue2349 (27cbc7b)
fix: harden neutral message boundaries after review (#2349) (d3d63da)
Reset exhausted load-balancer failover rotations (Fixes #2492) (#2495) (0c99622)
Add GPT-5.6 Sol, Terra, and Luna support (Fixes #2483) (#2497) (63b8855)
feat(providers): support configurable reasoning field name for Ollama streaming (Fixes #2488) (#2491) (ddff241)
fix(ui): accumulate subagent streaming text instead of replacing each token (Fixes #2008) (#2486) (ec14820)
Eliminate inverted core→tools dependency and barrel over-import chain (Fixes #2417) (#2473) (a4e7175)
feat(test): add Bun-backed root test orchestration entry point (Fixes #2463) (#2470) (ae7633d)
fix(launcher): add self-exit guards to prevent orphaned shim processes on terminal/PTY death (Fixes #2468) (#2474) (4fca1a4)
Fail loudly on unavailable profile provider; never save LB runtime as a standard profile (Fixes #2479, Fixes #2478) (#2480) (2d97afd)
fix: exclude prototype keys from executor template property test (#2349) (f1f5eab)
fix: expose placeholder model through core root (#2349) (2fbd53a)
Isolate tmux harness from nested sessions (#2471) (87e133d)
merge main into issue2349 and resolve neutral conflicts (cdf3c33)
fix(agents): activate subagent providers via isolated runtime, restore LB failover, guard Null message (Fixes #2410) (#2415) (e0a0ee8)
fix: keep empty merged content as human speaker (#2349) (12eb3ab)
fix: preserve hook telemetry metadata (#2349) (4ca3fcf)
fix: harden stream telemetry and validation paths (#2349) (a9deb44)
fix: address additional OCR migration findings (#2349) (3bd1558)
fix: address CodeRabbit migration comments (#2349) (6395d50)
fix: address neutral migration review findings (#2349) (1322381)
fix(agents): convert legacy Part-shaped arrays in toPartListUnion (#2349) (ec3e3ee)
fix: remove unnecessary type assertion in boundaryRecovery test (#2349) (913de60)
fix: remove @google/genai import from issue2410 test (#2349) (32cef38)
fix: normalize additionalContext injection to ContentBlock[] (#2349) (facc294)
fix(hooks): convert hook-modified contents back to IContent via toIContents (38ea315)
fix(format): prettier fix on auto-fixed test file (894ac54)
fix(post-rebase): resolve type errors from main merge (ba7a7d5)
fix(lockfiles): remove @google/genai from agents workspace in lockfiles (cc9f7a4)
fix(core): update tests for neutral IContent/ContentBlock shapes (49b2a3e)
fix(tests): update assertions for neutral ContentBlock shap...

Assets 3

Uh oh!

Releases: vybestack/llxprt-code

Release list

Release v0.10.0-nightly.260724.a20e30f14

Release v0.10.0-nightly.260724.a20e30f14

Installation

What's New

All Changes

Uh oh!

Release v0.10.0-nightly.260723.a25965779

Release v0.10.0-nightly.260723.a25965779

Installation

What's New

All Changes

Uh oh!

Release v0.10.0-nightly.260723.5868dac61

Release v0.10.0-nightly.260723.5868dac61

Installation

What's New

All Changes

Uh oh!

Release v0.10.0-nightly.260722.be8f36c6e

Release v0.10.0-nightly.260722.be8f36c6e

Installation

What's New

All Changes

Uh oh!

Release v0.10.0-nightly.260721.d58ee53aa

Release v0.10.0-nightly.260721.d58ee53aa

Installation

What's New

All Changes

Uh oh!

Release v0.10.0-nightly.260720.d69bda66a

Release v0.10.0-nightly.260720.d69bda66a

Installation

What's New

All Changes

Uh oh!

Release v0.10.0-nightly.260719.d69bda66a

Release v0.10.0-nightly.260719.d69bda66a

Installation

What's New

All Changes

Uh oh!

Release v0.10.0-nightly.260719.c2a2768a9

Release v0.10.0-nightly.260719.c2a2768a9

Installation

What's New

All Changes

Uh oh!

Release v0.10.0-nightly.260718.d69bda66a

Release v0.10.0-nightly.260718.d69bda66a

Installation

What's New

All Changes

Uh oh!

Release v0.10.0-nightly.260717.d69bda66a

Release v0.10.0-nightly.260717.d69bda66a

Installation

What's New

All Changes

Uh oh!