Skip to content

CHANGELOG.md

Latest commit

 

History

History
327 lines (224 loc) · 40.8 KB

CHANGELOG.md

File metadata and controls

327 lines (224 loc) · 40.8 KB
Raw

Changelog

0.8.11 (2026-05-13)

Correções

  • install: verboo install agora exibe mensagem útil em vez de tentar baixar binários do bucket GCS da Anthropic (que não tem @verboo/code → 404 silencioso). Orienta o usuário a usar npm install -g @verboo/code@latest ou bun install -g @verboo/code@latest.
  • doctor: removidas as warnings de mismatch de installation method que sugeriam verboo install como fix — o instalador nativo não funciona na nossa distribuição.
  • update: removido o aviso "Config expects X installation / Currently running Y / Updating the Y installation" que confundia usuários ("ele fala que mudou para native installer mas não mudou"). Config agora é silenciosamente corrigido para refletir o runtime real, com log em debug.

0.8.10 (2026-05-13)

Correções

  • build: causa raiz do auto-updater quebrado para TODOS os usuários — o Bun bundler substitui process.env.NODE_ENV por "development" por default. Isso fazia getCurrentInstallationType() retornar imediatamente 'development' no bundle distribuído, bloqueando completamente o auto-updater para qualquer instalação npm (qualquer usuário, qualquer versão desde o início). Comando verboo update exibia: "You are running a development build — auto-update is unavailable". Fix: adicionado 'process.env.NODE_ENV': JSON.stringify('production') ao define dos dois bundles (CLI e SDK) em scripts/build.ts. Validado: bundle agora gera if (false) {} no lugar de if (true) { return "development" }, e verboo update segue o fluxo normal de verificação no npm.

0.8.9 (2026-05-13)

Correções

  • query: regressão crítica do 0.8.8 — a abordagem "sempre nudge a menos que completionMarkers casem" disparava 3 nudges consecutivos em saudações simples (ex.: usuário digita hello, modelo responde Olá! Como posso ajudar?, e a lógica forçava o agente a "continue with the task" 3 vezes). Restaurada a lógica de continuationSignals (modelo precisa expressar intenção explícita de seguir adiante para disparar nudge), agora com cobertura ampla em inglês E português. Verbos de ação centralizados em VERBS_EN e VERBS_PT (50+ verbos cada), e construções idiomáticas como agora vou, bora, vamos, deixa eu, next I'll, let me, moving on, etc.
  • query: completionMarkers expandido com muitos casos en+pt-br para reduzir falsos positivos quando o modelo encerra naturalmente (ex.: como posso ajudar, tudo certo, tarefa concluída, aguardo suas instruções, wrapping up, over to you).

0.8.8 (2026-05-13)

Correções

  • query: parada silenciosa do agente em sessões pt-br — continuationSignals em inglês não casava com respostas em português e o loop encerrava sem aviso. Agora o nudge é o caminho padrão sempre que há texto sem tool_use, e completionMarkers reconhece marcadores pt-br (pronto, concluído, finalizado, etc.). Cap de MAX_CONTINUATION_NUDGES = 3 continua segurando loops fechados.
  • ui: barra de contexto inconsistente — número exibido (572/262.1k) e percentual (15%) agora usam o mesmo numerador (input_tokens). Antes o percentual somava cache_creation + cache_read enquanto o número mostrava só input_tokens.
  • autoupdater: notificação de update que nunca aparecia — useUpdateNotification usava o padrão setState-durante-render com retorno derivado, e o React descartava o JSX da primeira passada. Notificação agora é armazenada em state separado e persiste no commit. Default de initialVersion trocado de MACRO.VERSION (= 99.0.0 em verboo) para MACRO.DISPLAY_VERSION.
  • autoupdater: fallback stable → latest em getLatestVersion — @verboo/code só publica tag latest, então usuários com autoUpdatesChannel: 'stable' ficavam sem updates por 404 silencioso.

0.8.0 (2026-05-06)

Novidades

  • context: particionamento de contexto e pruning por relevância — compactação mais inteligente que preserva o que importa (#849)
  • raciocínio: suporte a xhigh como nível de esforço de raciocínio e forwarding de reasoning_effort para chat completions (#857)
  • subagentes: agente principal agora aguarda o resultado do subagente antes de continuar (#1032)
  • ripgrep: cadeia de resolução aprimorada — fallback npm, modo embedded em builds compilados
  • config: suporte a múltiplos diretórios de projeto (.verboo/, .claude/) no markdownConfigLoader
  • skills: novas skills built-in karpathy-guidelines (opt-in) e statusline-setup

Correções

  • router: strip de x-anthropic-billing-header no shim OpenAI — reduz desperdício de tokens (#1019)
  • cli: substituição de createRequire por import estático em teammate.js (#1026)
  • testes: resolução de instabilidade por vazamento de estado de módulo entre testes
  • tema: remoção de wrappers de memo obsoletos nos hooks de contexto de tema

Infraestrutura

  • Node.js >=22 obrigatório — alinhamento com dependências de runtime
  • Renomeação completa openclaude → verboo em URLs, strings de display e fixtures de teste
  • Extração de streamingMetricsRef para módulo sem dependências React (evita leak no SDK bundle)

0.9.2 (2026-05-06)

Bug Fixes

  • cli: replace createRequire with static import for teammate.js (#1026) (#1033) (c873725)

0.9.1 (2026-05-05)

Bug Fixes

  • theme: remove stale memo wrappers from theme context hooks (#534) (094f04c)

0.9.0 (2026-05-05)

Features

  • context partitioning and relevance-based pruning (#849) (ca676af)
  • rework release notes around GitHub releases (#981) (d948769)
  • SDK Runtime — Query Engine, Sessions, and Build Pipeline (#984) (60c76b6)
  • support self-hosted Firecrawl via FIRECRAWL_API_URL (#949) (a133e76)

Bug Fixes

  • groq: strip unsupported store field (#983) (6d0953a)
  • mcp: allow third-party providers to approve project-scope .mcp.json servers (#696) (#937) (dc3c065)
  • shims: strip x-anthropic-billing-header block before forwarding system prompt (#1019) (40ae1e7)
  • startup: make CLAUDE logo D distinct (#986) (35f86a9)
  • tests: resolve flakiness due to module leak and env state leakage (#988) (990a5a2)
  • web-search: surface diagnostic when adapter returns 0 hits and no native fallback (#1006) (1c74675)

0.8.0 (2026-05-02)

Features

  • add Opus 4.7 as default model and fix alias/thinking bugs (#928) (4c93a9f)
  • add streaming token counter (#797) (0ca4333)
  • api: deterministic request-body serialization via stableStringify (#882) (6ea3eb6)
  • cli: improve SSH interactivity detection via SSH_TTY and SSH_CONNECTION (#946) (aae96aa)
  • context preloading and hybrid context strategy (#860) (92d297e)
  • lsp: add first-class code intelligence setup (#950) (677d29f)
  • SDK Core — Permission System, Async Context, and Engine Extensions (#951) (a46b31c)
  • SDK Foundation — Type Declarations, Errors, and Utilities (#866) (91f93ce)

Bug Fixes

  • avoid legacy Windows PasswordVault reads by default (#941) (d321c8f)
  • errors: show actual host in 404 message instead of Ollama hint (#926) (#931) (4fab8b9)
  • input: strip leading ! when entering bash mode (#947) (5943c5c), closes #662
  • oauth: skip refresh for third-party providers (#955) (208c896)
  • openai-shim: don't label transport failures as HTTP 503 (#971) (#975) (cc0dab6)
  • openai-shim: strip store when baseUrl points at Gemini (#959) (0f0fd26), closes #664
  • plugins: sanitize env before spawning git so /plugin marketplace add works (#751) (#934) (5c4fdca)
  • provider: apply Codex OAuth session switch correctly (#974) (95a817f)
  • ripgrep: use @vscode/ripgrep package as the builtin source (#911) (#932) (ee0d930)
  • typecheck: make bun run typecheck actionable on main (#473) (#938) (8106880)
  • worktree: surface git stderr in rev-parse failure message (#690) (#954) (7711dda)

0.7.0 (2026-04-26)

Features

  • add model-specific tokenizers and compression ratio detection (#799) (e92e527)
  • add OPENCLAUDE_DISABLE_TOOL_REMINDERS env var to suppress hidden tool-output reminders (#837) (28de94d), closes #809
  • add streaming optimizer and structured request logging (#703) (5b9cd21)
  • add xAI as official provider (#865) (2586a9c)
  • api: expose cache metrics in REPL + normalize across providers (#813) (9e23c2b)
  • implement Hook Chains runtime integration for self-healing agent mesh MVP (#711) (44a2c30)
  • memory: implement persistent project-level Knowledge Graph and RAG (#899) (29f7579)
  • minimax: add /usage support and fix MiniMax quota parsing (#869) (26413f6)
  • model: add GPT-5.5 support for Codex provider (#880) (038f715)
  • tools: resilient web search and fetch across all providers (#836) (531e3f1)
  • zai: add Z.AI GLM Coding Plan provider preset (#896) (a0d657e)

Bug Fixes

  • agent: provider-aware fallback for haiku/sonnet aliases (#908) (a3e728a)
  • bugs (#885) (c6c5f06)
  • make OpenAI fallback context window configurable + support external model lookup (#861) (b750e9e)
  • mcp: disable MCP_SKILLS feature flag — source not mirrored (#872) (dcbe295)
  • normalize /provider multi-model selection and semicolon parsing (#841) (c4cb98a)
  • openai-shim: echo reasoning_content on assistant tool-call messages for Moonshot (#828) (67de6bd)
  • query: restore system prompt structure and add missing config import (#907) (818689b)
  • shell: recover when CWD path was replaced by a non-directory (#871) (a4c6757)
  • startup: show --model flag override on startup screen (#898) (d45628c)
  • startup: url authoritative over model name in banner provider detect (#864) (e346b8d), closes #855
  • surface actionable error when DuckDuckGo web search is rate-limited (#834) (3c4d843)
  • test: add missing teammate exports to hookChains integration mock (#840) (23e8cfb), closes #839
  • update: show real package version and give actionable guidance (#870) (6e58b81)

0.6.0 (2026-04-22)

Features

  • add model caching and benchmarking utilities (#671) (2b15e16)
  • add thinking token extraction (#798) (268c039)
  • api: compress old tool_result content for small-context providers (#801) (a6a3de5)
  • api: improve local provider reliability with readiness and self-healing (#738) (4cb963e)
  • api: smart model routing primitive (cheap-for-simple, strong-for-hard) (#785) (e908864)
  • enable 15 additional feature flags in open build (#667) (6a62e3f)
  • native Anthropic API mode for Claude models on GitHub Copilot (#579) (fdef4a1)
  • provider: expose Atomic Chat in /provider picker with autodetect (#810) (ee19159)
  • provider: zero-config autodetection primitive (#784) (a5bfcbb)

Bug Fixes

  • api: ensure strict role sequence and filter empty assistant messages after interruption (#745 regression) (#794) (06e7684)
  • Collapse all-text arrays to string for DeepSeek compatibility (#806) (761924d)
  • model: codex/nvidia-nim/minimax now read OPENAI_MODEL env (#815) (4581208)
  • provider: saved profile ignored when stale CLAUDE_CODE_USE_* in shell (#807) (13de4e8)
  • rename .claude.json to .openclaude.json with legacy fallback (#582) (4d4fb28)
  • replace discontinued gemini-2.5-pro-preview-03-25 with stable gemini-2.5-pro (#802) (64582c1), closes #398
  • security: harden project settings trust boundary + MCP sanitization (#789) (ae3b723)
  • test: autoCompact floor assertion is flag-sensitive (#816) (c13842e)
  • ui: prevent provider manager lag by deferring sync I/O (#803) (85eab27)

0.5.2 (2026-04-20)

Bug Fixes

  • api: replace phrase-based reasoning sanitizer with tag-based filter (#779) (336ddcc)

0.5.1 (2026-04-20)

Bug Fixes

  • enforce Bash path constraints after sandbox allow (#777) (7002cb3)
  • enforce MCP OAuth callback state before errors (#775) (739b8d1)
  • require trusted approval for sandbox override (#778) (aab4890)

0.5.0 (2026-04-20)

Features

  • add OPENCLAUDE_DISABLE_STRICT_TOOLS env var to opt out of strict MCP tool schema normalization (#770) (e6e8d9a)
  • mask provider api key input (#772) (13e9f22)

Bug Fixes

  • allow provider recovery during startup (#765) (f828171)
  • api: drop orphan tool results to satisfy strict role sequence (#745) (b786b76)
  • help: prevent /help tab crash from undefined descriptions (#732) (3d1979f)
  • mcp: sync required array with properties in tool schemas (#754) (002a8f1)
  • remove cached mcpClient in diagnostic tracking to prevent stale references (#727) (2c98be7)
  • use raw context window for auto-compact percentage display (#748) (55c5f26)

0.4.0 (2026-04-17)

Features

  • add Alibaba Coding Plan (DashScope) provider support (#509) (43ac6db)
  • add NVIDIA NIM and MiniMax provider support (#552) (51191d6)
  • add ripgrep to Dockerfile for faster file searching (#688) (12dd375)
  • api: classify openai-compatible provider failures (#708) (80a00ac)
  • vscode: add full chat interface to OpenClaude extension (#608) (fbcd928)

Bug Fixes

  • focus "Done" option after completing provider manager actions (#718) (d6f5130)
  • models: prevent /models crash from non-string saved model values (#691) (6b2121d)
  • prevent crash in commands tab when description is undefined (#730) (eed77e6)
  • strip comments before scanning for missing imports (#676) (a00b792)
  • ui: show correct endpoint URL in intro screen for custom Anthropic endpoints (#735) (3424663)

0.3.0 (2026-04-14)

Features

  • activate coordinator mode in open build (#647) (99a1714)
  • activate local-only team memory in open build (#648) (24d485f)
  • activate message actions in open build (#632) (252808b)
  • add allowBypassPermissionsMode setting (#658) (31be66d)
  • add Docker image build and push to GHCR on release (#656) (658d076)
  • implement /loop command with fixed and dynamic scheduling (#621) (64298a6)
  • implement Monitor tool for streaming shell output (#649) (b818dd5)
  • local feature flag overrides via ~/.claude/feature-flags.json (#639) (0e48884)
  • open useful USER_TYPE-gated features to all users (#644) (c1beea9)

Bug Fixes

  • bump axios 1.14.0 → 1.15.0 (Dependabot #4, #5) (#670) (a07e5ef)
  • extend provider guard to protect anthropic profiles from cross-terminal override (#641) (03e0b06)
  • improve fetch diagnostics for bootstrap and session requests (#646) (df2b9f2)
  • openai-shim: preserve tool result images and local token caps (#659) (30c866d)
  • replace broken bun:bundle shim with source pre-processing (#657) (adbe391)
  • resolve 12 bugs across API, MCP, agent tools, web search, and context overflow (#674) (25ce2ca)
  • route OpenAI Codex shortcuts to correct endpoint (#566) (7c8bdcc)

0.2.3 (2026-04-12)

Bug Fixes

  • prevent infinite auto-compact loop for unknown 3P models (#635) (#636) (aeaa658)

0.2.2 (2026-04-12)

Bug Fixes

  • read/edit: make compact line prefix unambiguous for tab-indented files (#613) (08cc6f3)

0.2.1 (2026-04-12)

Bug Fixes

  • provider: add recovery guidance for missing OpenAI API key (#616) (9419e8a)

0.2.0 (2026-04-12)

Features

  • add /cache-probe diagnostic command (#580) (9ccaa7a), closes #515
  • add auto-fix service — auto-lint and test after AI file edits (#508) (c385047)
  • Add Gemini support with thought_signature fix (#404) (5012c16)
  • add headless gRPC server for external agent integration (#278) (26eef92)
  • add wiki mvp commands (#532) (c328fdf)
  • GitHub provider lifecycle and onboarding hardening (#351) (ff7d499)

Bug Fixes

  • add File polyfill for Node < 20 to prevent startup deadlock with proxy (#442) (85aa8b0)
  • add GitHub Copilot model context windows and output limits (#576) (a7f5982), closes #515
  • add LiteLLM-style aliases for GitHub Copilot context windows (#606) (2e0e14d)
  • add store:false to Chat Completions and /responses fallback (#578) (8aaa4f2)
  • address code scanning alerts (#434) (e365cb4)
  • avoid sync github credential reads in provider manager (#428) (aff2bd8)
  • convert dragged file paths to @mentions for attachment (#382) (112df59)
  • custom web search — WEB_URL_TEMPLATE not recognized, timeout too short, silent native fallback (#537) (32fbd0c)
  • defer startup checks and suppress recommendation dialogs during startup window (issue #363) (#504) (2caf2fd)
  • display selected model in startup screen instead of hardcoded sonnet 4.6 (#587) (b126e38)
  • handle missing skill parameter in SkillTool (#485) (f9ce81b)
  • include MCP tool results in microcompact to reduce token waste (#348) (52d33a8)
  • ink: restore host prop updates in React 19 reconciler (#589) (6e94dd9)
  • let saved provider profiles win on restart (#513) (cb8f8b7)
  • normalize malformed Bash tool arguments from OpenAI-compatible providers (#385) (b4bd95b)
  • preserve only originally-required properties in strict tool schemas (#471) (ccaa193)
  • preserve unicode in Windows clipboard fallback (#388) (c193497)
  • rebrand prompt identity to openclaude (#496) (598651f)
  • replace isDeepStrictEqual with navigation-aware options comparison (#507) (537c469), closes #472
  • report cache reads in streaming and correct cost calculation (#577) (f4ac709)
  • restore default context window for unknown 3p models (#494) (69ea1f1)
  • restore Grep and Glob reliability on OpenAI paths (#461) (600c01f)
  • restore Ollama auto-detect in first-run setup (#561) (68c2968)
  • scrub canonical Anthropic headers from 3P shim requests (#499) (07621a6)
  • strip Anthropic params from 3P resume paths (#479) (4975cfc)
  • suppress startup dialogs when input is buffered (#423) (8ece290)
  • tui: restore prompt rendering on startup (#498) (e30ad17)
  • update theme preview on focus change (#562) (6924718)
  • web-search: close SSRF bypasses in custom provider hostname guard (#610) (a02c441)
  • WebSearch providers + MCPTool bugs (#593) (91e4cfb)