CHANGELOG.md

Changelog

0.21.0 (2026-06-30)

Features

• add /set-context-window and /clear-context-window commands (#1810) (5a79697)
• agents: add per-agent step limits (#1815) (1827d84)
• claude: add Opus 4.8 model support (#1769) (9c29811)
• cli: add headless heartbeat for print mode (#1789) (eea0a1a)
• ClinePass: add gateway provider with usage support (#1818) (985984b)
• effort: Add model-level reasoning effort routing (#1780) (cb689cc)
• report: generate deterministic session task reports (#1802) (a474933)
• session: add branch command for conversation forks (#1808) (9bf6aa2)
• session: harden fork-session branching (#1801) (8023356)

Bug Fixes

• atlas-cloud: vendor to gateway catalog correction and added reasoning support (#1785) (40faf25)
• auto-continuation overly biased toward Claude-style output — breaks with non-Claude models (#1713) (a5a8ccc)
• bash: correct off-by-one in truncated-line count (#1786) (0b40000)
• bg: preserve provider env-file values during prompt detection (#1767) (6fdb1d0)
• claude: make stream watchdog deterministic (#1823) (c1a9dad)
• compaction: make snip nudges model-aware (#1816) (d0843be)
• compaction: skip microcompact when compaction is off (#1800) (320d63c)
• copilot: auto-refresh Copilot token on 401 instead of only showing re-auth hint (#1766) (22fa5b4)
• core: join multi-block message text with a real newline (#1793) (4704cbc)
• env-file: collapse escaped backslashes in quoted values (#1773) (82fd237)
• hicap: improve model catalog and effort routing (#1790) (618e901)
• moonshot: Add verified Kimi effort metadata (#1796) (13cf30a)
• ollama: cap qwen3-coder-next:cloud output at 32768 (#1814) (185ffea)
• ollama: preserve chat history with native context (#1805) (259c7ec)
• openai-shim: recover GLM/Qwen XML tool calls emitted as text (#1791) (2083d1c)
• OpenClaude native launcher after Linux install (#1798) (ab8645d)
• plugins: keep marketplace reconciliation prototype-safe (#1821) (a7945e5)
• plugins: treat prototype-shadowing marketplace names as not found (#1787) (8859c5d)
• remote-session: match ingress host by hostname, not raw substring (#1792) (1cdca1c)
• status-line: estimate usage for unsupported providers (#1803) (ff8d47d)
• xAI: Update xAI model metadata and effort handling (#1795) (6f794f4)

Performance Improvements

• build: minify the CLI bundle (whitespace + syntax, keep identifiers) (#1743) (a723540)
• integrations: load descriptor catalog lazily on first registry read (#1742) (a1a3cfc)
• repl: batch streaming text, cache normalize, coalesce config writes (#1744) (13f7401)

0.20.1 (2026-06-25)

Bug Fixes

• bg: stream session logs with bounded memory (#1762) (bd00b3b)
• bridge: match loopback hostname exactly in HTTPS credential guard (#1760) (66ddbec)
• memory: bound memory-directory scanning work (#1757) (28bbec4)
• permissions: bound the speculativeChecks cache with FIFO eviction (#1724) (adcf5e5)
• prevent recursive debounce infinite loop in team memory sync (#1726) (669ecdf)
• query: prevent spurious Windows interruption prompt by passing 'interrupt' reason (#1733) (701b68c)
• resume: tolerate malformed legacy attachment records (#1768) (db66f41)
• surface swallowed error in plan file write (#1725) (02d43b6)
• update: stop false "development build" block on npm installs with NODE_ENV=development (#1781) (d32b6f0)
• worktree: base agent isolation worktree on parent HEAD, not origin/main (#1652) (3fb718f)

0.20.0 (2026-06-24)

Features

• Add session replay timeline (#1705) (1be9b86)
• agent-routing: assign a per-agent model from the /agents menu (#1632) (5471e4c)
• api: add OpenAI-compatible credential pool failover (#1706) (dd4c4ab)
• atlas-cloud: add GLM 5.2 to vendor catalog (#1755) (c2467ee)
• auto-detect and persist project conventions to wiki (#1010) (dc6a778)
• bughunter: make /bughunter public + add /bughunter-security & /bughunter-perf with robust fallback prompts (#1621) (1aabe26)
• cache: classify prompt-cache breaks by reliability (#1693) (4cf9812)
• cli: add local background sessions (#1642) (a1b3346)
• commands: add /update command with package-manager auto-detection (#1687) (c4aa756)
• config: add explicit provider env-file loading (#1668) (5af6f95)
• config: add OPENCLAUDE_CONFIG_DIR override (#1683) (2aad6fc)
• context-collapse: opt-in between-turns context collapse (span summarization) (#1619) (d5588ea)
• fireworks: add GLM-5.2 model support (#1728) (ae66c30)
• integrations: add GLM 5.2 as an Opengateway-routed model (#1772) (adafde3)
• memory: add memory.autoWrite alias for autoMemoryEnabled (#1326) (#1396) (b8c7c3b)
• permissions: allow npm/bun/tsc --version as read-only (#1759) (bcf9421)
• query: add lifecycle identity and terminal reasons (#1682) (23bc49a)
• ux: long-turn visibility + default-on stream hang safety net (#1758) (4dee44a)
• zai: add GLM-5.2 support (#1689) (b581bd9)

Bug Fixes

• atlas-cloud: sync static catalog with live /models metadata (#1754) (a23014b)
• bg: prefer exact session names over ID prefixes (#1753) (820015f)
• bughunter-security: align log-forging exclusion with A9 criteria (1aabe26)
• claude-desktop: add native Windows support for MCP server import (#1653) (e733908)
• code-indexing: guard command detection against prototype-chain names (#1710) (059ec5e)
• context: treat Opus 4.7 as 1M-context capable in modelSupports1M (#1670) (de6b6bd)
• deps: remove deprecated uuid install path by replacing vertex-sdk with local client (#1771) (9c0d5c6)
• format: roll formatFileSize over to the next unit at the 1024 boundary (#1703) (f5041e4)
• frontmatter: expand nested brace globs in paths: correctly (#1701) (ba85aa6)
• grep: relativize content-mode paths correctly on Windows (#1704) (4aec353)
• ink: reduce high-write-ratio diagnostic noise (#1699) (cc385a6)
• lsp: throttle diagnostic storms (#1698) (8cd4633)
• mcp: support draft 2020-12 tool schemas (#1740) (ab94a50)
• messages: make projections tool-pair safe (#1695) (df986c9)
• model: preserve [1m] tag for the 'best' alias (#1671) (da551e6)
• model: preserve [1m] tag for the codex aliases (#1709) (6c7d147)
• opencode-go: sync model catalog with opencode.ai/go (#1745) (38b0e27)
• preserve provider route context metadata (#1741) (5625f42)
• provider: Add Xiaomi MiMo token plan provider (#1751) (091571f)
• provider: centralize provider secret redaction (#1665) (29aea49)
• provider: honor explicit CLAUDE_CODE_USE_OPENAI=0 on fresh startup (#1690) (6fbbf2d), closes #1245
• provider: match xAI base URL by hostname, not 'x.ai' substring (#1669) (1b33fa6)
• query: add activity-aware query guard leases (#1686) (23cfc24)
• sdk: make stub-leak detection TDZ-safe + defer to next microtask (#1287) (#1398) (650fae9)
• session-title: harden generated title handling (#1691) (beab67b)
• settings: correct stale settings path references (#1666) (544b857)
• status: show active provider route instead of legacy provider bucket (#1673) (b9a5030)
• treat 5xx HTML overload pages as retryable provider_unavailable (#1750) (aed42df)
• type safety, defensive defaults, and unbounded retry prevention (#1553) (02ee7c6)
• upgrade shell-quote 1.8.3 -> 1.8.4 (CVE-2026-9277) (#1764) (3eb57c6)
• warnings: surface perf hooks buffer guidance (#1696) (916f247)
• WSL stdin handling (#1679) (3135e73)

Performance Improvements

• eliminate response.clone() memory doubling and cache lazy tool getters (#1478) (7d130e7)

0.19.0 (2026-06-16)

Features

• add NEAR AI provider integration (#1594) (eacc7d8)
• add redacted diagnostic issue reports (#1647) (7c034c5)
• add Vietnamese i18n for slash command descriptions (#1431) (89d0531)
• agent-routing: model-only agent routes (set the verifier or any agent's model on the current provider) (#1617) (9e902db)
• compact: auto-compact prompt on /resume + determinate progress bar (#1386) (716c1d4)
• config: add compactModel option to use a separate model for compaction (#1445) (#1629) (0c45e16)
• ctx: add /ctx context window visualization and token bars to /cost (#1610) (c2cf603)
• gemini-vertex: native Gemini Vertex client and auth helpers (1/3) (#1607) (b9c65de)
• provider: add Fireworks AI as official OpenAI-compatible provider (#1590) (0b24b60)
• web: rebuild landing as Astro static site with gitlawb theme and full docs (#1606) (d08593d)

Bug Fixes

• avoid file suggestion OOM on large repos (#1074) (8f92346)
• copilot: limit sub-agent concurrency to reduce Premium Request usage (#678) (#1534) (822eff3)
• gitDiff: keep hunk content lines beginning with -- or ++ (#1646) (241d52a)
• mcp: demote successful stdio startup stderr (#1622) (9fbcd75)
• Ollama max output token override (#1659) (bac74aa)
• ollama: parse text-based tool calls as fallback (#1053) (#1076) (a3f144b)
• plugins: prevent ENOENT on Windows marketplace cache finalization (#1500) (#1531) (a7d6580)
• read: improve oversized file guidance (#1626) (5fd4a09)
• rename "Claude" to "OpenClaude" in feedback survey prompt (#1598) (d00b105)
• sandbox temp dir fallback (#1662) (c3db798)
• security: bundle real sandbox runtime in open CLI (#1641) (bd3ad89)
• session: filter Anthropic-only params from 3P provider requests (#248) (#1533) (db2d093)
• startup provider validation fallback (#1658) (b036e9f)
• suggestions: stop slash-command dropdown freezing on a throwing getter (#1657) (00ff6de)
• tool-search: enable MCP tool deferral on converted-wire providers (#1608) (614a8d9)
• typecheck: correct fetch mock type casts in test files (#1592) (794ccd4)
• typecheck: expand cachedMicrocompact stub exports (#1591) (bb19392)
• typecheck: recreate missing CLI Transport interface (#1581) (3752dfe)
• typecheck: reduce error baseline by 89 across 8 files (#1595) (9db9427)
• vision handling for OpenAI-compatible models (#1663) (7be9dce)

Performance Improvements

• cli: restore --version fast path with dynamic provider imports (#1611) (a3a3c36)

0.18.0 (2026-06-10)

Features

• add .gitattributes to enforce LF line endings (#1550) (1c27957)
• enable HISTORY_SNIP — model-callable snip tool for context management (#1407) (cdc8057)
• extend --fallback-model to interactive REPL sessions (#1346) (#1419) (14e5a41)
• github: expose all 21 Copilot models with context window metadata (#822) (#1535) (1e8c1ac)
• goal: add session-scoped /goal continuation (#1293) (102cc30)
• opengateway: surface the gateway's "auto" smart-routing model in /model (#1588) (5040c49)
• provider: add Atlas Cloud as official OpenAI-compatible provider (#1585) (5b01c2b)

Bug Fixes

• add error context to silent catches and debugger detection (#1479) (e9b211a)
• agent-routing: support API model aliases (#1546) (9a342b6)
• api: honor OpenAI-compatible retry classification (#1547) (f1013df)
• api: retry once with provider-capped max_tokens (#1235) (edabf33)
• claudemd: gate User-scope external @include behind hasClaudeMdEx… (#1223) (fde7aa0)
• context: avoid noisy metadata fallback errors (#1544) (12cbe92)
• mcp: fail-closed on token count error + respect maxChars budget (#1113) (bd8c6b1)
• mcp: pass MCP stdio server args as separate array elements to pr… (#1222) (0d4e247)
• openai-shim: guarantee reasoning_content continuity for DeepSeek… (#1228) (d67c6d7)
• openai-shim: include text for image-only user content (#1410) (7c05453)
• provider: preserve explicit startup env (#1560) (754cb61)
• replace codex input_text with standard text type in generic resp (#1449) (be05594)
• typecheck: add MCP component view types (#1564) (548bffc)
• typecheck: add missing type aliases to message and tools stubs (#1510) (ea75a51)
• typecheck: add missing xai entry to deprecation retirement date maps (#1509) (1d90960), closes #1486
• typecheck: add plugin command view types (#1565) (fc0a4b5)
• typecheck: add proper type parameters to useState(null) hooks (#1513) (0e30ee8), closes #1486
• typecheck: add wizard agent creation types (#1566) (65034db)
• typecheck: annotate diff rendering props (#1568) (e53d612)
• typecheck: declare bundled markdown and macro fields (#1562) (5c239eb)
• typecheck: declare Ink JSX intrinsics (#1571) (38b2d83)
• typecheck: declare optional native modules (#1563) (f129dd0)
• typecheck: import bun test helpers in truncate test (#1520) (8527a04)
• typecheck: narrow remote agent SDK logs (#1573) (7727a9f)
• typecheck: recreate missing FeedbackSurvey utils (#1580) (62c2c5b)
• typecheck: recreate missing Spinner types (#1579) (553342c)
• typecheck: replace dead-code literal comparisons with isAntEmployee() (#1512) (7078853)
• typecheck: restore control protocol type exports (#1497) (3a308c1)
• typecheck: tighten permission rule UI types (#1567) (6ee24f7)
• typecheck: type beta header accumulator (#1516) (a692fca)
• typecheck: type cache-busted credential tests (#1525) (9e942da)
• typecheck: type Doctor screen state (#1522) (6cb3493)
• typecheck: type FileWrite rejection state (#1574) (fba949c)
• typecheck: type Grove dialog state (#1521) (f726685)
• typecheck: type gRPC stream messages (#1572) (c2cc6ed)
• typecheck: type MCP doctor test fixtures (#1527) (38c0fe4)
• typecheck: type MCP XAA auth storage (#1570) (bf2d540)
• typecheck: type provider diagnostic tests (#1517) (ebf38f7)
• typecheck: type secure storage command output (#1524) (8c2f585)
• typecheck: type session storage test fixtures (#1526) (491985a)
• typecheck: type stats dialog state (#1569) (499c702)
• typecheck: type tool test fixtures (#1518) (ddf9282)

Performance Improvements

• attachments: skip skill listings for utility forks (#1545) (2bed184)

0.17.1 (2026-06-05)

Bug Fixes

• typecheck: type GitHub app setup flow (#1523) (80607ca)
• typecheck: type search UI state (#1529) (47eea3f)

0.17.0 (2026-06-05)

Features

• add conversation cache and session persistence (#705) (353e306)
• memory optimization to prevent OOM in multi-session scenarios (#1437) (22b1a19)
• minimax: add MiniMax M3 model with 1M context window (#1470) (e2fa248)
• nvidia-nim: dynamic model discovery via integrate.api.nvidia.com (#1099) (#1177) (890456b)
• opengateway: add MiniMax M3 and Qwen 3.7 Max to the model catalog (#1515) (8705cd3)
• opengateway: Gemini 3.1 Flash Lite GA model id (#1537) (ea09176)
• provider: auto-switch on rate limit via providerFallbackChain (#768) (#1176) (cfbce38)
• sponsors: add Atlas Cloud sponsor and sponsored tip (#1536) (73a2833)
• xiaomi: retire deprecated MiMo V2 Pro and V2 Omni (#1538) (b1a8026)

Bug Fixes

• api: tighten reasoning_content heuristic to prevent false-positi… (#1201) (1fc5116)
• BashTool: include captured output in non-zero-exit error result (#1231) (#1249) (8416faa)
• cron: enforce MAX_CRON_PROMPT_CHARS cap on durable cron prompt l… (#1224) (f7d42c2)
• plugins: use mergeHooksSettings in marketplace supplement path (#1055) (#1167) (a7fc408)
• preserve raw mode across component re-renders (issue #843) (#1198) (3bf6ccd)
• promptinput: keep bash-mode ! out of the local mirror (#1179) (#1182) (ad3e208)
• security: prevent CRLF injection, path injection, and error message leakage (#1477) (7278cad)
• show all configured Mistral models and fix model selection priority (#1360) (#1418) (7df7cad)
• show vision-specific error when provider returns 404 for image requests (#1187) (8801a4c)
• test: stop use-input test from leaking a global stdin mock (#1501) (96ddec7)
• typecheck: make session history cache variant-safe (#1494) (e7abb81)
• typecheck: narrow hook event counts (#1496) (11e46af)
• typecheck: restore AppState hook generics (#1503) (343cd1a)
• typecheck: restore proactive module import surface (#1495) (e357d59)
• typecheck: restore typed add-dir source (#1504) (2c755d3)
• use mistral-vibe-cli-latest as default model for Mistral AI (#1188) (671faf2)
• vscode: send schema-valid permission responses (#1401) (8065f8d)

0.16.1 (2026-06-01)

Bug Fixes

• ci: build before unit tests in release workflow (#1463) (d35d468)

0.16.0 (2026-05-31)

Features

• doctor: warn local-model users about large context contributors (#1238) (11f0e02)
• enable MCP_SKILLS — discover skill:// resources as invocable skills (#1408) (f111eaa)
• provider: add OpenCode Zen/Go subscription support (#1350) (5a22d60)
• set process.title to 'openclaude' (#1425) (f6d7a58)

Bug Fixes

• agents: route configured agent model overrides (#1390) (a8632b4)
• autocompact: retry circuit breaker after cooldown (#1375) (11d59ec)
• bash: show output for ! shell commands (#1265) (#1395) (ac3ae10)
• build: restore /dream slash command in bundled CLI (#1399) (132539f)
• ci: scan PR head for intent checks (#1461) (276ec6a)
• docs: update Xiaomi MiMo API URL in README. (#1424) (7cc8eda)
• fork: render forked-worker messages, drop unmirrored /fork command (#1451) (b900364)
• ink: correct stringWidth JS fallback for symbol characters (#1244) (83abfa5)
• launcher: route direct Node launch paths through launcher (#1363) (363583f)
• loader: batch markdown reads + cap file size to unblock startup (#1240) (bfb0667)
• ollama: cap deepseek v4 pro cloud output tokens (#1348) (8dd7cb0)
• onboarding: bound preflight probe + recover from connectivity failure (#1017) (#1400) (c5ca847)
• provider: allow remote Ollama without OPENAI_API_KEY (#952) (01ffbb6), closes #369
• provider: require API key input when adding OpenGateway (#1384) (7c23fb7)
• query: keep tool failure guard across unrelated successes (#1277) (cf305cc)
• release: verify npm latest tag and document @latest install (#1378) (f3d41c6)
• repl: show permission prompts while draft input is present (#1393) (70b4b07), closes #651
• sandbox: guard annotateStderrWithSandboxFailures against missing runtime method (fixes Bash on builds without sandbox-runtime) (#1452) (479b0e8)
• teammate-progress: keep cumulative token+tool counts across prompts (#475) (#1402) (5247fb8)
• test: prevent providerProfiles config mock from leaking across files (#1432) (690b3f0)
• thinking: disable thinking for unsupported Ollama models (#1376) (8513178)
• third-party provider compat — update, metrics, and refusal message (#1406) (dda5ea3)

0.15.0 (2026-05-26)

Features

• agents: set active session agent from agents menu (#1349) (7419d38)
• configure API retry backoff (#370) (#1095) (d02c10b)
• query: robust multi-lingual and structural continuation nudge (#1280) (2f8aa50)
• safety: warn at startup when 3P provider + permissive mode skip the AI classifier (#1260) (4e8fa24)

Bug Fixes

• agent: allow custom model overrides (#1337) (785d3de)
• attribution: make git attribution opt-in by default (#1335) (6bc050e)
• codex-stream: recover tool args delivered only via done events (#1262) (2d26a46), closes #1259
• codex: allow credential storage fallback (#1347) (f15d236)
• json-schema: support top-level non-object roots via wrap/unwrap (#1261) (07d9b4f), closes #1256
• model: include profile models in descriptor picker (#1361) (2c87bfe)
• route MiniMax compacting through Anthropic-compatible API (#1154) (b3dc674)
• watchers: debounce skills and settings reload bursts (#1370) (ed91673)

0.14.0 (2026-05-23)

Features

• diagnostics: show request payload size breakdown (#1237) (0aff8de)
• opengateway: require API key on /v1/* and switch to bearer auth (#1322) (0fbfc12)
• xai: add xAI/Grok OAuth provider (browser + device-code) (#1284) (326f082)

Bug Fixes

• add 5-minute timeout to QueryGuard to prevent infinite spinner (#1255) (23254c2)
• allow non-OpenAI providers to skip OPENAI_API_KEY check (#1207) (4897d59)
• bash: preserve captured stdout in error message on non-zero exit (#1236) (a44a83f)
• compact: clear native tool results after time compaction (#1278) (9b3c904)
• grpc: register built-in agents so Agent tool isn't always empty (#1296) (c366f70)
• harden XAA OAuth callback state handling (#1299) (bafc2a1)
• input: preserve split utf8 keypresses (#1241) (a9f8642)
• MiMo remove unsupported body fields and preserve reasoning content (#1253) (aab2fbc)
• monitor: close permission dialog after selection (#1225) (1aa8aab)
• query: stop repeated tool-failure loops (#1219) (f71e769)
• recovery: keep thinking blocks on resume for reasoning-echo providers (#1248) (0d3c157)
• retry: adjust max_tokens on OpenRouter 402 credit shortfall (#1263) (892c054), closes #1125
• stdin,mcp: guard rawModeEnabledCount and defer MCP connections to prevent input freeze (#603) (#1268) (90360d3)
• TaskListV2: revert overflowX hidden that hides task text labels (#1215) (0fba154)
• treat blank Read.pages as omitted (#1269) (eca9dba)
• xml: guard escapeXml/escapeXmlAttr against null and undefined (#1250) (03f8791)

0.13.0 (2026-05-17)

Features

• export: add Markdown and JSON conversation exports (#1193) (271bad4)

Bug Fixes

• bashPermissions: apply MAX_SUBCOMMANDS cap in sandbox auto-allow path (#1057) (#1166) (c53ef18)
• gemini: parse raw tool call text (#1212) (2d20109)
• spinner: prevent layout shift during thinking and orphaned task icons (#1211) (8470832)
• websearch: surface adapter failure when auto mode falls back to native (#994) (#1168) (b3b7714)

0.12.1 (2026-05-16)

Bug Fixes

• entrypoint: apply --max-old-space-size=8192 universally, not just CCR (#1191) (4d0603e)
• gemini: preserve tool calls through opengateway (#1204) (13a0901)

0.12.0 (2026-05-16)

Features

• opengateway: add Gemini 3.1 Flash Lite + GLM 5.1 FP8 to catalog (#1194) (4d04f5b)

Bug Fixes

• openai-shim: surface in-stream errors and truncation hints (#1174) (6174d75)
• Reduce stable stringify heap usage (#1104) (c433d20)

0.11.0 (2026-05-14)

Features

• add sponsored tips with frequency-gated display (#1140) (a4cbb78)
• groq: dynamic model discovery with mapModel filtering and hybrid catalog (#1143) (a65bdb4)
• implement high-performance SQLite storage layer with JSON audit log (Phase 2 Masterpiece) (#1106) (e12432e)
• nvidia-nim: add latest chat models, remove duplicate Mixtral 8x22B entry. Verified against integrate.api.nvidia.com/v1/models on 2026-05-13. Tracks #1099. (#1145) (0f6668f)
• provider: add Gitlawb Opengateway as default provider with MiMo (#1165) (5b5ba88)
• provider: add Venice official provider (#1109) (f9621ab)
• provider: add Xiaomi MiMo integration (#1152) (18483e4)

Bug Fixes

• agent: prevent mid-flight peeking and taking over of forks (#1153) (74e3947)
• bashPermissions: block command substitution in array subscript position (#1111) (4a98a4a)
• bashSecurity: tighten fc -e detection to avoid long-flag false positives (#1107) (0c88def)
• codex: normalize empty MCP object schemas (#1121) (7ea74f2)
• errors: surface re-auth hint on OAuth token expiry 401s (#1042) (#1142) (921594e)
• hide missing-module slash command stubs (#1136) (cf33f03)
• integrations: cap gpt-5.5 context window at Codex effective limit (#1118) (#1141) (cac11dc)
• replace raw abort signal timeouts (#1123) (877b4dc)
• surface actionable error when fetch fails in _doOpenAIRequest (#447) (3af0924)
• update vulnerable dependencies (#1149) (5328f57)

0.10.0 (2026-05-11)

Features

• Add startup logo palette picker (#1072) (ed7b697)
• cli: honor --model alone without requiring --provider (#854) (7cfc8d5)
• incremental and cached token counting (#795) (4b1e516)
• knowledge: introduce local Orama persistence (feature-flagged) (#1015) (5873bc6)
• make Orama the default search engine with JSON-backed (#1094) (f443669)
• websearch: add first-class Brave adapter; fix Google + Brave presets; restore Exa snippets (#1044) (402cd3d)

Bug Fixes

• agent: ensure main agent waits for subagent completion (#1032) (6af709e)
• agents: coerce non-string whenToUse to prevent crash on save (#1086) (#1087) (fc89767)
• bashSecurity: reject nested heredoc ranges in stripSafeHeredocSubstitutions (#1050) (ebc9c70)
• effort: persist xhigh and send reasoning_effort on chat_completions (#857) (feb5791)
• openai-shim: redact ?auth=, ?passwd=, ?pwd= in diagnostic URLs (#1070) (20bc6ae), closes #1069
• openai-shim: strip store for local providers (vLLM, custom) (#1048) (4830d6f)
• openai-shim: strip store when baseUrl points at Cerebras (#1040) (0adf97d)
• replace unsupported Unicode glyphs with widely available alternatives (#1088) (e1e277a)
• resolve two bugs making interactive mode unusable with plugin ecosystems (#825) (#830) (e438c89)
• validate plugin component paths (#1096) (9fed6ae)

Performance Improvements

• local: add OPENCLAUDE_LOCAL_FAST_PATH to skip cloud-only transforms (#1068) (4fad5d2), closes #1016

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)