Maka 0.1.11

This release contains 375 merged pull requests from 24 contributors.

Highlights

• Expanded Runtime Host from a local execution service into the shared authority for multiple connected Hosts, remote project registration, live run state, and archived session lifecycle (#3097, #3145, #3079, #3074, #3151).
• Added the installable Maka CLI package and its protected staged npm release pipeline, including cross-platform artifact and Eval validation (#3169, #3173, #3185, #3188, #3192, #3197, #3200, #3201).
• Added brokered Windows AppContainer sandbox support and tightened local IPC ownership and ACL enforcement (#2961, #3179, #3182).
• Added Work Board storage foundations, Host-scoped task creation, prompt-history completion, and first-run viewport containment (#3028, #3122, #1874, #3195).
• Added native Desktop and TUI locale authorities and Qwen3.8 Max Token Plan support (#2686, #2691, #3157).

Reliability and developer experience

• Preserved live turns across refresh and projected authoritative live execution state through Runtime Host (#3189, #3079).
• Kept restored tasks safe from concurrent removal, queued busy-raced sends as steering, bounded summarizer inputs, and retired obsolete compact and compatibility paths (#3056, #3032, #3113, #3128, #2742).
• Hardened MCP rediscovery, provider failure diagnostics, rate-limit handling, session stream completion, and scheduled-task ownership (#2989, #2675, #3115, #2682, #2655).
• Added fair multi-arm Eval infrastructure and the DeepSeek Harness benchmark arm, while isolating subject metering from framework accounting (#2668, #2971, #3176).
• Strengthened Windows installer, crash-recovery, remote service restart, and release artifact coverage (#2650, #2562, #3186, #2941).

Fixed

• Disabled TUI taskbar-progress keepalives by default on native Windows and
  Windows Terminal sessions, where repeated OSC 9;4 updates can make Explorer's
  taskbar unresponsive. MAKA_TASKBAR_PROGRESS=1 restores the prior behavior,
  while MAKA_TASKBAR_PROGRESS=0 disables it explicitly.

Distribution

• Ships for Apple Silicon macOS as a signed and notarized DMG and ZIP, and for Windows x64 as an unsigned NSIS installer and ZIP, built and verified in the same release run.
• The bundled Computer Use skill ships with the app, but the Computer Use executor remains excluded from this release.

Pull requests

• #1661 refactor(mcp): migrate the legacy client path to SDK v2 — @me2seeks
• #1874 feat(ui): complete the composer draft from prompt history with Tab — @ARE404
• #1932 refactor(headless): unify held in task set hash(#1409) — @chinawch007
• #2163 perf(runtime): prune superseded tool observations — @xxhZs
• #2346 feat(inspector): say what the context is full of, not just how full — @ARE404
• #2445 fix(storage): serialize operational schema migrations — @Astro-Han
• #2511 fix(runtime): land the terminal fact past a corruption latch, type the sealed-run refusal — @Benjamin-eecs
• #2518 fix(runtime): gate Responses replay on encrypted reasoning — @me2seeks
• #2552 fix(desktop): announce bot channel status once — @me2seeks
• #2553 docs(windows): define crash durability boundary — @liugddx
• #2562 ci(windows): require crash recovery evidence — @liugddx
• #2566 fix(ui): separate session row actions from navigation — @me2seeks
• #2567 fix(cli): keep autocomplete inside short terminals — @me2seeks
• #2569 fix(runtime): preserve CJK PowerShell output on Windows — @me2seeks
• #2572 fix(ui): keep the active prompt tick visible after resize — @me2seeks
• #2575 fix(ui): settle drag selections after pointer release — @me2seeks
• #2597 feat(memory): extract on automatic compaction — @YayoiNanoka
• #2605 refactor: replace Headless with minimal Eval kernel — @Astro-Han
• #2611 fix(runtime): make the Stop/timeout lifecycle race deterministic (#2514) — @chinawch007
• #2636 fix(cli): build headless before isolated workspace tests — @MoonOld
• #2639 refactor(runtime): settle provider steps once — @Colafornia
• #2642 refactor(runtime-host): centralize workspace authority — @M4n5ter
• #2646 fix(desktop): show agent Automation crons on the scheduled-tasks page — @jackwener
• #2648 feat(scheduled-task): unify 定时任务 into a single catalog — @jackwener
• #2649 fix(cli): avoid Windows taskbar progress stalls — @me2seeks
• #2650 test(windows): exercise installer lifecycle — @liugddx
• #2651 test: wait for active session in slash command e2e — @Astro-Han
• #2655 refactor(scheduled-task): make Runtime Host the sole global authority — @jackwener
• #2656 refactor: remove obsolete fixtures and JSONL import — @jackwener
• #2657 test: remove orphaned legacy fixtures — @jackwener
• #2658 test(windows): verify installer upgrades — @liugddx
• #2665 feat(storage): add managed Secret Store and activation injection — @MicroGery
• #2668 feat(eval): add fair multi-arm execution infrastructure — @Astro-Han
• #2671 fix(runtime-host): preserve settled hosted outcomes — @Astro-Han
• #2673 docs: define contribution policy — @Astro-Han
• #2674 fix(eval): harden multi-arm evaluation infrastructure — @Astro-Han
• #2675 fix(runtime-host): preserve provider failure diagnostics — @M4n5ter
• #2676 docs(provenance): record Computer Use and bundled Skill origins — @hqhq1025
• #2680 fix(desktop): disable scroll animation for reduced motion — @Colafornia
• #2681 fix(desktop): remove stale scrollbar fallback — @MoonOld
• #2682 fix(runtime-host): preserve assistant stream completion — @M4n5ter
• #2683 docs(runtime-host): align architecture guide with current boundaries — @M4n5ter
• #2686 feat(desktop): unify native locale authority — @me2seeks
• #2688 feat(desktop): localize Plan Mode — @me2seeks
• #2689 test: remove obsolete compatibility and script tests — @jackwener
• #2690 test: remove persisted compatibility coverage — @jackwener
• #2691 feat(cli): add a TUI locale authority — @me2seeks
• #2692 test: trim redundant happy-path coverage — @jackwener
• #2693 test: remove low-value text and script suites — @jackwener
• #2694 test: remove auxiliary harness coverage — @jackwener
• #2695 chore(skills): remove bundled Skills with unresolved provenance — @likun666661
• #2696 test: reduce provider and plumbing matrices — @jackwener
• #2697 test: remove pure snapshot suites — @jackwener
• #2698 test: trim presentation projection matrices — @jackwener
• #2699 test: remove obsolete backfill and stress coverage — @jackwener
• #2700 refactor: remove legacy runtime policy migration — @jackwener
• #2701 fix(desktop): point link token at solid accent — @MoonOld
• #2703 test: remove redundant happy-path harnesses — @jackwener
• #2707 test: trim transcript and replay compatibility cases — @jackwener
• #2708 test: remove dead fixtures and diagnostic det...

Maka CLI 0.1.0-beta.1

Install with:

npm install --global maka-agent@next

Source commit: 5d9a5b0
Stage workflow run: https://github.com/maka-agent/maka-agent/actions/runs/32123297725 (attempt 1)
SHA-256: 9daa78ced074d00a32fad4d179cb0aa37d6c431bf4b53e188446c14cf29a6ae2

Apple Silicon macOS and Windows x64 release built from commit 28817be.

Before publishing, download these draft assets on another Apple Silicon Mac and on a Windows x64 machine, and complete .github/RELEASE_CHECKLIST.md.

The Windows build is unsigned: SmartScreen warns on first launch, and the download has to be checked against its .sha256 file.

Bundled Git source materials and their checksum are attached to this draft. The packaged applications also carry the GPLv2 license and written source offer.

Known limitation: Computer Use is not included in this release.

Apple Silicon macOS and Windows x64 release built from commit 808f3e4.

Before publishing, download these draft assets on another Apple Silicon Mac and on a Windows x64 machine, and complete .github/RELEASE_CHECKLIST.md.

The Windows build is unsigned: SmartScreen warns on first launch, and the download has to be checked against its .sha256 file.

Bundled Git source materials and their checksum are attached to this draft. The packaged applications also carry the GPLv2 license and written source offer.

Known limitation: Computer Use is not included in this release.

Apple Silicon macOS and Windows x64 release built from commit 8066188.

Before publishing, download these draft assets on another Apple Silicon Mac and on a Windows x64 machine, and complete .github/RELEASE_CHECKLIST.md.

The Windows build is unsigned: SmartScreen warns on first launch, and the download has to be checked against its .sha256 file.

Bundled Git source materials and their checksum are attached to this draft. The packaged applications also carry the GPLv2 license and written source offer.

Known limitation: Computer Use is not included in this release.

Apple Silicon macOS and Windows x64 release built from commit 4243e4f.

Before publishing, download these draft assets on another Apple Silicon Mac and on a Windows x64 machine, and complete .github/RELEASE_CHECKLIST.md.

The Windows build is unsigned: SmartScreen warns on first launch, and the download has to be checked against its .sha256 file.

Bundled Git source materials and their checksum are attached to this draft. The packaged applications also carry the GPLv2 license and written source offer.

Known limitation: Computer Use is not included in this release.

Apple Silicon macOS and Windows x64, built from 2e4c1aa.

Runtime Host reaches opt-in parity and reopens Computer Use on the desktop. The runtime gains restricted CodeMode and consistent file_diff output, managed workspaces bundle a hermetic Git, and the chat surfaces keep settling - tool details, composer pickers, and the prompt anchor rail all moved this release.

The Windows build is not code-signed yet. SmartScreen will warn on first launch. Verify your download against the matching .sha256 file. Known limitation: Computer Use is not included in this release.

Runtime Host

• Add the Runtime Host Bot adapter - #2192 by @M4n5ter
• Compose the Runtime Host candidate - #2194 by @M4n5ter
• Complete Runtime Host opt-in parity - #2216 by @M4n5ter
• Restore the Desktop Computer Use capability - #2233 by @M4n5ter
• Restore computer use presentation parity - #2254 by @M4n5ter
• Resolve OAuth tokens for connection effects - #2246 by @M4n5ter

Runtime and tools

• Add restricted CodeMode execution - #1648 by @Nyvo-io
• Make Edit/Write/FormatJson emit file_diff results for both renderers - #2264 by @Astro-Han
• Enforce swarm failure boundaries - #2203 by @likun666661
• Keep PTY tool results JSON-safe - #2209 by @Astro-Han
• Replay provider-facing MCP arguments - #2243 by @M4n5ter
• Preserve reasoning across provider steps - #2249 by @Astro-Han
• Make the tool-result archive capability indivisible - #2255 by @Astro-Han

Managed workspace

• Bundle and verify a hermetic Git runtime for managed workspaces - #2199 by @zhiiw
• Gate managed workspace reads behind owner-bound Runtime Host execution - #2129 by @zhiiw

Storage

• Implement the Session Bundle filesystem codec - #2013 by @MicroGery

Composer and chat surfaces

• Rebuild the tool detail panel - one surface, one type scale, real diff colours - #2227 by @Astro-Han
• Keep tool details visible during live turns - #2244 by @M4n5ter
• Keep a tool-call group collapsed until the reader opens it - #2226 by @Astro-Han
• Keep composer model/thinking pickers mounted mid-turn and unify them as ghost menus - #2230 by @Astro-Han
• Drop the branch picker and rebuild the project picker on Selector - #2217 by @Astro-Han
• Pin the prompt anchor rail to the Astryx chat scrollport - #2161 by @ARE404
• Give the prompt anchor rail its motion - #2215 by @ARE404
• Land a switched-to session at its latest turn instead of flying there - #2239 by @ARE404
• Draw the reasoning chevron from the Astryx icon registry - #2228 by @Astro-Han
• Preserve Astryx List accessible names - #2225 by @me2seeks
• Strip ATX heading markers from quote chip display - #2214 by @cat0825
• Drop the dead [data-trow] chat rules - #2231 by @Astro-Han
• Localize the Astryx chrome adopted since #1795 - #2202 by @GabrielDrapor

Settings and desktop shell

• Align the Models header and back button - #2198 by @Colafornia
• Align the Settings chat-defaults column, with a resizable assistant-tone textarea - #2256 by @GabrielDrapor
• Keep modal dialogs out of the window drag region - #2210 by @Astro-Han
• Complete the macOS quit lifecycle - #2251 by @Astro-Han
• Preserve a new task across renderer reloads - #2174 by @liugddx
• Clear the "New messages" indicator at the bottom and per conversation - #2211 by @cat0825
• Expose thinking levels for discovered Codex OAuth models - #2250 by @M4n5ter
• Replace the unhealthy opencode-free bootstrap model - #2166 by @liugddx
• Repair the five regressions from the 定时任务 rebuild - #2261 by @jackwener

Performance

• Stream OpenAI Responses incrementally - #2238 by @likun666661
• Coalesce partial stream persistence - #2157 by @xxhZs
• Mount a switched-to transcript progressively - #2191 by @Benjamin-eecs
• Seed transcript geometry before progressive fill - #2237 by @Benjamin-eecs
• Bound progressive fill idle so long transcripts settle - #2259 by @Astro-Han

Headless and Harbor

• Let the Maka arm run where its competitors run - #2177 by @Astro-Han
• Bind managed shell sessions to the in-container tool surface - #2258 by @Astro-Han
• Record the four-arm Terminal-Bench 2.1 comparison - #2208 by @Astro-Han

Tests, CI, and tooling

• Pin GitHub Actions and configure Dependabot - #2212 by @Colafornia
• Add a non-blocking Windows baseline lane - #2173 by @liugddx
• Bump the AI SDK and re-port the tool-call tracker patch - #2193 by @Astro-Han
• Make attachment drop readiness atomic in e2e - #2229 by @M4n5ter
• Extract the shared sanitizeUnicodeText into core - #2190 by @chinawch007

Contributors

Thanks to @ARE404, @Astro-Han, @Benjamin-eecs, @Colafornia, @GabrielDrapor, @M4n5ter, @MicroGery, @Nyvo-io, @cat0825, @chinawch007, @jackwener, @likun666661, @liugddx, @me2seeks, @xxhZs, and @zhiiw.

Apple Silicon macOS and Windows x64, built from 3ad479d.

Windows arrives this release. Search now runs on your own model provider, instructions you write once apply in every workspace, and updates stop interrupting your work. Astryx reached most of the remaining surfaces.

The Windows build is not code-signed yet. SmartScreen will warn on first launch. Verify your download against the matching .sha256 file. Known limitation: Computer Use is not included in this release.

Windows x64

• Package, verify, and ship a Windows x64 build from the same release run as macOS — #2182 by @Astro-Han
• Establish the Windows support baseline — #2156 by @liugddx

Web search, on your own provider

• Add provider-native web search: one routing layer over OpenAI, Anthropic, or explicit Tavily, with durable replay, citations, and privacy-mode coverage across Desktop, CLI, Runtime Host, and Headless — #2152 by @hqhq1025

Instructions, skills, and updates

• Load ~/.maka/AGENTS.md (or CLAUDE.md / GEMINI.md) before the project's own instruction files — #2183 by @cat0825
• Bundle and auto-install a Computer Use skill, never over a copy you modified — #2147 by @hqhq1025
• Make app updates background and task-aware, so restart tells you what it would interrupt — #1992 by @me2seeks

Runtime Host

• Own plan turn transitions — #2169 by @M4n5ter
• Own plan runtime state — #2082 by @M4n5ter
• Own Daily Review runtime — #2113 by @M4n5ter
• Own Deep Research state — #2099 by @M4n5ter
• Own derived Session effects — #2066 by @M4n5ter
• Add the Desktop client foundation — #2134 by @M4n5ter
• Add the Session adapter core — #2140 by @M4n5ter
• Add Session catalog IPC — #2143 by @M4n5ter
• Add Session execution IPC — #2149 by @M4n5ter
• Add Session domain IPC — #2164 by @M4n5ter
• Provide Desktop's native Browser and Computer Use capabilities back to the host — #2178 by @M4n5ter
• Expose the effective pricing projection — #2073 by @me2seeks
• Add the Desktop Pricing adapter — #2148 by @me2seeks
• Honor TMPDIR for control endpoints, and sweep only dead directories — #2160 by @Benjamin-eecs

Workspace authority

• Gate managed workspace execution behind owner-bound authority — #2106 by @zhiiw
• Seal managed workspace baseline admission with a durable Git receipt — #1872 by @zhiiw
• Make the execution boundary the sole file-path authority — #2087 by @Astro-Han
• Decide workspace path containment in one realpath space — #2059 by @Astro-Han

Settings and shell, rebuilt on Astryx

• Give each capability its own disclosure — #2100 by @jackwener
• Let a settled value stay a row until you ask to edit it — #2139 by @jackwener
• One expandable row component, not two — #2141 by @jackwener
• Drop the specs and jargon nobody can act on — #2118 by @jackwener
• Stop printing what only the copy button is for — #2116 by @jackwener
• Stop repeating one fact fourteen times — #2120 by @jackwener
• Use Astryx Link for the last bare anchors, and refresh the memory seed — #2138 by @jackwener
• Give every section the same 920px content column — #2080 by @jackwener
• Stop the row hairline bending on Astryx Item's corners — #2111 by @jackwener
• Keep project files out of settings — #2081 by @jackwener
• Empty the junk drawer and seat the workbar toggle — #2123 by @jackwener
• Let project and branch join the composer footer control row — #2168 by @jackwener
• Let Astryx clickAction own the in-flight button state — #2089 by @jackwener
• Register global shortcuts through Astryx useHotkeys — #2091 by @jackwener
• Rebuild 定时任务 on Astryx's tracker archetype, so 计划提醒 and 每日回顾 become one page — #2155 by @Astro-Han
• Rebuild the session workbar on Astryx primitives — #2158 by @Astro-Han
• Rebuild the MCP page with Astryx — #2035 by @xuruiray
• Simplify the Skills page — #1973 by @me2seeks
• Clarify the Usage page hierarchy — #2024 by @me2seeks
• Restore the native application menu and unify platform command routing — #2098 by @Astro-Han

Inspector

• Filter the trace by what a reader actually asks — #2115 by @ARE404
• Close the follow-ups left on the trace panel — #2041 by @ARE404

Performance

• Lazy-mount sidebar session trees — #2021 by @likun666661
• Eliminate the swarm ledger full-scan hot path — #2040 by @likun666661

Reliability

• Branch copied tool authority facts — #2061 by @liugddx
• Preserve stream liveness across filtered parts — #2124 by @Astro-Han
• Return proxied responses at headers instead of body EOF — #2159 by @Benjamin-eecs
• Land the terminal fact a stop claims, and stop dropping non-terminal prior runs — #2078 by @Astro-Han
• Keep realtime session creation single-flight — #2135 by @me2seeks
• Honor the default permission mode immediately — #2109 by @jackwener
• Reset incompatible operational state — #2122 by @jackwener
• Release settled bot typing listeners — #2150 by @me2seeks
• Resolve permission overlay assets in bundled dev — #2045 by @Colafornia
• Make thinking strength follow the chosen level on Kimi, StepFun, and MiniMax coding plans — #2067 by @Astro-Han
• Rehydrate an unanswered user question the surface never received — #2086 by @Astro-Han
• Give selection quote actions an opaque popover surface — #2154 by @GabrielDrapor
• Centre the empty-chat hero in the reading column — #2153 by @Astro-Han
• Align traffic lights with the sidebar icon column — #2144 by @jackwener
• Stop the collapsed rail repainting a third tone in dark mode — #2187 by @Astro-Han
• Localize the field required/optional marker app-wide — #2184 by @Astro-Han
• Put the 计划提醒 inspector two tab stops from any row — #2185 by @Astro-Han
• Seed plan reminders with distinct createdAt — #2186 by @Astro-Han
• Flag aria-label on elements whose role cannot hold a name — #2108 by @jackwener

Headless and Harbor

• Bound teardown and mark the output replay — #2151 by @Astro-Han
• Reap the scoped command's output replay — #2146 by @Astro-Han
• Keep Harbor archive ids within ref grammar — #2130 by @me2seeks
• Stop handing graded competitor arms this repo's tree — #2121 by @Astro-Han
• Stop misfiling graded competitor cells as infrastructure failures — #2090 by @Astro-Han
• Let an authoritative reward survive a missing agent self-report — #2112 by @Astro-Han
• Make the settlement window reachable at the Harbor boundary — #2107 by @Astro-Han
• Make the smoke harness reach the settlement window — #2114 by @Astro-Han
• Honour the command-timeout floor over the Harbor bridge — #2110 by @Astro-Han
• Resolve host.docker.internal for in-container arms — #2062 by @Astro-Han
• Point the AHE snapshot at the moved workspace-instructions source — #2092 by @Astro-Han
• Unify canonicalJson into a single serializer — #2005 by @chinawch007
• Add the Reasonix benchmark arm — #2079 by @Astro-Han
• Raise the harness A/B pair-concurrency cap to 16 — #2075 by @Astro-Han

Tests and tooling

• Replace fixed waits with explicit barriers in runtime tests — #2162 by @Benjamin-eecs
• Own the temp namespace in the test runners — #2068 by @Astro-Han
• Isolate the dev bootstrap environment — #2131 by @me2seeks
• Seed a real project so the sidebar rename spec stops racing — #2022 by @jackwener
• Eliminate Desktop smoke false failures — #2101 by @xuruiray
• Fix the serial workspace batch that never ran after a parallel failure — #2136 by @jackwener
• Stabilize Skill draft restoration — #2137 by @M4n5ter
• Await invocable Skill projection — #2105 by @M4n5ter
• Build an invalid UTF-8 path with Git plumbing — #2097 by @M4n5ter
• Prove the Storybook autoplay contract the smoke harness stands on — #2071 by @UncertaintyDeterminesYou4ndMe

Removed

• Remove dead storage modules — #2104 by @XmchxUp
• Remove the consumer-less Desktop IPC bridge surface — #2065 by @UncertaintyDeterminesYou4ndMe
• Remove dead shell CSS recipes and close the check-dead-css blind spots — #2070 by @UncertaintyDeterminesYou4ndMe

Contributors

Thanks to @ARE404, @Astro-Han, @Benjamin-eecs, @Colafornia, @GabrielDrapor, @M4n5ter, @UncertaintyDeterminesYou4ndMe, @XmchxUp, @cat0825, @chinawch007, @hqhq1025, @jackwener, @likun666661, @liugddx, @me2seeks, @xuruiray, and @zhiiw.

Apple Silicon macOS release. Download Maka-0.1.4-mac-arm64.dmg, verify it against the .sha256 file, then drag Maka to Applications.

Known limitation: Computer Use is not included in this release. It was rebuilt this cycle around the maka-cu executor, but maka-cu does not ship inside the DMG yet.

Computer Use, rebuilt around maka-cu

• Make maka-cu a selectable executor backend, then the executor that actually runs — #1953, #1958 by @hqhq1025
• Rewrite what the model reads: observation, action surface, refusals — #1952 by @hqhq1025
• Make the agent cursor land where it is aimed, and stop it curling — #1883 by @hqhq1025
• Mirror the driven window instead of competing for the screen — #1882 by @hqhq1025
• Let a model read back the Computer Use call it actually made — #1891 by @hqhq1025
• Give driving the computer its own conversation row, and say what it did — #1888 by @hqhq1025
• Show that the machine is being driven, stop when it locks, and keep it awake while it runs — #1889 by @hqhq1025
• Redact an observation id on the same terms as the fields beside it — #2044 by @hqhq1025
• Hold the two argument schemas against each other — #1870 by @hqhq1025
• Measure a rendering change against recorded trajectories, and name what a model got stuck on — #1884 by @hqhq1025

Storage: SQLite as the operational authority

• Make SQLite the sole operational authority — #1994 by @jackwener
• Make the project catalog part of the operational database — #2016 by @Astro-Han
• Converge concurrent fresh WAL initialization — #2003 by @zhiiw
• Record why a moved device number cannot be repaired on its own — #1868 by @hqhq1025

Runtime Host

• Own input and artifact preparation — #2031 by @M4n5ter
• Add interactive session actions — #2023 by @M4n5ter
• Add safe-boundary continuation authority — #1996 by @M4n5ter
• Host sandbox boundary interactions — #1935 by @M4n5ter
• Preserve Agent Graph history across revisions — #2039 by @M4n5ter

Agents and inspection

• Project a per-session causal trace over both ledgers — #1956 by @ARE404
• Surface the session trace as a workbar tab — #2018 by @ARE404
• Add configurable subagent model routing — #1975 by @likun666661
• Add a subagent settings page — #1999 by @likun666661, rebuilt on the settings route idiom in #2008 by @Astro-Han
• Support 32-way agent swarm concurrency — #1963 by @likun666661
• Force tool-free child summaries — #1989 by @likun666661

Editor and providers

• Add Mermaid rendering to Markdown — #1929 by @ARE404
• Switch Codex OAuth to the ChatGPT device-code flow — #1950 by @Astro-Han
• Harden Codex and add Grok PKCE — #1946 by @jackwener
• Derive thinking-strength controls from models.dev reasoning options — #2050 by @Astro-Han

Performance

• Stream the transcript incrementally instead of re-deriving it per token — #2034 by @Astro-Han
• Subscribe to session UI state at the granularity each surface reads — #1998 by @Astro-Han
• Stop idle-session event-health polling from re-rendering the whole shell — #1988 by @Astro-Han

Reliability

• Give each turn its own execution scope — #2006 by @Astro-Han
• Resolve a streamed tool call delta against the aliases that still claim it — #1995 by @Astro-Han
• Keep streamed tool calls alive when a gateway skips index 0 — #1971 by @Astro-Han
• Read AgentRun ledgers written by any version — #1986 by @Astro-Han
• Accept usage_recorded in the AgentRun event schema — #1945 by @songlairui
• Stop one undefined field killing every turn that calls a tool — #1867 by @hqhq1025
• Tell a refused tool call what to do next, not just what went wrong — #1885 by @hqhq1025
• Tell a refused Automation pause/resume which cause stopped it — #1948 by @hqhq1025
• Let a Claude the generated table has not heard of still read images — #1949 by @hqhq1025
• Treat an empty Read ref as "no ref" — #1955 by @Astro-Han
• Align workspace path handling across file tools for Windows WSL — #1696 by @woyue2
• Make macOS dev permission grants reliable — #1920 by @liugddx
• Make the macOS dev TCC grant durable, behind an opt-in — #2032 by @Astro-Han
• Say which startup step has not come back — #1869 by @hqhq1025
• Take the running turn from the run, not from session status — #1987 by @Astro-Han
• Close lifecycle races exposed by CI — #1934 by @M4n5ter
• Unify cron expression authority — #1937 by @me2seeks
• Clear the CI reds on main — #2001 by @Astro-Han

Interface

• Make Astryx the sole streaming renderer — #1960 by @Astro-Han
• Rebuild Settings on the Astryx open-group idiom — #1972 by @jackwener
• Componentize the settings page container — #1991 by @jackwener
• Move the bot-chat cluster onto the Astryx settings kit — #1993 by @jackwener
• Finish Astryx convergence for the Claude card and WeChat QR panes — #1984 by @jackwener
• Deep-review round over the settings convergence and shell seam — #1997 by @jackwener
• Rebuild the plan-reminder form as a grouped-row panel — #2002, match the reference sheet layout — #2004, deep-review the sheet — #2007, and bring list rows to the premium checklist — #2011 by @jackwener
• Extensions page joins the premium checklist — #2012 by @jackwener
• Skill category chips become Tokens — #2014, and skill empty states gain their next-step actions — #2019 by @jackwener
• Remove a misleading sample Skill action — #1966 by @jackwener
• Align composer model and voice controls — #1740 by @sunheyi6
• Restore Astryx composer focus ownership — #1968 by @Astro-Han
• Preserve sent inline references with Astryx — #1927 by @Astro-Han
• Restore line wrapping in the deep-thinking disclosure body — #1938 by @Astro-Han
• Render a running tool the same way as a settled one — #1961 by @Astro-Han

Headless and benchmarks

• Put every benchmark arm under the same tool surface and model budget — #2046 by @Astro-Han
• Register ArchiveRead when Harbor archives tool results — #2028 by @Astro-Han
• Require validated verifier grades — #1644 by @me2seeks
• Classify a benchmark trial by who ended it, not by how the exception reads — #2051 by @Astro-Han
• Benchmark three agents on DeepSeek V4 Flash — #1983 by @Astro-Han

Removed

• Remove cua-driver now that maka-cu is the Computer Use executor — #1959 by @hqhq1025
• Remove isolated runtime utilities — #2009 by @jackwener
• Retire the Cursor subscription — #2037 by @me2seeks

Testing and CI

• Parallelize heavy test lanes — #1965 and parallelize execution host integration coverage — #1964 by @M4n5ter
• Give the alignment audit its own lane — #1974 by @Astro-Han
• Keep heavy storage probes off default runs — #1947 by @jackwener
• Prune duplicate Electron journeys — #1936 by @jackwener
• Stabilize asynchronous lifecycle fixtures — #1969 by @Astro-Han
• Clean before build:test so orphan dist cannot mask deleted sources — #1944 by @Astro-Han
• Assert the attachment token's centering contract, not its pixel geometry — #2048 by @Astro-Han
• Drive the shell-run flush timer from the test instead of the wall clock — #2054 by @Astro-Han

Release engineering

• Prepare 0.1.4 — #2053 by @Astro-Han
• Notarize the final macOS DMG — #1962 by @jackwener
• Clear the high-severity advisories blocking the release audit — #2055 by @Astro-Han
• Regenerate third-party notices for the upgraded production tree — #2056 by @Astro-Han
• Upgrade ip-address past the new SSRF advisories — #2058 by @Astro-Han
• Bump the pinned Codex CLI to 0.146.0 — #2057 by @Astro-Han

Contributors

Thanks to everyone who shipped in this release: @Astro-Han, @jackwener, @hqhq1025, @M4n5ter, @likun666661, @me2seeks, @ARE404, @zhiiw, @woyue2, @sunheyi6, @songlairui, @liugddx.

Full changelog: v0.1.3...v0.1.4

Apple Silicon macOS release built from commit 76e1476.

Before publishing, download this draft asset on another Apple Silicon Mac and complete .github/RELEASE_CHECKLIST.md.

Known limitation: Computer Use is not included in this release.