v0.3.18

What's Changed

This release adds managed Codex CLI setup and MCP server management, introduces a built-in Codex worker template, and delivers a major Web UI navigation refresh alongside broader manager and runtime stability improvements.

Highlights:

• Added managed Codex CLI runtimes with startup integration, runtime APIs, and Computer page controls.
• Added managed MCP servers across Codex, OpenClaw, and PicoClaw, including catalog management, batch installation, and preservation of existing configurations.
• Added a built-in Codex worker template with workspace provisioning, synchronized skills, and sandbox availability checks.
• Redesigned the Web UI sidebar and page shell, with addressable sections, persistent resizing, refined settings, and improved responsive layouts.
• Stabilized Codex manager and agent setup flows, including Windows runtime handling, task titles, downloads, and activity state.

Full Changelog

• feat(web): refactor sidebar navigation and page shell (#309) (1903d5a)
• fix(web): tighten task activity polling (#310) (2c32bfc)
• feat(web): improve provider presets and editing UX (#307) (74aabb4)
• feat(web): refine sidebar and task creation flows (#311) (0c2f7a0)
• fix(web): use full language labels (#312) (c617dca)
• fix(manager): stabilize codex manager and agent setup (#314) (0ebf8a8)
• fix(web): align sidebar visuals with design (#315) (cce929c)
• feat: add Codex worker template (#313) (efcb441)
• feat(web): refine workspace sidebar navigation (#316) (c0a0dec)
• fix(agent): stabilize codex agent setup flows (#318) (199985a)
• chore: add module CODEOWNERS (96c1a3a)
• fix(web): keep sidebar nav icons embedded (#319) (78a2446)
• fix(codex): clean up Windows runtime locks (#320) (c5cde79)
• feat: add managed agent runtimes (#323) (40bb351)
• feat(mcp): add managed MCP servers for agents and runtimes (#252) (8970711)
• fix(web): improve MCP resource navigation and dialog (248dd06)
• fix: improve agent list responsiveness and connector UI (#324) (f55e4ba)
• fix(windows): Make it compatible with codex.cmd (#326) (8a1c198)
• fix(agent): preserve existing MCP servers during managed updates (#325) (f753361)
• fix(windows): Use HTTP/1.1 to download codex (#327) (2bcac27)
• fix(windows): revert to use HTTP/2 to download codex (#328) (7b7abbd)
• feat: Add more logs for downloading codex (#329) (7332c5c)

v0.3.17

What's Changed

This release switches the built-in Manager agent to the Codex CLI runtime, introduces scheduled task automation, and improves reliability across the web UI, Windows runtime paths, and OpenClaw workspaces.

Highlights:

• Switched the built-in Manager agent to run on the Codex CLI runtime, with readiness reporting when Codex is unavailable.
• Added scheduled task automation with storage, scheduler execution, run history, APIs, and workspace management UI.
• Added GitHub connector support, including backend OAuth/storage APIs and web UI flows for attaching GitHub context.
• Added OpenCSG login environment selection for production, stage, and custom sites, with environment-aware endpoints and account UI.

Full Changelog

• fix(web): refine model provider labels (#292) (2613999)
• feat(connector): add GitHub connector support (#293) (7acfb71)
• feat(web): add long message collapse (#291) (d122e9a)
• fix(agent): harden Windows runtime handling (#294) (5155308)
• feat(auth): support OpenCSG login environments (#297) (df34d7a)
• fix(web): stabilize long message collapse (#296) (fd68014)
• feat(task): add scheduled task automation (#295) (936aac6)
• fix(web): clear stale agent activity state (#298) (68e2001)
• fix(tasks): show agent names in task assignees (#300) (f25fbe4)
• fix(web): stabilize chat auto-follow scrolling (#299) (8d061d8)
• fix: support non-ascii Windows install paths (#302) (96d74a2)
• fix(web): refresh generated scheduled tasks (#304) (de07367)
• fix(web): improve scheduled task active state (#305) (25b0d47)
• fix(tasks): improve scheduled task run feedback (#308) (362d43d)
• feat(agent): run manager on Codex with connector leases (#301) (0bc9d07)
• fix(openclaw): sync agent instructions into workspace (#306) (ab53395)
• ci(release): skip image ref checks for codex templates (6fd808f)

v0.3.16

What's Changed

This release improves runtime and agent creation consistency, adds better agent activity visibility in the web UI, and tightens Windows, upgrade, and Codex integration reliability.

Highlights:

• Added an agent activity dashboard in the web UI, including agent event visibility, active working status in conversations, and chat entry points into agent details.
• Streamlined agent creation and runtime handling across templates, APIs, and profile details, with clearer runtime normalization and remembered defaults.
• Renamed the frontend Hub surface to Resources and updated related badges, icons, and navigation assets.
• Improved upgrade and runtime reliability with better upgrade diagnostics, unified runtime config compatibility, Windows-safe task state writes, and more reliable Codex runtime cleanup on delete.
• Refined Codex and skill integration by delivering tool activity via metadata, improving community skill installation behavior, and allowing OpenCSG providers for agent models.

Full Changelog

• fix(runtime): unify worker runtime config compatibility (#272) (9d89cf9)
• fix(upgrade): improve failure diagnostics (#273) (71d2b1b)
• fix(agent-profile): remove image field from profile view (#274) (d657ae1)
• fix(web): improve skill upload dialog dark mode contrast (#275) (833782b)
• feat(web): rename hub to resources and update badges and icons (#270) (a23e96d)
• fix(web): update hub icon (#276) (48721b2)
• fix(web): rename resources navigation icon asset (#277) (8b75e1c)
• refactor(web): rename Hub to Resources in the frontend (9e983a7)
• feat(web): add agent activity dashboard (#278) (7e4fe29)
• fix(taskcore): use standard JSON state writes (#280) (a0e5673)
• feat(agent): streamline runtime handling and agent creation (#281) (9cdd331)
• fix(web): align agent delete dialog dark mode (#282) (94f1ed4)
• fix(codex): retry runtime cleanup on delete (#283) (604e293)
• build(windows): add build scripts for windows (#284) (77ce522)
• feat(web): reorganize agent detail tabs (#285) (7d6a6da)
• fix(web): keep agent basics editable in template mode (#286) (f159865)
• fix(web): refine agent details tab layout (4c9cfe3)
• fix(web): stabilize template agent creation flows (#287) (2d00038)
• fix(openclaw): use compact gateway activity verbosity (#288) (672e3b5)
• fix(codex): deliver tool activity and final replies via metadata (4f58d3b)
• fix(skill): refine community skill installation (#289) (0cd96b2)
• fix(codex): allow OpenCSG providers for agent models (9d46778)

v0.3.15

What's Changed

This release expands team and task workflows, adds official remote skill installs, improves Windows OpenClaw compatibility, and fixes several web UI stability issues.

Highlights:

• Added broader agent and team task workflows across shared task APIs, CLI, HTTP flows, realtime updates, and persistent task storage.
• Improved team execution workflows and task status presentation across task APIs, CLI, and the web UI.
• Added support for installing official remote skills from the Hub, with safer archive handling, pagination, and invalid-skill filtering.
• Fixed Windows OpenClaw workspace isolation and approval-file mounting so Codex/OpenClaw flows behave correctly on Windows.
• Fixed several web UI issues, including model provider scrolling, contextual error routing, and env editor draft crashes.

Full Changelog

• fix(web): keep model provider page scrollable (#259) (e521600)
• feat(team): improve agent team workflows (#260) (ffdf821)
• feat(skill): support official remote skill installs (#261) (0c8a713)
• fix(agent): install default system skills by agent id (#262) (04b3865)
• fix(openclaw): isolate workspace mounts for Windows (#263) (a019494)
• fix: harden LLM bridge and agent cleanup flows (#264) (a9b9b9b)
• fix(openclaw): stop mounting approvals on windows (#265) (4164af9)
• chore(web): polish hub and model provider UI (#268) (1a0ed02)
• fix(web): route operation errors to contextual views (#269) (106aa30)
• feat(task): support agent and team task workflows (#267) (8ebe45c)
• fix(web): prevent env editor draft crashes (#271) (690a7c0)

v0.3.14

What's Changed

This release strengthens template and skill distribution, modernizes local storage and migration safety, and smooths several web, runtime, and Windows-specific workflows.

Highlights:

• Added embedded system skills and centralized shared skill creation so default skills are provisioned consistently and protected from accidental replacement.
• Switched official template loading to CSGHub with lazy-loaded workspace contents and file previews, while keeping the template listing resilient to invalid remote entries.
• Migrated local state to typed IDs with updated API, CLI, and web surfaces, while adding safer migration backups and symlink preservation.
• Improved agent and template flows across the UI by restoring the add-skill button, fixing profile draft synchronization, filtering manager templates during agent creation, and showing clearer template details.
• Fixed Codex usability on Windows by correcting command path handling, and hardened runtime behavior with legacy sandbox recreation, runtime-specific worker defaults, failed gateway cleanup, and more reliable Feishu reconnect recovery.

Full Changelog

• fix(web): restore the add-skill button in agent profile (b60f3c8)
• feat(web): add GitHub feedback link (#241) (aa19766)
• refactor(template): rename hub package and CLI to template (#243) (d7d0f57)
• fix(config): preserve Windows hub registry paths (#244) (47579db)
• feat(skill): add embedded system skills (#235) (9e9c6b7)
• feat(hub): lazy load official template workspaces (#246) (3d64b34)
• refactor(skill): centralize skill-creator as system skill (#247) (14e284e)
• refactor(web): align design token naming with figma (#236) (b764bb0)
• fix(web): sync agent detail draft on profile updates (#248) (3ef7258)
• fix(config): migrate legacy hub settings (#250) (0ee5585)
• fix(web): derive the agent profile selector from agent_profile (a6319d7)
• fix(codex): improve command path handling for Windows (#251) (4fac538)
• feat(store): migrate local state to typed IDs (#245) (c3d7e82)
• fix(agent): recreate legacy-named sandboxes on startup (#253) (0413cfc)
• fix(localstore): preserve symlinks in migration backups (#254) (e272f79)
• fix(config): back up store before migration rewrites (#255) (2727c19)
• fix(template): include human in manager-created rooms (#249) (53cf054)
• feat(template): rename builtin worker templates (dd85371)
• fix(web): hide manager templates in agent creation (554646c)
• fix(config): back up store before migration rewrites (#257) (79e4fec)
• fix: harden runtime defaults and Feishu reconnects (#258) (08f4e26)

v0.3.13

What's Changed

This release expands skill and model management across the product while improving authentication, Feishu integration, and runtime upgrade behavior.

Highlights:

• Added end-to-end SkillHub management, including global skill browsing, file preview, upload, deletion, and agent-scoped skill assignment.
• Introduced account login and centralized model provider management, including OpenCSG support, provider CRUD, connectivity checks, and improved same-tab sign-in behavior.
• Improved Feishu integration with clearer bot invite failure reporting, reconnect fixes, exposed bot IDs for API invites, and better tool activity formatting.
• Added Windows launcher auto-upgrade support and mounted csgclaw-cli into sandbox runtimes from the host for easier runtime refresh and shared image usage.
• Polished the web UI with agent profile section navigation, more stable skill preview rendering, and dialog/layout fixes.

Full Changelog

• fix(channelbridge): format Feishu tool activity messages (#221) (5f9f8ea)
• feat(skillhub): add global skill browsing to hub (65cb11d)
• feat(skillhub): add skill upload and removal workflow (7a88b35)
• fix(web): stabilize hub skill preview layout (b16e8a7)
• feat(agent): add agent-scoped skill management (93172df)
• fix(web): default hub skill path to SKILL.md on selection (1b1fe5a)
• fix(codex): keep managed AGENTS.md refresh under CODEX_HOME (4c7a4b0)
• fix(feishu): surface bot invite failures and setup guidance (#222) (31e38ac)
• fix(feishu): Fixed the Feishu reconnect error (#223) (2ab082b)
• fix(web): improve spacing in the hub skill delete dialog (08c39d1)
• feat(upgrade): add Windows auto-upgrade support for launcher installs (#225) (521c6fd)
• fix(openclaw): expose Feishu bot IDs for API invites (#226) (cbbf310)
• feat(runtime): mount sandbox csgclaw-cli from host (#228) (8614908)
• feat(auth): add login flow (#230) (583c1c6)
• feat(model): add centralized provider catalog (#233) (6578798)
• feat(web): add agent profile section navigation (#227) (3fb5766)
• fix(openclaw): route runtime LLM through CSGClaw bridge (#238) (b61994f)
• refactor(hub): consolidate templates and local skill helpers (#237) (834a0e0)
• feat(model): add built-in OpenCSG provider (#239) (2d9fc9c)
• fix(auth): restore same-tab OpenCSG login (#242) (2e94b25)

v0.3.12

What's Changed

This release expands Codex and Feishu integration, improves workspace management flows in the web UI, and hardens upgrade and runtime behavior across local deployments.

Highlights:

• Added Codex provider support in OpenClaw, including image input handling and richer streaming usage metadata.
• Expanded Feishu binding and channel management with a dedicated agent binding workflow, registration/polling/finalization support, idempotent admin human binding, bound-channel visibility in the API/UI, and runtime refresh or reconnect handling when bindings change.
• Introduced a Feishu local Codex bridge with threaded replies plus more resilient reaction and final-reply handling.
• Improved the web experience with a floating manager chat, better room member management, a cleaner agent creation form, and more consistent profile behavior.
• Hardened local runtime and release behavior by moving managed AGENTS.md into agent CODEX_HOME, and requiring managed official installs for upgrades.

Full Changelog

• fix(agent): use effective CSGHub Lite defaults (#208) (bf46df2)
• feat(openclaw): support Codex provider and image input (#211) (f7325c9)
• fix(web): keep agent overview after delete (#212) (a2e0b48)
• feat(web): add floating manager chat (#214) (4dd4395)
• feat(channelbridge): add Feishu local Codex bridge (#209) (0a07e7d)
• fix(upgrade): require managed official installs for release updates (#215) (10efe5b)
• fix(feishu): keep codex bridge reactions and final replies best-effort (#217) (f710b71)
• fix(boxlite): serialize cli commands per home (#218) (fb2fb56)
• feat(feishu): add agent binding workflow (#216) (0369838)
• fix(build): handle empty docker embed manifest args under nounset (27ce68b)
• feat(web): improve room member management and tooltip behavior (#213) (eb00a2b)
• feat(web): refine agent creation form experience (#219) (6de266a)
• fix(codex): move managed AGENTS.md to agent CODEX_HOME (f5fb1fc)
• feat(feishu): add human channel management (#220) (adf9598)
• fix(web): align human profile layout with agent profile (1180b11)

v0.3.11

What's Changed

This release expands agent profile management across the web UI and Codex runtime, with stronger avatar and human profile flows, schema-driven runtime configuration, and cleaner participant-based Feishu credential binding.

Highlights:

• Added end-to-end avatar management for agents and humans, including upload/cropping, built-in avatar assignment, and safer save behavior for partial profile edits.
• Introduced human profiles and DM profile actions in the web UI, with avatar updates synced back to IM user records.
• Added agent instructions support for Codex agents and generated managed AGENTS.md content in local workspaces.
• Expanded Codex runtime configuration with schema-driven runtime options, local workspace directory support, host directory picking, and narrower restart behavior.
• Bound Feishu credentials through participants across the CLI, API, provider wiring, and related docs.

Full Changelog

• feat(web): add agent avatar management (#201) (1de03ed)
• fix(web): prevent avatar saves from submitting incomplete profiles (#202) (c1e22bd)
• feat(web): add human profiles and DM profile action (#203) (73fe511)
• feat(agent): support instructions and managed AGENTS.md for codex agents (1ffa681)
• fix(codex): inject env and decode app-server stdout as JSON stream (c26486f)
• feat(web): assign built-in avatars on participant creation (#205) (2cd91c7)
• feat(web): streamline agent profile and creation forms (1e10d48)
• feat(feishu): bind credentials through participants (#204) (135ebcd)
• chore(templates): update embedded manager image tags (929f75b)
• feat(web): show agent skills in profile (7754aa1)
• feat(codex): add schema-driven runtime options for local workspace dir (674f93b)
• refactor(codex): move runtime config handling into the runtime controller (da1792f)
• fix(agent): use field masks for agent patch updates (de185b6)
• feat(api): add host directory picker for agent workspace fields (50da7a5)
• fix(codex): only restart on workspace dir changes (1406561)
• fix(web): inline runtime options into agent profile basics (a684367)

v0.3.10

What's Changed

This release strengthens the platform’s core agent and team model while improving runtime resilience, configuration workflows, and day-to-day web UI behavior.

Highlights:

• Replaced legacy bot identities with a participant-based model across APIs, CLI, runtime wiring, web UI, configs, and docs, with follow-up fixes for legacy participant IDs and admin state migration.
• Added a settings save-and-restart flow so common config.toml fields can be edited in the UI, validated before save, and applied with restart guidance.
• Improved team and task workflows by refactoring participant mapping, moving task execution logic into internal/team, and exposing friendly agent display names in task views.
• Upgraded Codex and agent runtime behavior by replacing ACP session wiring with the app-server runtime, unifying runtime layout handling, and restoring persisted Codex sessions after server restart.
• Polished operator and user experience with better slash command handling, thread/tool-call rendering fixes, manager setup and upgrade refresh improvements, and safer cleanup/profile save behavior.

Full Changelog

• fix(im): preserve mention context for bot commands (#178) (d8be5e0)
• fix(openclaw): fix plugin permissions and kanban scrolling (#179) (717918d)
• refactor(participant): replace bot identities with participants (#155) (a4cfd35)
• Chore(image): Upgrade PicoClaw image to picoclaw:2026.6.8 (#183) (a748546)
• feat(config): add settings save and restart flow (#182) (caa3994)
• chore(web): tighten frontend config and typings (#181) (b81e060)
• fix(participant): repair legacy agent participant IDs (#184) (da0d4cc)
• ci: allow normal markdown body (#185) (fb7450e)
• fix(participant): migrate admin state and restore upgrade action (#186) (b59d8dc)
• chore: update OpenClaw runtime image tag (#187) (f2ef846)
• docs(license): add logo restriction to LICENSE (fbf79e2)
• feat(web): optimize slash command picker (#190) (83b0666)
• refactor(templates): derive PicoClaw image tags from agent.toml (#188) (8e061bd)
• fix(threads): attach tool calls to the response root (#189) (15ef6b4)
• fix(api): delete unreferenced CSGClaw users on agent cleanup (#191) (cc43ed4)
• Fix(UI): Refresh agent detail after upgrade action (#192) (f216911)
• fix(agent): apply gateway profile on save and fix GLM token limits (#193) (1b08d4b)
• fix(web): show manager setup notice and refresh workspace (#194) (fe514c4)
• feat(ci): add shared Docker embed image pipeline (#195) (9862ca6)
• feat(team): refactor participant mapping and task workflows (#196) (977420c)
• refactor(codex): replace ACP session wiring with app server runtime (0c704e9)
• feat(agent): use embedded templates for image upgrades (#198) (57a5091)
• refactor(agent): unify runtime layout for workspace, skills, and host logs (43c6e73)
• fix(api): separate agent skills browsing from workspace paths (99a4ba5)
• feat(team): expose agent display names in tasks (#200) (44f86e0)
• build(templates): gate image bumps and embed builds (#199) (0db883d)
• fix(codex): restore app-server sessions after server restart (c5c6735)

v0.3.9

What's Changed

This release improves agent and team workflows, expands UI capabilities, and hardens Codex runtime integration and local workspace behavior.

Highlights:

• Added agent avatars end to end, including models, API/service flows, bot creation, and web UI display.
• Improved team task execution and board interactions with clearer planning, task-room projection, parent-child task details, and start controls.
• Added cross-runtime /new conversation reset support and IM room message clearing.
• Improved the web UI with safer agent detail actions, better chat overlay layering, and preserved composer caret behavior during slash tokenization.
• Hardened Codex runtime integration and local execution by stabilizing the provider bridge, isolating runtime shells from host startup environment, and preserving workspace ownership for bind-mounted workspaces.

Full Changelog

• fix(codex): stabilize runtime provider bridge (#171) (5924b4f)
• feat(web): improve agent detail actions (#172) (d81e416)
• feat(agent): add avatar support (#140) (f66d011)
• refactor(ci): embed PicoClaw template dist (#173) (e4f9c9b)
• refactor(agent): resolve workspace roots through registered runtimes (a669992)
• feat(im): add room message clearing (#168) (3a4179c)
• docs: update AGENTS.md (#123) (0a7ca0d)
• refactor(templates): centralize builtin template registry (161f35c)
• fix: preserve workspace ownership and refactor Makefile (#175) (ebc8bc8)
• fix(web): improve chat overlay layering (#176) (84e08d2)
• feat(team): improve task completion and board interactions (#177) (dee5682)
• feat(agent): support /new conversation reset across runtimes (#169) (cb9ba46)
• fix(codex): isolate runtime shells from host startup env (ea8c229)
• fix(web): preserve composer caret after slash tokenization (909d144)