OpenClaw had its Microsoft week
June 5, 2026 · stable releases + Windows + Scout
The codebase kept hardening, but the bigger story is distribution: Windows Companion, Microsoft Scout, and three public videos that pushed OpenClaw from project chatter into platform gravity.
- 2026.5.28 stable
- 2026.6.1 stable
- Next-train content stays out until it ships
- Windows Companion made OpenClaw native on Windows
- Microsoft Scout made the enterprise wedge explicit
- Skill governance, provider bounds, channel reliability, and proof lanes kept maturing
OpenClaw Weekly Dev Experience · slide 1 / 17
June 5, 2026
High velocity, less hand-waving
Sampled window
The raw activity is still noisy. The useful read is where the work clustered: release proof, channel reliability, skill governance, provider bounds, and mobile state.
- 3,000+ commits sampled
- 168 authors
- 185 release bullets
- 206 issue refs
- Peter Steinberger 1,559
- Vincent Koc 759
- Shakker 195
- Ayaan Zaidi 41
- Dallin Romney 20
- 1,192 fixes
- 907 docs
- 261 tests
- 261 refactors
- 116 chores · 99 perf · 79 features
OpenClaw Weekly Dev Experience · slide 2 / 17
June 5, 2026
Two stable releases only.
5.28 + 6.1 stable only
This deck keeps the public story to the two stable releases and saves next-train content for the week it lands.
- Codex runtime recovery
- Safer channel identity
- iOS Pro UI refresh
- Stricter browser and automation inputs
- Provider/media expansion plus faster-fail CLI and auth
- Cleaner interrupted tool recovery
- Steadier mobile and channel delivery
- Bounded provider and plugin requests
- Skill Workshop and Workboard
- Control UI startup resilience
OpenClaw Weekly Dev Experience · slide 3 / 17
June 5, 2026
The biggest release may not be in the changelog
Windows Companion + Scout
The changelog says the platform got safer. Microsoft made the market story louder: OpenClaw now has a Windows desktop path and a first-party enterprise agent built on top of it.
- Windows Companion moves OpenClaw onto the default enterprise desktop
- Scout turns OpenClaw into a Microsoft 365 work-assistant story
- This is distribution, not a feature drop
- Local desktop control
- Microsoft 365 context
- Enterprise identity and policy
- Open-source foundation still visible underneath
OpenClaw Weekly Dev Experience · slide 4 / 17
June 5, 2026
OpenClaw Windows Companion landed
Native Windows path
The Windows launch is the practical adoption story: tray app, setup, chat, diagnostics, node mode, and installers for x64 and ARM64. Not glamorous. Extremely useful.
- Native WinUI Windows companion / hub
- Signed installer path for x64 and ARM64
- System tray status, local setup, chat, Command Center diagnostics
- GitHub repo: openclaw/openclaw-windows-node
- Windows can become a first-class OpenClaw node
- Canvas, screen, camera, notifications, device status, STT/TTS, and controlled system.run are in scope
- Launch video: Scott Hanselman + Samantha Song demo OpenClaw on Windows with Peter joining
OpenClaw Weekly Dev Experience · slide 5 / 17
June 5, 2026
Microsoft Scout made the enterprise bet explicit
Always-on personal agent
Scout is Microsoft putting a name on the enterprise version of the idea: an always-on agent with files, shell, browser, Microsoft 365 context, approvals, identity, and policy.
- Scout is framed as an Autopilot: always-on, autonomous, operating under user and org controls
- It connects across Teams, Outlook, OneDrive, SharePoint, calendar, chats, contacts, cloud, desktop, and web
- Microsoft says it is powered by OpenClaw open-source technology
- Governed Entra identity instead of anonymous service-account mush
- Sensitive actions can require approval
- Purview and organizational controls stay in the loop
- Frontier preview, not general availability yet
OpenClaw Weekly Dev Experience · slide 6 / 17
June 5, 2026
Three videos, one live thread
Today + next week
This week had enough video signal for its own run. Cover Windows today. Go deeper on Peter's build-system talk and the GitHub fireside next week.
- Use in today's presentation
- OpenClaw + Windows: Microsoft Build 2026
- Scott Hanselman and Samantha Song demo the Windows path; Peter Steinberger joins
- Peter Steinberger, BRK245
- Deep-dive next week
- Core thesis: build tools that close the loop for agents, then let the tools compound
- Deep-dive next week
- Peter Steinberger, Dave Morin, Omar Shaheen
- Maintainer panel: Val, Jacob Tomlinson, Sally, Brad, Vincent, Josh
OpenClaw Weekly Dev Experience · slide 7 / 17
June 5, 2026
Skill creation got a review lane
Skill Workshop
Skills now move through proposal, review, approval, quarantine, rollback metadata, and support-file approval. Good. Markdown roulette was not a governance model.
- Proposal lists, review states, and revision handoff
- Searchable previews and support-file approval
- Visible who-approved, what-changed, what-allowed trail
- Skill creation stops being drop-a-file-and-pray
- Reviewers can reverse installs and inspect support files
- Skill lifecycle becomes auditable end to end
OpenClaw Weekly Dev Experience · slide 8 / 17
June 5, 2026
Channels and mobile became the product
Delivery is not plumbing
Channel bugs are the user experience. If the final reply vanishes, nobody cares that the model was clever. This week pulled more delivery failures into durable, inspectable paths.
- Telegram, WhatsApp, iMessage, Slack, Discord, Teams, Google Chat, Meet
- Context preserved; duplicate previews reduced
- Final replies durable; late cleanup stops masking success
- iOS Pro UI refresh
- Hosted push relay defaults
- Android helpers and mobile session diagnostics
- Mobile is now a command surface, not a notification afterthought
OpenClaw Weekly Dev Experience · slide 9 / 17
June 5, 2026
Providers got wider without going feral
Coverage + bounded failures
More model and media surface area landed with bounded timers, retries, OAuth lifetimes, polling, and failure shapes. Boring constraints, blessedly useful.
- Timers, retries, OAuth and device-code lifetimes capped
- Media downloads and generated-content polling bounded
- Provider catalog and reasoning-output failures get clearer
- Claude Opus 4.8
- Fal Krea schemas and NVIDIA featured models
- Streaming music and provider-backed voice catalogs
- Encrypted PDF extraction and structured tool content
OpenClaw Weekly Dev Experience · slide 10 / 17
June 5, 2026
Long runs stopped leaving bodies behind
Agent + Codex recovery
The runtime paid down failures where an agent half-completes the job and leaves everyone staring at a corpse called "session still running".
- Subagents keep cwd and workspace separation
- Hook context stays prompt-local
- Session locks release on timeout abort; live locks survive cleanup
- Interrupted tool calls, stale bindings, compaction handoffs, auth-profile failover recover more cleanly
- App-server and helper failures no longer tear down shared runtime state
- Delegated workflows have a clearer plugin/supervisor path
- Fewer zombie sessions and false live switches
OpenClaw Weekly Dev Experience · slide 11 / 17
June 5, 2026
Workboard, Chat, and Control UI got less fragile
Continuity over flash
Streams stay alive, drafts stay local, sends reconcile, and the planning surface is starting to coordinate actual multi-agent work.
- Sends survive history loading
- Stream deltas incrementally; markdown work waits until it should
- Drafts stay local while typing
- Completed sends reconcile instead of going dark
- Orchestration primitives and task-backed board runs
- Task comments in the edit modal
- Keyboard movement controls
- Board is moving toward multi-agent coordination, not static task furniture
OpenClaw Weekly Dev Experience · slide 12 / 17
June 5, 2026
Plugin installs need policy, not vibes
ClawHub + external packaging
External plugin packaging, ClawHub trust surfaces, and install policy are replacing the old dangerous-code-scanner theatre.
- GitHub Copilot and Tokenjuice moving to official external plugin packaging
- npm and ClawHub metadata
- Plugin display names, skill verification, trust surfaces on ClawHub
- Explicit policy, lifecycle, doctor checks, metadata, rollback
- Install decisions become inspectable
- The next bar is auditability from click to rollback
OpenClaw Weekly Dev Experience · slide 13 / 17
June 5, 2026
Safety moved into runtime behavior
Fail closed, leave receipts
Malformed input gets rejected earlier. Release, CI, Docker, and E2E lanes cap logs, response bodies, probe waits, and artifact checks.
- Browser timeouts, viewport and tab indices, Gateway ports
- Cron retry handling, Discord component ids, schema array refs
- Telegram callback pages and channel progress callbacks
- Release, CI, Docker, E2E, plugin install, update, doctor, diagnostics, security lanes
- Capped logs, response bodies, readiness probes, child workflow waits, artifact checks
- Failures report bounded proof instead of hanging or false-greening
OpenClaw Weekly Dev Experience · slide 14 / 17
June 5, 2026
Easier to trust when state is visible
DX snapshot
DX improved because more state moved into view. The next fight is one canonical explanation across UI, channel, provider, plugin, and delegated runtime.
- Providers time out more cleanly
- Channels preserve final replies better
- Mobile sessions keep more state
- Skill Workshop gives skill changes a visible review path
- One run can touch mobile, channel, provider auth, skill install, Workboard task, and Codex helper
- When one surface drifts, users need one explanation: model, auth, context, channel, workspace, tools, policy, proof
OpenClaw Weekly Dev Experience · slide 15 / 17
June 5, 2026
Three questions for builders and maintainers
Operating discipline
Aim the conversation at operating discipline: safer installs, visible state, and proof packets that survive the chaos-goblin phase of agent work.
- Where should install policy be strict by default, and where should expert builders get escape hatches?
- Which state must be visible before a run starts: model, provider, auth, context budget, channel target, workspace, tools, policy scope?
- What is the smallest proof packet every agent task should leave?
- Skill Workshop adoption: are proposals, review states, support-file approvals, and rollback metadata actually used?
- Mobile and channel durability under real Telegram, WhatsApp, iMessage, Discord, Talk usage
- Provider bounds: can non-maintainers understand and fix timeout/catalog failures?
OpenClaw Weekly Dev Experience · slide 16 / 17
From agent runs to agent operations
Close
The weekly story is bigger than a changelog. Stable releases made OpenClaw more survivable. Microsoft made the distribution story real. The next bar is simple: visible state, explicit policy, durable delivery, and proof every time.
- 5.28 and 6.1 moved recovery, channels, providers, media, skills, and proof lanes forward.
- Windows Companion and Microsoft Scout show OpenClaw becoming desktop and enterprise infrastructure, more than a clever dev tool.
- Every agent run should say what it used, why it chose it, what changed, and which evidence proves it worked.
OpenClaw Weekly Dev Experience · slide 17 / 17