WeeklyCLAWEp. 17
Weekly Claw · Ep. 17

OpenClaw Change log & Dev Experience
Weekly Claw Ep. 17

Two stable releases, 1,265 commits, and a week mostly spent tightening the runtime boundaries.

Release window
  • v2026.6.5 on June 9
  • v2026.6.6 on June 12
  • June 5 to June 12, 2026
Main story
  • Security surfaces fail closed more often
  • Channels survive restarts more cleanly
  • Release proof got stricter
week17-boundaries.html

OpenClaw Weekly Dev Experience · slide 1 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Community pulse

The week in raw signal
Throughput

A fix-heavy, test-heavy week with broad contribution and two stable releases.

Numbers
  • 1,265 commits
  • 131 authors
  • 187 release issue refs across 6.5 and 6.6
Commit mix
  • 794 fixes
  • 191 tests
  • 117 docs
  • 30 features
Top contributors
  • Vincent Koc 525
  • Shakker 172
  • Peter Steinberger 118
  • Ayaan Zaidi 55
  • joshavant 52

OpenClaw Weekly Dev Experience · slide 2 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Release map

Two stable releases, one hardening arc
6.5 + 6.6

The release train split into two useful stories: richer provider/channel support in 6.5, then tighter boundaries in 6.6.

v2026.6.5
  • 70 release bullets
  • 83 issue refs
  • QQBot reasoning cleanup
  • Parallel web search provider
  • Auth and plugin install durability
v2026.6.6
  • 34 release bullets
  • 104 issue refs
  • Security boundary hardening
  • Telegram and iMessage recovery
  • Control UI startup speed

OpenClaw Weekly Dev Experience · slide 3 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Security

Boundaries got tighter, and they fail closed
Runtime safety

The largest 6.6 theme was reducing silent trust leaks across transcripts, tools, channels, and delegated runtimes.

Locked down
  • Transcript and image redaction tightened
  • Sandbox binds and host env inheritance restricted
  • MCP stdio and loopback tools hardened
  • Codex HTTP and native search policies narrowed
Fail closed
  • Exec approvals fail closed on timeout
  • Elevated sender checks improved
  • Deleted-agent ACP bypass closed
  • Discord moderation and Teams group actions bounded

OpenClaw Weekly Dev Experience · slide 4 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Channels

Delivery that survives restarts
Channel durability

Messaging reliability moved from best effort toward durable routing, dedupe, and restart recovery.

Telegram
  • Account-scoped topics route to the right agent
  • Streamed text survives tool calls
  • Generic ingress supports /compact
  • Durable dispatch dedupe moved into the SDK
Other channels
  • iMessage always-on inbound restart
  • WhatsApp replies attach to the successor controller
  • Feishu rate limits retry
  • Mattermost thread replies preserved

OpenClaw Weekly Dev Experience · slide 5 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Browser and MCP

Real sessions, safer edges
Connectivity

Browser and MCP integrations gained reach without turning every tool boundary into a trust fall.

Browser
  • Existing-session CDP support
  • Discovered WebSocket validation
  • Default-profile cdpUrl handling
  • Browser output boundaries tightened
MCP
  • Streamable HTTP loopback transport
  • OAuth and SSE authorization corrected
  • Richer tool results coerced before provider calls
  • Lease and timestamp handling got stricter

OpenClaw Weekly Dev Experience · slide 6 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Control UI

Startup got faster, first replies got visible
Latency

The UI work was not cosmetic. It attacked startup wait, first-event mystery, and model catalog churn.

Speed
  • Cached model metadata at startup
  • Startup catalog wait removed
  • Slash commands load lazily
  • TUI runtime plugins prewarm
Diagnostics
  • First-assistant-event traces
  • Slow-reply warnings
  • Active connection host visible in TUI
  • Derived-registry rescan storms reduced

OpenClaw Weekly Dev Experience · slide 7 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Providers

More providers, fewer weird failures
Model surface

Provider support expanded while reasoning replay, compaction ownership, and local-model behavior got cleaner.

New support
  • OpenRouter OAuth onboarding
  • Claude Fable 5 adaptive thinking
  • Parallel bundled as web_search
  • Google Vertex ADC rows and runtime resolution restored
Correctness
  • Codex compaction ownership preserved
  • Local models skip guardian review
  • Gemma 4 reasoning replay preserved
  • Provider cooldown recovery improved

OpenClaw Weekly Dev Experience · slide 8 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Cron and plugins

Jobs and installs left better receipts
Lifecycle

Scheduled runs, plugin installs, and ClawHub flows became harder to fake-green.

Cron
  • Active task runs cancel cleanly
  • Terminal timeout state is preserved
  • No-deliver tool warnings recover
  • Legacy cron JSON stores migrate during doctor preflight
Plugins
  • Dry runs skip publish approval
  • Trusted hooks can be declared
  • Managed plugin drift is reported
  • Official npm plugin pins stay intact

OpenClaw Weekly Dev Experience · slide 9 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Release proof

Release proof got stricter
Evidence lanes

The release system now fails louder when proof is missing, oversized, or impossible to inspect.

Gates
  • Beta pages stay draft until npm and plugin proof pass
  • QA Lab requires runtime tool-call evidence
  • Resource ceilings are enforced, not only logged
  • Docker and E2E diagnostics are bounded
Receipts
  • npm package and tarball integrity
  • Release SHA and CI run links
  • Full evidence reports for 6.5 and 6.6
  • Plugin publish proof separated from core publish proof

OpenClaw Weekly Dev Experience · slide 10 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
DX review

The week through a builder's eyes
Dev Experience

Builders got faster starts, safer installs, and fewer channel ghosts. Still not glamorous. Correctly so.

Less friction
  • Faster Control UI startup
  • First replies are easier to diagnose
  • OpenRouter OAuth reduces setup drag
  • Dry-run plugin flows avoid noisy approvals
More trust
  • Approvals fail closed
  • Releases carry verifiable evidence
  • Tool output boundaries are stricter
  • Local model paths avoid unnecessary review

OpenClaw Weekly Dev Experience · slide 11 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Video signal

Build the thing that builds the thing
Peter Steinberger · BRK245

Peter's talk is the missing through-line for Week 17: OpenClaw is not only shipping features; it is building the feedback loops that let agents ship safely.

Worth calling out
  • Build tools for agent feedback, not just faster human coding
  • Close the loop with screenshots, tests, video, and maintainer receipts
  • Treat issues and PRs as prompt requests that can be triaged repeatedly
  • Use project invariants and vision files so fresh agents know the rules
Why it fits this week
  • Week 17 hardened approvals, channels, cron, plugins, and release evidence
  • The talk names the same bar: autonomy with proof, not autonomy by vibes
  • Crab Box, Mantis, Auto Review, and prompt provenance map directly to the DX story

Source: “Build the thing that builds the thing” / BRK245 · slide 12 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Community signal

The community is testing the edges
Signal

The busiest labels and release notes point to the same pattern: session state, proof, compatibility, and maintainer review pressure.

Repeated pressure
  • Session-state issues stayed prominent
  • Compatibility risk labels stayed loud
  • Proof supplied and proof sufficient both showed up heavily
  • Maintainer review queues remained visible
Good sign
  • A fix-heavy week is not stagnation
  • It means real users are finding seams
  • The runtime is being forced to explain itself

OpenClaw Weekly Dev Experience · slide 13 / 15

WeeklyCLAWEp. 17
OpenClaw Community
June 12, 2026
Signal map

Where to watch next
Next bar

The next useful work is not more surface area. It is clearer state, safer defaults, and receipts that survive handoff.

Watch
  • Provider onboarding and OAuth failure clarity
  • Channel delivery after restarts
  • Cron no-deliver and timeout reporting
  • Plugin install trust and rollback paths
Ask
  • Can a new builder explain the failure state?
  • Can maintainers audit the run without logs archaeology?
  • Can every risky action show its boundary?
  • Can proof stay small enough to inspect?

OpenClaw Weekly Dev Experience · slide 14 / 15

Summary

OpenClaw is becoming operational infrastructure
Close

Week 17 was not a shiny-feature week. It was better: the runtime got stricter about who can act, what can leak, how channels recover, and what proof must exist before a release ships.

Bottom line
  • Two stable releases
  • 1,265 commits
  • Security and delivery hardening
  • A clearer path from agent demos to agent operations

OpenClaw Weekly Dev Experience · slide 15 / 15