OpenClaw Change log & Dev Experience
Weekly Claw Ep. 18
A week of richer channels, safer routing, and recovery loops that make agent work visible instead of mysterious.
- v2026.6.6 on June 12
- v2026.6.8 on June 16
- June 12 to June 19, 2026
- Channel delivery became less brittle
- Agent runs recovered more often
- Model/provider routing got safer
OpenClaw Weekly Dev Experience · slide 1 / 14
June 19, 2026
The week in raw signal
Throughput
The window stayed fix-heavy, but the center of gravity moved: channel delivery, session recovery, model routing, and provider/plugin packaging.
- 1,000 commits sampled
- 69 authors
- 500 changed issues / PRs
- 459 fixes
- 367 refactors
- 81 tests
- 24 features
- Vincent Koc 818
- github-actions[bot] 18
- Josh Lehman 17
- Shakker 17
- joshavant 13
OpenClaw Weekly Dev Experience · slide 2 / 14
June 19, 2026
Stable releases covered
6.6 + 6.8
Week 18 source capture found two stable releases plus a fresh beta carrying the next wave of delivery and recovery work.
- Published 2026-06-16T16:32:26Z
- openclaw 2026.6.8
- Current stable release in window
- Published 2026-06-12T11:04:42Z
- openclaw 2026.6.6
- Earlier stable release in window
OpenClaw Weekly Dev Experience · slide 3 / 14
June 19, 2026
Channels got less brittle
Delivery
Telegram and WhatsApp changes point at a simple operator promise: the agent can produce rich work, and the channel should not flatten it into mush.
- Tables, lists, expandable blockquotes, line breaks, stickers, media, and CLI-backed replies are preserved more faithfully
- DM sends, thread replies, Mattermost context, Slack final replies, and Discord action handling were tightened
OpenClaw Weekly Dev Experience · slide 4 / 14
June 19, 2026
Agent runs recovered more often
Less dead air
The release notes keep circling the same concern: partial turns, interrupted sessions, yielded subagents, compaction, and replay all need visible endings.
- Thinking-only retries
- Duplicate-hook prevention
- Partial JSON repair
- Truncated-response detection
- More runs end with an artifact
- Failures are easier to name
- Long-running work has clearer handoff points
OpenClaw Weekly Dev Experience · slide 5 / 14
June 19, 2026
Model routing got safer
Provider surface
GLM-5.2 entered the catalog, provider IDs were normalized, and tool-schema recovery kept model/provider boundaries less sharp-edged.
- Model catalog carries GLM-5.2 with normalized provider-qualified IDs
- Cron primary restored to citadel-all/zai/glm-5.2 for the GLM variant
- Gemini CLI OAuth behind proxies
- Bedrock embeddings
- Codex Spark OAuth routing
- OpenAI/Anthropic replay recovery
OpenClaw Weekly Dev Experience · slide 6 / 14
June 19, 2026
Official providers became real packages
Packaging
External provider packages and startup discovery are boring in the best way: less bespoke install state, more repeatable fleet operations.
- Official providers can ship as independent npm releases
- External channel plugins load at Gateway startup
- Managed plugin recovery hardened
- Package validation improved
- StepFun boundary stays explicit
OpenClaw Weekly Dev Experience · slide 7 / 14
June 19, 2026
Failure modes became legible
Dev Experience
Week 18 is not just feature velocity. It is work on the places agents usually embarrass themselves: context, delivery, replay, and status.
- Cron status reports SQLite storage paths
- CLI usage errors stay classified as usage failures
- Session history repair preserves debugging evidence
- Credential-aware usage footers
- Compaction usage preservation
- Delivery receipts instead of ghost runs
OpenClaw Weekly Dev Experience · slide 8 / 14
June 19, 2026
The support story is shifting
Field signal
The week’s fixes read like a field report from users running agents in real channels, not demos.
- Telegram
- Mattermost
- Slack
- Discord
- Mobile clients
- Codex Hosted Search landed
- Key-free search providers stayed explicit opt-ins
- Control UI, iOS, Android, and WebChat reduced handoff friction
OpenClaw Weekly Dev Experience · slide 9 / 14
June 19, 2026
Autonomy with receipts
Proof loops
The strongest thread is the same one Weekly Claw keeps returning to: agents need proof loops, not just bigger prompts.
- Issues, PRs, channel threads, and release notes become promptable operating surfaces
- Screenshots, tests, status, usage, replay, and delivery receipts close the loop
- Humans can inspect what changed, what failed, and what shipped
OpenClaw Weekly Dev Experience · slide 10 / 14
June 19, 2026
Build the thing that builds the thing
BRK245
Peter Steinberger’s talk remains the right lens: build the harness that lets agents prove their work.
- Crab Box, Mantis, Auto Review, and prompt provenance map onto the current DX direction
- Agent tools should make work observable, testable, and repeatable
- The loop is the product
OpenClaw Weekly Dev Experience · slide 11 / 14
June 19, 2026
What operators should notice
Receipts
The useful work is in the dull seams: delivery, replay, provider boundaries, and explicit proof that a run did what it claimed.
- Rich channels
- Recovered sessions
- Provider-qualified model IDs
- Packageable providers
- Cron must execute tools, not narrate plans
- Discord and Answer Overflow sampling should be deeper
- Canonical changelog route needs DNS/live proof
OpenClaw Weekly Dev Experience · slide 12 / 14
June 19, 2026
Week 18 variant handling
GLM + Opus
This correction preserves both model-route variants and stops the cron from overwriting one with the other.
- GLM-primary preserved route
- Cron primary restored to citadel-all/zai/glm-5.2
- Opus-route preserved copy
- Opus remains available as fallback/explicit run
- Do not replace week18 without explicit selection
- Keep changelog aliases generated
OpenClaw Weekly Dev Experience · slide 13 / 14
The useful theme: less magic, more receipts
Close
OpenClaw’s week was about making agent work survive the messy middle: channels, sessions, models, plugins, and humans checking the result.
- Two stable releases
- 1,000 commits sampled
- 69 authors
- Channel and recovery hardening
- No Google Slides artifact
OpenClaw Weekly Dev Experience · slide 14 / 14