OpenClaw v2026.6.8

Schema Quarantine · GLM-5.2 · Kimi K2.7 · Telegram Rich Text

June 16, 2026 · Stable Release · 143 PRs · 66 Issues

At a Glance

143
Pull Requests
66
Linked Issues
34K
Changelog chars

Top Highlights

🛡️ Tool Schema Quarantine (15+ PRs)

OpenAI/Anthropic tool schemas are now quarantined when unreadable or post-hook. Sustained security pass across quarantine unreadable OpenAI/Anthropic tool schemas series — 8+ Anthropic PRs, 10+ OpenAI PRs.

🤖 Three New Model Providers

GLM-5.2 (Zhipu AI), Kimi K2.7 Code (Moonshot), Claude Haiku 4.5 catalog entries. Plus OpenRouter/Vertex provider-prefix normalization, OAuth image routing through Codex, LM Studio binary-thinking support.

📨 Telegram Rich Text

Structured rich text: tables, lists, expandable blockquotes, preserved intentional line breaks, prompt-preserving CLI backend delivery, safer rich-media boundaries. WhatsApp now honors configured ACP bindings.

🔐 Reasoning-Signature Recovery

Anthropic thinking-signature recovery on replay + OpenAI reasoning-signature recovery. Storeless OpenAI Responses replay compatibility preserved. Eliminates a major class of replay bugs.

New Model Support

GLM-5.2 (Zhipu AI)

Full provider support added. Chinese frontier-class model. #92796

Kimi K2.7 Code (Moonshot)

Coding-specialized variant. Long-context reasoning. #92554

Claude Haiku 4.5

Anthropic's fast/cheap tier — now in static catalog. #90116

OpenRouter / Vertex Normalization

Provider-qualified model IDs normalized across OpenRouter and Google Vertex paths. #92627, #91218

Channel Engineering

Telegram — Rich Text

Tables, lists, expandable blockquotes, line breaks, prompt-preserving CLI delivery, native draft migration retired, safer rich-media boundaries.

WhatsApp — ACP Bindings

Honors configured ACP bindings. Login fix: no longer reports success before auth durably persisted.

Slack — Hooks + Validation

Outbound message_sent hooks. Warns on name-keyed channel map. Same-channel final reply mirroring.

Discord — Auto-thread Budgets

60s timeout + 4,096-token reasoning-model budget for auto-thread titles. Default-enabled (no plugins.entries.discord.enabled required).

Schema Quarantine — Sustained Security Pass

A 15+ PR series hardening the agent runtime against malformed tool schemas from upstream providers.

OpenAI Family (10+ PRs)

  • quarantine unreadable tool schemas — #92921
  • guard OpenAI transport tool descriptors — #89016
  • guard OpenAI tool schema conversion — #89378
  • harden OpenAI strict schema inspection — #89543
  • guard OpenAI strict tool diagnostics — #90200
  • skip unreadable OpenAI completion tool schemas — #90283
  • skip unreadable OpenAI responses tool schemas — #90286, #90397
  • guard post-hook tool payloads — #92928
  • guard responses tool payload names — #89703

Anthropic Family (8+ PRs)

  • quarantine invalid direct tool schemas — #92896
  • quarantine invalid projected tools — #89418
  • guard Anthropic tool descriptors — #89221
  • guard Anthropic tool schema conversion — #90228
  • skip unreadable Anthropic tool schemas — #89622, #90278
  • guard Anthropic provider tool descriptors — #89229
  • quarantine unreadable Anthropic payload tools — #92908

Reasoning-Signature Recovery

Anthropic thinking-signature recovery (#92286, #92916, #92373)

Embedded runner: freshly streamed thinking signatures intermittently invalid on replay. Recovery wrapper now fires for genericized error text. History thinking-block stripping when thinking is disabled.

OpenAI reasoning-signature recovery (#92941)

Recover invalid reasoning signatures in OpenAI Responses API. Storeless replay compatibility preserved. gpt-5.5 tool reasoning effort omitted where unsupported.

What this means

Replays of long agent sessions no longer fail when thinking-block signatures degrade across provider round-trips. Reliability for long-horizon agents materially improved.

Agent & Gateway Runtime

/btw CLI Command

Now supported in CLI-backed sessions. CLI usage-error exits classified as usage failures, not successful runs.

Trusted Subagent Override

Trusted subagent thinking override clamped through provider/model fallback. Yielded media completions preserved.

Admin-Gated Surfaces

HTTP session/model override requires admin privileges. session_kill endpoint protected. Subagent pauses on abort signal.

OpenAI Agent Selector

Unknown OpenAI agent selectors explicitly rejected. wrapToolWithBeforeToolCallHook made idempotent to prevent double execution.

Memory, State & Diagnostics

Embedding Batches Split

Oversized OpenAI embedding batches split before 431s. Nested embedding fetch failures now surfaced.

QMD Memory in Transient Mode

memory_search stays available in transient QMD mode. Skipped short-term recall hits explained.

SQLite NFS Hardening

WAL avoided on NFS state volumes. Full reindex preserves rollback/cache recovery.

Recovery Scheduling

Stuck-session recovery no longer resets warning backoff. Reset archive fallback for missing async transcripts.

UI / Mobile / TUI

WebChat Backscroll

Survives streaming — previously lost when long responses came in.

Workspace Files Collapse

Can collapse and start collapsed. Cleaner default state.

Sidebar Session Picker

Remains interactive above desktop workbench.

iOS Foreground Gateway

Stale foreground gateways force reconnect. TUI parent stdin stays paused after exit.

Web Search Provider Opt-In

Key-free providers no longer auto-selected (#93616)

Previously, Parallel Free, DuckDuckGo, Ollama, and Codex Hosted Search were automatically chosen when no API-backed provider was configured. This silently degraded result quality for many users. They are now explicit opt-ins — defaults now require a configured API-backed provider.

Other changes

  • Hono bumped to 4.12.25 — patched runtime for OpenClaw and ACPX packages
  • /models and models.list catalog loading now bounded — fixes perf regression from v2026.6.1
  • Profile auth honored for SecretRef model entries
  • OpenClaw cron status reports SQLite storage path (not legacy jobs.json)

Migration & Upgrade

Standard upgrade path

npm install -g @openclaw/cli@latest
openclaw update stable
openclaw doctor

Heads-up for plugin authors

  • OpenAI/Anthropic tool schemas may now be quarantined if malformed — ensure your plugin emits valid JSON Schema
  • Key-free web search providers require explicit opt-in — set web_search.provider in config
  • Discord channel is now default-enabled — disable explicitly if you don't want it

What does NOT need migration

  • Existing channel configs (Telegram/WhatsApp/Slack/Feishu) — all backwards-compatible
  • Existing model selections — GLM-5.2, Kimi K2.7, Haiku 4.5 are additive
  • Memory store layout — same on-disk format

Verdict

Release scale MAJOR — 143 PRs, 66 issues
Security posture IMPROVED — 15+ schema-quarantine PRs
Model coverage +3 NEW (GLM-5.2, Kimi K2.7, Haiku 4.5)
Channel richness TELEGRAM RICH TEXT — tables, lists, blockquotes
Agent reliability REASONING-SIGNATURE RECOVERY
Behavioral change KEY-FREE WEB SEARCH NOW OPT-IN

Resources