OpenClaw
RTT release coverage backfilled and hardened
openclaw-rtt backfills missing RPC and Control UI coverage for the 2026.6.1 beta while making release imports rebase against their workflow ref, closing a known coverage gap.
read --wire →
$ tail -f topics/ai-agents.log
AI Agents news, every morning.
Daily updates from the AI-agent and agentic-coding ecosystem - LangChain, OpenHands, Cline, and LlamaIndex. What shipped this week in the frameworks and coding agents developers are building on.
61 updates across 2 projects this week.
Your week in AI Agents, every Monday.
One calm review of what shipped across AI Agents - the commits, releases, and security advisories that matter. Every Monday, with security advisories same-day. Free, unsubscribe in one click.
We'll start you on the top AI Agents projects - refine anytime. · Read a sample issue →
Recent wires
OpenClaw
Control UI chat startup coalesces model and command metadata
A new negotiated chat.metadata Gateway RPC coalesces model catalog and slash command loading for Control UI chat startup, cutting the first-message path free of that dependency graph.
read --wire →
OpenClaw
Channel model overrides now honored in agent ingress
OpenClaw now applies channel-level model overrides before the default model reaches fallback or auth in agent-command ingress.
read --wire →
OpenClaw
Copilot compaction moves to SDK state with no auth breakage
OpenClaw routes Copilot session compaction through SDK-backed state, dropping marker sidecars while preserving auth and session binding, and Windows nodes now reconnect after restart with the real app version.
read --wire →
OpenClaw
Node bootstrap pairing completed, generated-media prompts clarified
OpenClaw fixed the Windows node bootstrap handoff and reshaped how generated media is delivered across session types.
read --wire →
OpenClaw
Extracting LLM core, surfacing chat errors, hardening Windows bootstrap
OpenClaw extracts shared LLM contracts into new packages while the Windows gateway gets hardened reconnects and visible chat error messages.
read --wire →
OpenClaw
Windows node hardens signing; Crabbox adds recovery and pond mesh
Windows node release signing is now scoped to OpenClaw-owned EXEs with a verifier, while Crabbox can recover from AWS security group rule exhaustion and introduces a new pond grouping primitive.
read --wire →
OpenClaw
Preserving signed thinking payloads across providers
OpenClaw now preserves provider-owned signed thinking text byte-for-byte, fixing the regression where replay altered Anthropic thought blocks.
read --wire →
OpenClaw
OpenClaw unifies reply bypass logic, Crabbox adds delegated run output and Windows hydration
OpenClaw consolidates source-reply acknowledgement handling into one core change, Crabbox gains a delegated run session output flag, and Windows hydration support lands across the stack.
read --wire →
OpenClaw
Runtime auth snapshots preserved, AWS SSH ingress churn fixed
OpenClaw preserves runtime-only external OAuth profiles across saves, and Crabbox stops pruning active AWS SSH leases during refreshes.
read --wire →
OpenClaw
OpenClaw hands compaction back to Codex, crabs get experimental Wayland
OpenClaw stops preempting Codex-owned compaction, Crabbox adds an experimental Wayland desktop, and Codex usage-limit messaging grows actionable recovery paths.
read --wire →
OpenClaw
Linear-time glob matching closes O(n^k) ReDoS in session visibility
OpenClaw replaced regex wildcard matching with a linear-time glob matcher in session-visibility, fixing a polynomial backtracking vulnerability in the agentToAgent allow-pattern check.
read --wire →
OpenClaw
Slack approval QA checkpoints and Telegram poison fix land
OpenClaw ships Slack approval checkpoints for verifiable artifact proof and stops Telegram's isolated-ingress loop on missing harnesses.
read --wire →
OpenClaw
Awaiting agent_end hooks, preserving sessions after compaction failures
OpenClaw now awaits agent_end hooks in local CLI and Codex terminal paths, and auto-reply sessions survive compaction failures.
read --wire →
OpenClaw
Doctor warns on plaintext secrets, remote TUI startup skips plugin load
openclaw's doctor now flags plaintext secrets in config and the remote TUI skips plugin-heavy startup work, trimming cold start.
read --wire →
OpenClaw
ClawHub publish contention hardened, doctor checks go provider-aware
Package publishes now survive Convex write contention, and doctor checks validate Cloudflare and other providers read-only.
read --wire →
OpenClaw
ClawScan security classification now flows through Codex
ClawScan security classification now routes through a queued Codex worker that overrides VirusTotal decisions, making telemetry-only and trusting verified @openclaw/* packages by default.
read --wire →
OpenClaw
Harden skill transfers and sanitize Codex payloads
OpenClaw hardens skill transfer acceptance against banned requesters and sanitizes Codex image payloads against replay attacks.
read --wire →
OpenClaw
Stale bootstrap fix, Feishu streaming, checkpoint ledger
OpenClaw ships a stale-bootstrap fix, Feishu CardKit stream handling, a checkpoint ledger, and a dry-run diff for path edits.
read --wire →
OpenClaw
OpenClaw centralizes provider drift policy and hardens config repair forensics
Provider drift classification is now a shared policy across live tests, plus a config-repair fix closes a gap where the latest broken config copy was silently dropped.
read --wire →