Changelog — June 4, 2026

claude agents --json now includes a waitingFor field that shows exactly what a waiting session is blocked on — for example, a pending permission prompt — making it easier to script or monitor multi-agent pipelines. Slash-command autocomplete changes behaviour: clicking a command in the menu now fills it into your prompt instead of running it immediately; press Enter to confirm. The /effort command now confirms when your chosen level will persist as the default for new sessions, removing the previous ambiguity about whether the setting was sticky. Remote Control moves from a startup banner to a persistent footer pill with a direct session link.

Two notable IDE changes: explicitly listing Grep or Glob in --tools now correctly provides the dedicated search tools on native builds (previously silently ignored), and Windsurf is renamed to Devin Desktop in the /ide menu, /terminal-setup, and /scroll-speed following the editor's rebrand.

Key bug fixes: startup no longer hangs when the config directory is read-only or unwritable — Claude Code now falls back to in-memory config and surfaces the error at startup instead of showing a blank screen. WebFetch permission rules now correctly apply to preapproved domains; explicit deny/ask/allow rules take precedence over the auto-allow list. Windows path permission rules now match paths spelled with backslashes or case-variant forms. An Esc interrupt sent at the very start of a turn is no longer silently dropped in stream-json/SDK sessions. MCP per-server timeout values below 1000 ms are no longer floored to a 1-second watchdog that was aborting fast MCP calls.

Release notes → Affects: /openclaw/, /openclaw/setup/, /openclaw/skills-guide/, /openclaw/configuration/, /openclaw/security/, /openclaw/cost-optimisation/, /claude-cowork/, /claude-cowork/setup/, /claude-cowork/skills-guide/, /claude-cowork/system-prompts/, /claude-cowork/pricing/

Temperature is now plumbed through the Responses API on the IronClaw web interface, giving web users the same temperature control previously only available via direct API calls. A bug where v1 conversation history was not correctly scoped for channel conversations is fixed — this could previously cause incorrect context bleeding between channels. WeCom is added as a new release distribution target for teams using that platform. CI improvements update the benchmark tracking reference and scope token permissions for reusable workflows.

Release notes → Affects: /ironclaw/, /ironclaw/setup/, /ironclaw/configuration/, /ironclaw/security/, /ironclaw/skill-allowlisting/, /ironclaw/vs-openclaw/

v7.3.28 ships several quality-of-life changes: free models that may use your prompts for training are now marked with a brain-circuit icon in the model picker, so the data-use implications are visible before you select a model. Marketplace skill installation is now resilient to missing project directories and concurrent install attempts. Post-compaction tool calls and follow-up messages now appear in the correct order after the compaction summary in both the CLI and VS Code transcript — previously these could appear out of order. Cloud Agent session transcripts are restored in VS Code session previews, and stalled cloud session loading no longer hangs indefinitely.

v7.3.29 focuses on JetBrains: hidden session UIs are now disposed after a configurable timeout, reducing memory usage for long sessions. Session stability is improved by keeping controller subscription state on the UI thread. Markdown code blocks now render as full-height multiline boxed editors instead of inline snippets. Streaming performance is improved by retaining existing markdown and code-block views while a response streams in, and raw fence markers are no longer shown during updates.

Two JetBrains release candidates also landed: rc.5 adds editor-backed code blocks and optimizes session UI layout; rc.6 ensures the model picker highlights models that can be used for training, consistent with the VS Code behaviour in v7.3.28.

Release notes → Affects: /kilocode/, /kilocode/setup/, /kilocode/models/, /kilocode/orchestrator/, /kilocode/security/, /kilocode/vs-claude-code/

NemoClaw's v0.0.58 release documentation is now published, covering new CLI commands and skill updates. Several important fixes ship alongside. A Hermes bug where AIAgent._strip_think_blocks was being converted to a staticmethod by the messaging response normalizer is fixed — this caused Hermes chat completions to fail with a missing content argument on Bedrock-compatible backends. Proxy environment variables (HTTP_PROXY, HTTPS_PROXY, NO_PROXY) are now correctly forwarded into the sandbox when running nemoclaw onboard, so network tooling inside the sandbox can reach corporate proxies.

GPU support on Windows ARM sees significant improvement: nemoclaw now accepts WSL2 + Docker Desktop GPUs on N1X hardware when a bounded CUDA proof succeeds using the correct aarch64 image (cuda-sample:vectoradd-cuda12.5.0). The previous proof image shipped an x86-64 ELF under its arm64 manifest tag, causing exec format error on real N1X hardware. The proof timeout is configurable via NEMOCLAW_WSL_GPU_PROOF_TIMEOUT_MS, and explicit exec format error output now distinguishes an image-architecture problem from a missing GPU. nemoclaw status also now reports the actual CUDA proof result rather than treating any configured GPU as healthy.

Two startup fixes: the Hermes startup sequence now clears a retryable Tirith download_failed marker before dispatching explicit commands, preventing a stale .tirith-install-failed file from persisting across restarts. When Docker is unreachable during gateway startup, nemoclaw onboard now fails immediately with platform-specific recovery guidance (colima start / sudo systemctl start docker / Docker Desktop) instead of waiting several minutes and producing generic diagnostics. Documentation updates clarify that adding a host to the sandbox egress allowlist does not bypass OpenShell's separate SSRF protection.

Commit log → Affects: /nemoclaw/, /nemoclaw/setup/, /nemoclaw/policy/, /nemoclaw/local-gpu/, /nemoclaw/switching-providers/, /nemoclaw/skills/

OpenAI announced the deprecation of three platform features: reusable prompt objects, the Evals platform, and Agent Builder. Shutdown timelines and migration guidance are on the OpenAI deprecations page. Teams that built evaluation pipelines on the Evals platform or created Agent Builder automations should review the migration path before the shutdown date. The container billing change that took effect June 2 (per-minute billing with a 5-minute minimum) continues to apply.

Release notes → Affects: /chatgpt/, /chatgpt/pricing/, /chatgpt/custom-gpts/, /chatgpt/tips/