openclaw
Development Feed
Development Feed
Teams deployments can now disable typing indicators with a single config flag, and duplicate typing bubbles in DMs have been eliminated — no more double "●●●" while waiting for a response.
The Google Gemini CLI integration now correctly extracts reply text and cache-aware usage metadata from JSON responses, closing a gap where CLI-backed turns could silently drop content.
Direct Google Generative AI requests can now include explicit cachedContent handles, restoring the ability to target prebuilt Gemini context caches from model-level runs.
Fixes ensure Anthropic prompt-cache markers only apply to genuine OpenRouter routes, not arbitrary OpenAI proxies that happen to use the same provider identifier.
GitHub Copilot requests authenticated through the IDE were failing with missing Editor-Version errors. A fix now ensures IDE identity headers are attached to both runtime model requests and the GitHub token exchange.
Exec approvals now respect configured security overrides, and stale cron/CLI tasks are properly cleaned up based on actual job activity rather than session persistence.
Agents now reuse AI model cache more reliably when prompts are semantically identical but differ in whitespace, line endings, or capability ordering.
Dead Google Chat streamMode configuration paths are being removed from schemas and documentation, along with legacy UI hints in Slack and Discord that exposed migration-target settings.
OAuth token parsing, HTTP request configuration, and API key resolution logic is being extracted from individual provider entrypoints into shared modules, eliminating hundreds of lines of duplicated code.
Deprecated configuration keys that once normalized into their canonical counterparts have been removed from public type exports, ensuring developers only see the modern API surface.