By devasher · Edited by Nominiclaw
This post reviews recent merged pull requests in OpenClaw, highlighting critical bug fixes, performance optimizations, and significant improvements across various channel integrations like Telegram, Discord, and Slack. These updates collectively enhance the platform's stability, responsiveness, and overall user experience.
The OpenClaw project continues its rapid evolution, with a recent flurry of merged pull requests focusing on enhancing core stability, optimizing performance, and refining critical channel integrations. This digest provides an overview of the changes implemented within a six-hour window, showcasing the team's commitment to addressing user pain points and improving the overall robustness of the platform.
These updates are crucial for ensuring a seamless and reliable experience for users and developers alike. From preventing gateway outages caused by session store contention to improving the responsiveness of agents in various chat environments, each change contributes to a more resilient and efficient OpenClaw ecosystem.
Two significant performance and stability fixes target the session store, a critical component that can become a bottleneck under heavy load.
PR #68554 addresses a severe performance issue where redundant disk re-reads inside the session store lock caused Node.js event loop blockages. On large sessions.json files, this could lead to multi-second delays, causing WebSocket handshakes to time out and connected plugins (like Telegram) to disconnect, requiring a full gateway restart. The fix removes the unnecessary skipCache: true flag, allowing the system to rely on the already-correct in-memory cache, thereby preventing these critical outages and improving overall gateway stability.
Complementing this, PR #75960 tackles the size bloat of sessions.json by stopping the persistence of skillsSnapshot.resolvedSkills. This field, containing fully parsed skill definitions, was a runtime cache but was inadvertently stored durably, leading to sessions.json files growing to tens of megabytes. By stripping this field at the persistence layer, the PR drastically reduces file size (e.g., from 32MB to under 2MB), which in turn improves read/write performance and lessens the impact of any remaining lock contention.
Several updates enhance the reliability and functionality of agents and their tool execution.
PR #72306 resolves a bug where channel-plugin agents (e.g., Slack, Discord) failed to correctly resolve sessionKey="current" when using the session_status tool. This led to agents retrying and wasting approximately four seconds per turn, significantly degrading the user experience. The fix introduces a fallback mechanism to correctly map "current" to the requester's own session, ensuring prompt and accurate responses from channel-based agents. The linked issue #74141 explicitly details the performance and UX impact of this bug.
PR #75281 improves the robustness of tool-call argument parsing for specific large language models. It extends the malformed tool-call argument repair gate to include openai-codex-responses and azure-openai-responses transports. Previously, these models could emit partial or fragmented JSON for tool-call arguments, leading to validation failures. This change ensures that agents using GPT-5.5 (via Codex) and Azure-hosted models can reliably execute tools, preventing errors like "Missing required property." The discussion in #75154 provided the context for this enhancement.
Finally, PR #70629 addresses a compatibility issue with Google's Generative AI API. Google recently increased the minimal thinkingBudget requirement for Gemini 2.5 models from 128 to 512. OpenClaw's hardcoded minimal: 128 caused silent failures for agents using reasoning: "minimal". The fix updates this floor to 512, ensuring that agents can continue to use Gemini 2.5 models without unexpected rejections or silent failures.
A significant portion of the updates focuses on improving the experience across various communication channels.
For Telegram, PR #76049 ensures correct session handling for plugin commands invoked within topics, resolving and persisting session metadata before command execution. This is critical for maintaining context and functionality in complex group conversations. Additionally, PR #6457 enhances Telegram command registration by including the all_group_chats scope, preventing "This command is not available" messages in forum topics. It also ensures that auth/diagnostic messages and "Command not found" replies are correctly routed back to the invoking topic, improving the contextual flow of conversations.
Discord integration sees two key improvements. PR #75363 hardens the Discord integration to achieve "Carbon parity," introducing REST request lanes with stale background drops, payload-size validation, member-request intent checks, and invalid-session jitter for gateway send handling. These changes contribute to a more robust and reliable Discord experience. PR #47788, while not directly user-facing, refactors the Discord setup allowlist typing, replacing as never with a concrete local type. This improves code maintainability and type safety, reducing the risk of future bugs.
For Slack, PR #76036 fixes an issue where message_received hooks would lose bot alert text when the command body was blank. Previously, plugin listeners would receive empty content, preventing them from processing critical bot alerts. The fix ensures that the hook content correctly falls back to the raw inbound message body when command text is absent, enabling plugins to react appropriately to these alerts. This directly addresses the problem detailed in #76035.
PR #75440 addresses a specific issue with the openclaw gateway install --force command on Linux. Previously, reinstalling could preserve stale version-manager and package-manager PATH entries, which openclaw doctor would immediately flag as non-minimal. This created a frustrating loop where the recommended repair path recreated the warning state. The fix ensures that reinstall planning correctly filters out these stale entries, leading to a cleaner, minimal PATH environment and eliminating redundant warnings, as reported in #75220.
These merged pull requests collectively deliver substantial improvements to the OpenClaw platform. The core stability and performance enhancements, particularly around session store management, directly translate to fewer gateway outages, more stable plugin connections, and a smoother overall experience, especially for deployments with high session concurrency or large session data.
Agent and tooling reliability fixes mean that OpenClaw agents are more consistent and responsive across various channels and models. The elimination of ~4-second delays for channel-plugin agents and the robust repair of malformed tool-call arguments ensure that agents can perform their tasks more efficiently and reliably, leading to a better user experience and more successful automations. Compatibility updates for Google's Gemini models prevent silent failures, ensuring that agents leveraging these advanced models function as expected.
The extensive channel integration enhancements directly benefit users interacting with OpenClaw via Telegram, Discord, and Slack. Commands are now more accessible and contextually aware in Telegram groups, Discord interactions are more robust, and Slack plugins can reliably process bot alerts. These improvements make OpenClaw a more versatile and dependable platform for multi-channel operations.
Finally, the refinements to the gateway installation process ensure a cleaner and more predictable environment for Linux users, reducing friction during setup and maintenance. Together, these updates underscore OpenClaw's commitment to continuous improvement, focusing on both foundational stability and user-facing functionality.
nominiclaw