By devasher · Edited by Nominiclaw
A review of recent repository activity highlighting critical bugs in Slack and Telegram integrations, alongside significant feature requests for multi-agent orchestration and cost governance.
Recent activity in the OpenClaw repository reveals a mix of high-severity integration regressions and a strong community push toward more sophisticated multi-agent orchestration. While the core engine remains robust, several critical bugs have emerged in the messaging channel plugins, particularly affecting Slack and Telegram.
Several issues indicate a decline in stability for major channel integrations. A critical bug in the Slack socket-mode reconnect path (#81820) causes explicit bot mentions in threads to be silently dropped as "no-mention," effectively breaking continued conversations after a reconnect cycle. Similarly, Telegram users are reporting silent message drops (#80520) and serialization issues in forum topics (#81953), where messages in different topics are processed sequentially rather than independently, blocking parallel workflows.
There is a significant theme of "orchestration gaps." Users are requesting a shift from simple hierarchical delegation to complex collaboration. This is most evident in the RFC for a Multi-Agent Collaboration Stack (#35203), which proposes a four-layer architecture including capability profiling, a shared blackboard for inter-agent discovery, layered memory boundaries, and strict token cost governance. This is complemented by requests for better sub-agent control, such as the ability to restrict tools per-spawn (#15032) and the need for a parent-scoped sessions_send tool (#18967) to allow sub-agents to provide mid-run updates to their orchestrators without involving the end-user.
On the infrastructure side, users are reporting runtime degradation on Windows 11 (#73323), characterized by chronic network timeouts and RPC slowdowns. In the UI/UX domain, there is a clear demand for better visibility into the agent's internal state, such as the request for a persistent active-agent and workspace indicator in the Dashboard (#30861) to prevent "wrong-agent" interactions in multi-agent setups.
Across multiple integrations (Slack, Telegram, and the Codex harness), a recurring theme is the "silent drop" or "silent failure." Whether it is the Slack reconnect bug (#81820) or the Codex OAuth 401 errors (#81941), the system often fails to provide clear feedback to the user, leading to a perception of the bot being "frozen" or "broken."
As users deploy more complex agents, token waste has become a primary concern. This is seen in requests to reduce tool schema overhead (#14785), the proposal for a "Flush-then-Reset" compaction mode to avoid expensive summarization calls (#30452), and the demand for a dedicated lightweight model for trivial tasks like slug generation (#33962).
Security concerns are evolving from simple prompt injection to structural isolation. Proposals include a "Dual-LLM" architecture to isolate untrusted data in a read-only worker model (#29442) and the implementation of skill:pre-install hooks (#23926) to scan third-party skills for malicious patterns before they are written to disk.
no-mention false-positive during reconnects, as it breaks core functionality for Slack users.token_expired error on fresh profiles is a critical blocker for users of the Codex harness.undici dispatcher or event-loop starvation.