OpenClaw Issue Digest: Session Concurrency, Routing Stability, and Resource Management

Open Issues

Recent activity in the OpenClaw repository reveals several critical stability and behavioral issues, primarily centered around session concurrency, routing logic, and resource management.

Session Concurrency and State Corruption

One of the most severe reported issues is the EmbeddedAttemptSessionTakeoverError (#84829, #83615). This occurs when a second message arrives in the same agent session before a previous LLM call completes. Because the session lock is released during inference, the second message can overwrite the session file, invalidating the first call's fingerprint and causing the first response to be silently discarded. This results in burned tokens and a broken user experience where the agent appears frozen.

Additionally, there are reports of session bloat (#45718), where skillsSnapshot and systemPromptReport fields are accumulated on every run, causing sessions.json to grow unbounded and eventually trigger context overflow errors.

Routing and Integration Regressions

Routing stability remains a concern across several channels:

• Telegram: Issues include DMs polluting the agent:main:main session (#41165) and a regression where requestHeartbeat with intent: "immediate" fails to wake idle agents (#84841).
• Discord: A significant preflight/mention-gating bug (#44025) allows messages to pass through when they should be dropped, such as when another configured agent is mentioned or when mentions appear only in quoted text.
• Feishu: Reports indicate that read tool results for images are lost before the final outbound payload is assembled (#41744).

Resource Pressure and Gateway Stability

Several issues highlight systemic pressure on the gateway process:

• Event-Loop Starvation: CPU saturation is causing fetch timeouts to fire significantly late (#84835), while zombie Chrome processes are blocking CDP ports and causing gateway timeouts (#41750).
• Memory and I/O Pressure: Large compaction checkpoint artifacts are driving excessive read I/O and memory pressure (#84822). Furthermore, a leak of unclosed fs.FileHandle references on session locks is causing fatal crashes on Node.js 24+ (#84820).
• Log Bloat: The gateway.err.log can grow to several gigabytes due to repeated WebSocket heartbeat timeout lines without built-in rotation (#41676).

Key Themes

1. The "Race to the Lock"

There is a recurring theme of failures stemming from how OpenClaw handles session locks. Whether it is the EmbeddedAttemptSessionTakeoverError or the ERR_INVALID_STATE crashes on Node 24, the common thread is a failure to safely manage the lifecycle of session-store access during concurrent or high-load operations.

2. Tool-Call Reliability and UX

Several reports point to a gap between tool execution and user-facing reporting. For instance, the "confabulation" issue on Telegram (#41824) shows agents claiming to have completed actions that never actually executed, likely due to a lack of enforced tool-first ordering on chat surfaces.

3. Configuration Fragility

Users are reporting that unrecognized keys in openclaw.json can cause the gateway to enter a crash-loop (#41372), and that certain CLI commands for cron management do not match the provided documentation (#41372).

Action Required

High Priority / Blocked

• Fix EmbeddedAttemptSessionTakeoverError (#84829): This is a critical failure in the core session-locking mechanism that leads to silent data loss and token waste.
• Resolve Node 24 FileHandle Crashes (#84820): This is a process-level failure that causes the gateway to crash every few hours on newer Node.js runtimes.
• Address Discord Mention-Gating (#44025): This is a security and routing failure that allows unauthorized or incorrect agent triggers.

Immediate Contributor Attention

• Implement Log Rotation (#41676): To prevent disk exhaustion on long-running production gateways.
• Fix sessions_yield Logic (#84839): Correct the bug where parents are awoken with a yield continuation instead of a child completion event, which breaks sub-agent delegation workflows.
• Sanitize Session Keys for Plugins (#84789): Fix the crash in the active-memory plugin caused by colons in Telegram forum session keys.