Description

When a user manually stops an agent, the session file lock is never released, causing subsequent turn attempts to fail with SessionWriteLockTimeoutError for 60+ seconds.

Steps to Reproduce

1. Start any agent conversation
2. While agent is mid-response (e.g., waiting on LLM API), manually stop the agent
3. Immediately try to send another message to the same agent

Expected Behavior

Agent should recover and accept new messages immediately after abort.

Actual Behavior

Root Cause

File: selection-C4e-Qn9W.js (bundled), function abortRun:

const abortRun = (isTimeout = false, reason) => {
    aborted = true;
    // ...
    abortCompaction();
    abortActiveSession();
    // BUG: releaseHeldLockForAbort() only called for timeout abort
    if (isTimeout) sessionLockController.releaseHeldLockForAbort().catch(...)
};

releaseHeldLockForAbort() is guarded by if (isTimeout). When user aborts manually, isTimeout=false, so the lock is never released.

Failure Chain

1. User manual abort → abortRun(isTimeout=false) → releaseHeldLockForAbort() SKIPPED → lock remains held
2. Run cleanup → acquireForCleanup() → acquireCleanupLock() → takeHeldLockAfterRetainedIdle() fails (lock in use)
3. Falls back to acquireLock() → lock already held → waits 60s → SessionWriteLockTimeoutError
4. cleanupEmbeddedAttemptResources() never reached → lock NEVER released
5. Only freed by watchdog (maxHoldMs, default 300s) or gateway restart

Relevant Logs

Suggested Fix

- if (isTimeout) sessionLockController.releaseHeldLockForAbort().catch(...);
+ sessionLockController.releaseHeldLockForAbort().catch((err) => {
+     log.warn("failed to release session lock on abort: runId=" + params.runId + " " + String(err));
+ });

Environment

• OpenClaw version: 2026.5.27
• Node: v22.22.1
• OS: Linux (Ubuntu 7.0.0-15-generic)
• Provider: DeepSeek (deepseek-v4-pro)