[P1][Reliability] Persist task state for crash recovery

[aWN4Y25pa2EK]

Summary

TaskManager is in-memory only. If process crashes mid-execution, task state is lost and tasks get stuck in limbo.

Source

Adversarial Security Assessment

Location

• impl/mvp/src/tasks/manager.ts:78-130
• impl/mvp/src/holons/root.ts:461-492

Problems

• Process crash → task orphaned (not DONE, not FAILED)
• No recovery mechanism on restart
• Manual cleanup required
• Resource leaks (Linear sub-issues created but task "lost")

Impact

• Tasks stuck indefinitely after crash
• Inconsistent state between Linear and HCA
• Manual intervention required
• Resource leaks

Severity

High | Likelihood: High (processes crash: OOM, deployments, exceptions)

Recommended Fix

1. Persist task state to Redis/database
2. Implement recovery on startup
3. Add TTL to orphaned tasks
4. Log state transitions for debugging

class PersistentTaskManager extends TaskManager {
  constructor(private redis: Redis) { super(); }
  
  async createTask(opts: CreateTaskOptions): Promise<Task> {
    const task = super.createTask(opts);
    await this.redis.hset(`task:${task.id}`, task);
    return task;
  }
  
  async recoverOrphanedTasks(): Promise<void> {
    const keys = await this.redis.keys('task:*');
    for (const key of keys) {
      const task = await this.redis.hgetall(key);
      if (task.state.status === 'PLANNING' || task.state.status === 'WORKING') {
        // Recover or fail orphaned tasks
        await this.failTask(task.id, 'Recovered after crash');
      }
    }
  }
}

---

Generated by HCA Architecture Assessment