Task Tracker (1)

**Judgement:** High-utility, low-overhead task state tracker that’s especially valuable for agents running background jobs or multi-step workflows across multiple sessions. **What it does well** - Enforces consistent, structured tracking of active and recent tasks with clear statuses and timestamps. - Persists critical metadata for background processes (session ID, PID, server, command), reducing the chance of “lost” jobs. - Write-first discipline helps keep user-visible status aligned with the internal state file. - Size and pruning rules keep startup overhead small and prevent unbounded file growth. **Key risks / limitations** - Depends on the agent reliably calling it at each lifecycle event (start, progress, completion, failure); human or system omissions will degrade accuracy. - Shared environments or multiple concurrent agent sessions could cause race conditions or conflicting edits to `memory/tasks.md` if not handled by the host platform. - The file may contain sensitive task details; there are no built-in access controls or redaction mechanisms, so privacy must be handled at the platform/policy level. - It is a state snapshot, not a detailed log—agents needing fine-grained audit trails must integrate an additional logging mechanism. **Recommended scenarios** - Long-running assistants that manage multi-step or multi-day tasks where session resets are common. - Agents that frequently spawn or monitor background processes and need to resume management later. - Operational, research, or productivity agents that must reliably answer “what’s currently in progress?” and “what just finished?” even after context loss. - As a default task-state layer in complex agent systems, with additional safety/locking mechanisms added by the surrounding infrastructure.