Shared Task State
One durable ledger for many agents to read, claim, and update.
HZLThe Missing Task Layer for OpenClaw and Multi-Agent Teams
Coordinate many agents through shared, durable task state with built-in human visibility.
Skip to content
The Problem in OpenClaw
OpenClaw-style deployments always start with one main agent, then expand into specialist agents (writer, researcher, engineer, marketer).
As soon as those agents collaborate, coordination pain appears:
- session memory resets break continuity,
- handoffs become prompt-dependent,
- parallel workers can collide on the same work,
- progress tracking fragments across chat threads and notes.
Why Existing Approaches Fail
Common fallbacks create new problems:
- Chat-based coordination is fast but brittle and hard to audit.
- Human PM tools are usually too heavyweight for agent execution loops.
- Bespoke scripts and trackers become long-term maintenance burden.
- Memory or markdown task files can work for small/single-session setups, but they usually break down once work spans sessions or agents:
- no atomic claiming (two agents can take the same task),
- stale copies and merge conflicts in task files,
- no lease/stuck-task recovery model,
- handoffs depend on prompt discipline instead of enforced state transitions,
- weak auditability when debugging "what happened?"
You track your work. HZL lets your agents track and coordinate theirs when state must be durable, shared, and safe under parallel execution.
How HZL Works in Stateless Systems
Every OpenClaw agent run is a fresh session. Any agent can wake first, check its own work, create/assign tasks, and hand off to others.
Lifecycle (for any agent):
- Session starts and reads shared task state (
in_progress,ready,blocked). - Agent resumes owned work or claims next eligible task.
- Agent updates progress, creates follow-on work, or delegates to another agent/project queue.
- Task completion is written durably, and completion signals are queued for delivery.
Next session starts from that same durable state instead of memory.
Scope Boundary
HZL is intentionally narrow:
- It is a shared task ledger and coordination primitive layer.
- It does not orchestrate runtime behavior: OpenClaw and your agents already handle session wakeups, scheduling, triggers, and coordination logic. HZL stays focused on durable state instead of duplicating a control plane.
- It does not plan or decompose work: agents are responsible for strategy, prioritization, and task breakdown. HZL records and enforces task state transitions.
- It does not replace built-in coding-tool todo lists for quick local loops: those are optimized for single-session execution. HZL is for work that must survive session resets and coordinate across multiple agents.
Use each system where it is strongest: orchestrator for runtime control, agents for planning, and HZL for durable shared task state across sessions.