/install agent-todo
agent-todo Skill
Use this skill as an execution queue, not as a passive reminder list.
核心原则:每个 agent 只维护并消费自己 workspace 下的任务队列;需要跨 agent 分发时,再按 OpenClaw 已注册的 workspace 进行发现和投递。
Core commands
bash ./script.sh add "Publish release" \
--task-type publish \
--source "forum:#19/reply:88" \
--next-action "Push main to GitHub and publish ClawHub version" \
--success-criteria "GitHub and ClawHub are both updated"
bash ./script.sh dispatch "Review release" \
--to-agent reviewer \
--task-type review \
--source "chat:direct" \
--next-action "Review release artifacts" \
--success-criteria "Feedback delivered"
bash ./script.sh run-pending --claim
bash ./script.sh done \x3Cid> --note "what was completed"
bash ./script.sh report \x3Cid>
bash ./script.sh block \x3Cid> --reason "why blocked"
bash ./script.sh setup-heartbeat --write
bash ./script.sh setup-heartbeat --all --write
Workflow
- Add tasks with enough execution context:
task_typenext_actionsuccess_criteriasource
- For composite goals, prefer
planto split them into concrete steps. - During heartbeat, run:
bash ./script.sh run-pending --claim
- If it returns
EXECUTE_NOW, do the task immediately. - Prefer continuing a
runningtask before opening a freshpendingone. - To assign work to another agent, use
dispatch --to-agent \x3Cagent_id>. - After execution:
- success →
done - generate reply text →
report - cannot continue →
block - no longer needed →
cancel
- success →
Storage model
- Current workspace queue:
.agent-todo/tasks.json - Optional local identity:
.agent-todo/local.json - Workspace discovery source:
~/.openclaw/openclaw.json - Heartbeat wiring: managed block in
HEARTBEAT.md
Do not hand-write workspace paths in normal usage. Let the script resolve the current workspace and discover registered workspaces from OpenClaw.
Notes
- Single-workspace mode works out of the box after install.
- Multi-agent routing is opt-in: it only matters when you call
dispatch. setup-heartbeat --all --writeappends or updates a managed block for every discovered workspace instead of overwriting the full file.reportgenerates different output shapes for forum sources and direct chat sources.
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install agent-todo - After installation, invoke the skill by name or use
/agent-todo - Provide required inputs per the skill's parameter spec and get structured output
What is Agent Todo?
Turn follow-up promises into an execution queue for agents. Use when chats or discussions create tasks that should be claimed and executed during heartbeat,... It is an AI Agent Skill for Claude Code / OpenClaw, with 173 downloads so far.
How do I install Agent Todo?
Run "/install agent-todo" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Agent Todo free?
Yes, Agent Todo is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Agent Todo support?
Agent Todo is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Agent Todo?
It is built and maintained by zoujiejun (@zoujiejun); the current version is v1.2.1.