Guide · Agent skill
How to use the Human In The Loop agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The human-in-the-loop skill teaches an AI agent when a person needs to make a decision and when the work should simply keep going. It is written for developers running OpenRig multi-agent systems and for teams designing human review queues. After this guide you will know how to install the skill, how to ask for it with natural requests, and what it changes about the way an agent ends a work slice.
Human In The Loopmvschwarz/openrig · Apache-2.0 Download the skill Quick start
- 1Download and unzip. Download the ZIP from this skill's page and unzip it so that SKILL.md sits in a human-in-the-loop folder.
- 2Place the folder. Move it to ~/.claude/skills/ for personal use or to .claude/skills/ inside a project.
- 3Or upload in the apps. In the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.
- 4Confirm you have a daemon rig. Check that you run the OpenRig daemon with its queue, workflow, and Mission Control components, since the skill depends on them.
- 5Ask in plain language. Request a closeout classification, a decision routing, or a queue design in everyday words.
- 6Review the decision item. Check that the agent's human request includes proof, the decision, a recommended default, and outcomes.
What the human-in-the-loop skill makes the agent do
At the end of a work slice, the skill has the agent sort the next step into one of three classes: auto-continue, human gate, or park. Auto-continue means the slice closed cleanly and the next named slice in the plan can start. A human gate is used only for a genuine decision, such as usage limits, provider auth, product-intent ambiguity, or a roadmap tradeoff. Park means the work is intentionally stopped, for example while waiting on something external, and the agent records a reason and a way to resume.
- Auto-continue: mark the slice closed and create the next owner's queue item from the plan.
- Human gate: create a human queue item with proof, the decision text, a recommended default, and the possible outcomes.
- Park: stop with a reason and a resumption path.
- The skill says a human should not be asked for routine status updates. Humans are for decisions, not narration.
Install the skill (ZIP download or Claude apps upload)
Download the ZIP from the skill's page on this directory and unzip it. The skill is delivered as a folder containing SKILL.md. Place the unzipped folder in your personal skills folder at ~/.claude/skills/, or in a project's .claude/skills/ folder if you want it available only inside one project.
If you use the Claude apps instead, open Settings, go to Capabilities, and upload the skill under Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills
Trigger the skill with natural requests
You do not need a special command. The skill's description tells the agent when to use it, so ask in plain language about closeouts, decisions, or queue design. Phrases that match the skill's stated uses include classifying a slice closeout, routing a real decision to a person, and designing a human queue or dashboard surface.
- "Classify this slice closeout as auto-continue, human gate, or park."
- "This needs a real decision from a person. Write the human queue item with proof and a recommended default."
- "Design a dashboard that shows the decision queue first and rig state second."
- "Should this closeout wait for approval, or can the orchestrator move to the next slice?"
What a good human request contains
When the skill routes a decision to a person, it asks for four things so the human can decide without reading the whole history: proof of the situation, the decision to make in plain English, a recommended default, and the outcomes each choice leads to. The skill lists the Mission Control verbs a human can use: approve, deny, route, annotate, hold, drop, and handoff. Approval returns the work to orchestration or the chosen owner, and feedback creates a new durable queue item rather than only changing the source file.
- Proof: what happened and where it can be inspected.
- Decision text: the single choice in plain English.
- Recommended default: what the agent would do if the human agrees.
- Action outcomes: what happens after each verb.
Know when not to use it
The skill is deliberately restrictive. If a closeout is clean and the progress file already names the next safe slice, the default conveyor continues and no human gate should be created. If the next owner is another agent, the skill says to use the queue-handoff skill instead. The skill also states that a primitive that only wakes humans is not trustworthy, because it must also know when not to.
- Clean closeout with a named next slice: continue automatically.
- Status update only: do not escalate.
- Another agent is the next owner: use queue-handoff.
Requirements and safety notes
The skill assumes an OpenRig daemon with its queue, workflow, and Mission Control components. The referenced Mission Control and queue files are not included in this file, so the skill alone will not give you a working queue. The skill file itself does not run scripts and does not use the network. Its guidance depends on a daemon-backed rig, so for simple one-off chat tasks it adds nothing.
- Requires: OpenRig daemon and its queue, workflow, and Mission Control components.
- Runs scripts: no.
- Uses network: no.
- Without a daemon-backed rig, the queue and dashboard guidance has nothing to act on.
What people use it for
- Deciding whether a finished slice should auto-continue to the next named slice or wait for a human.
- Writing a clear human queue item for a usage-limit or provider-auth decision.
- Designing a Mission Control style dashboard where the decision queue is the main view.
- Parking a workstream intentionally while waiting on an external dependency, with a documented resume path.
Frequently asked questions
Is the human in the loop agent skill free?
The skill is published under the Apache-2.0 licence, as stated on this page. Check the licence file in the source repository for the full terms before you reuse it.
How do I install the human-in-the-loop skill?
Download the ZIP from this page and unzip it. Put the folder in ~/.claude/skills/ for personal use, or in .claude/skills/ inside a project. In the Claude apps, upload it under Settings → Capabilities → Skills.
What do I need to run this skill?
You need the OpenRig daemon with its queue, workflow, and Mission Control components. The skill references these, but the referenced files are not included in this file, so the skill alone is not enough.
Does the skill run scripts or use the internet?
According to the skill's listing, it runs no scripts and uses no network. It only gives the agent guidance on classifying closeouts and writing human requests.
Will the agent ask me to approve every finished task?
No. The skill says approval is not required for every clean closeout, and the default conveyor continues unless an explicit human gate is reached. Asking for approval on routine work is one of the failure modes it is meant to prevent.
What are the alternatives if this skill is not right for me?
For handing work from one agent to another, the skill itself points to the queue-handoff skill. For simple one-off chat tasks that do not need queues or a daemon, this skill is not the right fit.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you build or operate multi-agent systems on the OpenRig daemon and want agents to escalate only real decisions with clear context. Pick something else if you want an approval step after every task, need a simple chat helper, or do not run a daemon-backed rig, since the skill depends on components that are not included here.
Download the skill