A specialized skill for implementing structured Human-in-the-Right-Loop (HITRL) workflows that ensure AI agents obtain human approval before and after executing critical tasks.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install draft-agent-loop
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install draft-agent-loop using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Draft Agent Loop (HITRL)?
This skill enables developers to move beyond blind box agent interactions by enforcing a rigorous, phase-based lifecycle. It leverages the Draft CLI to create a persistent, remote Task Journal that acts as the single source of truth for planning, logging, and verification. By using this entry in the library of Openclaw Skills, users can ensure that high-stakes tasks are executed with precise oversight and clear evidence.
The primary value of integrating this with Openclaw Skills is the ability to manage remote agent environments safely. It mandates a headless execution model, preventing unauthorized local filesystem writes and keeping all project metadata within secure Draft pages. This approach fosters transparency and reliability in autonomous agent operations by requiring explicit human gates at every critical juncture.
Draft Agent Loop (HITRL) Use Cases
- Requiring a detailed technical approach review before an agent begins a complex coding task.
- High-stakes deployments where every execution step must be logged and verified by a human supervisor.
- Collaborative troubleshooting where the agent provides real-time evidence of its actions via a remote dashboard.
- Maintaining a durable audit trail of agent activities for compliance without polluting local disk space.
How Draft Agent Loop (HITRL) Works
- Setup: The agent initializes a connection using the Draft CLI and confirms a healthy headless session status.
- Planning: A Task Journal page is created remotely, and the agent submits a detailed implementation plan using a mandatory template for user approval.
- Approval Gate: The agent enters a blocking state and waits for a specific approval trigger like APPROVED or LGTM in the chat before moving forward.
- Execution and Logging: The agent performs the authorized tasks, appending non-sensitive execution logs to the Task Journal to maintain transparency.
- Verification: Final results and artifacts are summarized on the Draft page, and the agent waits for a final sign-off from the user.
- Archiving: The workflow is finalized with a summary of outcomes and key decisions stored permanently in the remote Task Journal.
Draft Agent Loop (HITRL) Setup
Ensure you have the draft-cli installed and configured as a prerequisite for these Openclaw Skills.
npm install -g @innosage/draft-cli
To initiate the HITRL lifecycle, the agent must start the daemon and verify the runtime environment:
draft start-server
draft status --json
Configuration can be further customized using the GLOBAL_INVITE_CODE environment variable to manage page publishing permissions.
Draft Agent Loop (HITRL) Data Schema & Taxonomy
The skill organizes its data within remote Draft pages to ensure a zero-footprint local workflow. The schema includes:
| Component | Description | Format |
|---|---|---|
| Task Journal | The primary container for the workflow, identified by a unique Page ID. | Draft Page |
| Implementation Plan | Detailed context, goals, and acceptance criteria. | Markdown Header |
| Execution Log | Timestamped actions, status indicators, and non-sensitive evidence. | Appended List |
| Final Summary | Outcome description, key decisions, and links to external artifacts. | Markdown Summary |
Draft Agent Loop (HITRL) Advanced Features
- Iteration Loop Support: Seamlessly handles feedback by appending new iteration plans to the existing Task Journal without losing context.
- Headless Runtime Enforcement: Specifically designed for remote Openclaw Skills environments using the v2 runtime to ensure isolation.
- Non-Sensitive Logging: Automatically filters out credentials and PII from execution logs to maintain security best practices.
- Blocking Handoff Logic: Prevents the agent from proceeding through the lifecycle until a human provides explicit chat-based authorization.
SKILL.md
Loading
Related Openclaw Skills
Zeelin Script Gen Skill V4
A professional AI-driven skill that converts text documents into structured film scripts, character profiles, and scene-by-scene storyboards.
Xiangyun Table OCR Agent
An intelligent OCR agent that converts images, scans, and PDFs into structured data formats like Excel, Word, and Markdown with high-fidelity layout analysis.
Web Access Skill
A comprehensive networking skill that enables AI agents to perform advanced search, dynamic web scraping, and authenticated browser automation.
Web Presentation Pattern Selector
An AI-powered architectural tool for selecting and implementing optimal web presentation patterns like MVC, Front Controller, and Template View.
GCP Global Search
A specialized technical retrieval skill for querying real-time Google Cloud product data, documentation, and pricing directly from official sources.
Free Image Editing — Edit Images and Export Videos
An AI-powered image processing skill that allows users to edit images and export them as professional videos using natural language commands.
Featured*
