Hepha is an autonomous agent skill that executes iterative delivery loops using a plan-execute-check-review-commit workflow for unattended coding tasks.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install hepha-skill
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 hepha-skill 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 Hepha?
Hepha is a specialized Openclaw Skills agent designed for high-autonomy software development and continuous delivery. It transforms complex requirements into a series of small, shippable sub-tasks, managing the entire lifecycle from planning to git commits without requiring constant user intervention. By leveraging a strict loop protocol, Hepha ensures that every code change is backed by rigorous engineering checks and browser-based validation.
This skill is particularly valuable for developers who want to delegate small-step implementations and technical research to an AI agent that maintains high quality standards. Hepha operates under non-negotiable rules, ensuring no code is committed until it passes both automated tests and user-perspective reviews, making it a robust addition to any Openclaw Skills collection.
Hepha Use Cases
- Executing unattended autonomous loops for complex feature implementation.
- Performing automated refactoring with continuous validation and small-step commits.
- Conducting technical research and documenting architectural decisions before implementation.
- Running browser-based UI testing and interaction validation using MCP or Playwright.
- Maintaining a structured development backlog and decision log automatically within a project.
How Hepha Works
- Plan: The agent analyzes requirements to decompose them into a task graph stored in .autopilot/backlog.md with defined acceptance criteria and risk levels.
- Research: If the task involves new technology or architectural changes, the agent gathers evidence and documents tradeoffs in .autopilot/decision-log.md.
- Execute: The agent implements the specific sub-task with minimal code changes, focusing on keeping functions small and reusable.
- Check: All relevant project quality gates, such as linting, unit tests, and type checking, are executed to ensure engineering quality.
- Review: For UI or interaction changes, the agent uses browser tools to verify page loads, interaction paths, and visual regressions.
- Commit: A conventional git commit is created only after all checks pass, and the task status is updated in the backlog.
Hepha Setup
To start using this Openclaw Skills capability, provide the agent with a clear requirement and a starter prompt. The agent will handle the directory setup automatically.
# Starter Prompt Example
1. Enable hepha mode.
2. Run loop: plan -> execute -> check -> review -> commit.
3. Perform web research before technical choices.
4. Requirement: <Your Requirement Here>
The agent will automatically create a .autopilot/ directory in your project root and copy necessary templates for backlog, progress, and decision logging.
Hepha Data Schema & Taxonomy
Hepha organizes its operational data within the .autopilot/ directory to ensure transparency and traceability during autonomous runs.
| File | Purpose | Key Metadata |
|---|---|---|
backlog.md |
Task management | task ID, state, dependencies, acceptance criteria, risk |
progress.md |
Execution log | selected task, check results, review evidence, commit hashes |
decision-log.md |
Technical research | options compared, evidence links, tradeoffs, final rationale |
Hepha Advanced Features
- Auto-Decomposition: Automatically splits high-level requirements into vertical slices or risk-first task graphs using predefined patterns.
- Intelligent Re-Planning: Triggers an automatic re-evaluation of the task graph if dependencies change or repeated failures occur.
- Browser-Based Validation: Seamlessly integrates with MCP tools and Playwright to provide screenshots and interaction logs as evidence of success.
- Conventional Commit Enforcement: Standardizes all automated git history with clear explanations of purpose and rationale.
- Schema Validation: Continually verifies the integrity of the backlog to prevent circular dependencies or missing acceptance criteria.
SKILL.md
Loading
Related Openclaw Skills
minimax-xlsx
A rigorous spreadsheet production system that leverages Python and .NET tools to generate, validate, and audit high-fidelity Excel workbooks.
China Telecom Mail
An automation-ready email management skill for China Telecom accounts supporting secure POP3 and SMTP protocols.
Bytedance TOS Image Processing
A comprehensive image processing skill for Bytedance TeraObjectStore that automates resizing, format conversion, and watermarking via the official SDK.
Water Safety & Swimming
A comprehensive AI survival skill focused on drowning prevention, water competency assessment, and emergency rescue techniques.
Values Clarification & Purpose
A structured framework for AI agents to help users identify their core values and align their daily actions with their life purpose.
Tool Fluency
A foundational meta-skill for mastering basic hand and power tools to handle any home repair or DIY project with confidence.
Featured*
