Hanesu is an innovative experimental workflow layer designed specifically for AI coding agents. This framework integrates principles from Harness Engineering to enhance productivity in repository-level software development. Rather than relying on lengthy prompts for coding tasks, Hanesu organizes the workflow into structured task files, clearly defined phases, role handoffs, quality gates, and progress artifacts.

Key Features

• Structured Interaction: Hanesu provides a visible process for agents, promoting efficiency and predictability throughout the development cycle.
• Compatibility: It complements existing tools such as OpenCode, Codex, and Claude Code, providing a controlled environment for these agents to operate within a set framework.
• Focused Application: While Hanesu may not be necessary for straightforward edits, it excels in handling complex tasks that require thoughtful deliberation, such as ambiguous, risky, or multi-step projects where logical progression and verification are crucial.

Workflow Principles

Hanesu operates on several guiding principles that prioritize effective software development:

• Defined Workflow: Agents follow a structured path rather than engaging in extensive conversations for each task.
• Contextual Awareness: A search for relevant context precedes code edits or interactions with source files.
• Documentation of Decisions: Critical choices are recorded as artifacts to maintain clarity and prevent essential information from being lost in conversation history.
• Human Oversight: Gates are implemented where decision-making is significant, ensuring a layer of review for high-risk tasks.

How It Works

Workflow in Hanesu is designed to progress through defined stages, maintaining clarity and minimizing errors. For each task, the pipeline is set by type, ensuring a consistent approach:

{
  "id": "f1",
  "type": "feature",
  "status": "running",
  "phase": "search"
}

This model emphasizes a single task focus to facilitate better context and more deliberate handoffs across roles:

• First Phase - Context Scout: Before any coding occurs, a context scout evaluates the codebase, streamlining efforts by providing relevant file candidates and minimizing unnecessary exploration.

Daily Workflow Steps

1. Task Definition: Register tasks within the .hanesu/feature.json file, outlining the specific requirements for each coding effort.
2. Implementation Request: Direct the AI to execute specific tasks using concise commands such as Implement f1 using Hanesu. This triggers the agent to follow the specified workflow, ensuring adherence to the project’s structure.
3. Progress Tracking: Each phase is monitored and documented in files like current.json to guarantee clear visibility into the project's status.
4. Artifact Review: Outputs from various phases are documented for review, helping the team assess progress and outcomes efficiently.
5. Task Completion: Once a task is marked as done, the orchestrator automatically identifies and selects the next task from the backlog.

Configuration and Customization

Project customization allows teams to adapt Hanesu to their specific needs:

• Customize files such as .hanesu/config.md for preferences and .hanesu/feature.json for backlog management.
• Define roles and create new tasks through existing architecture in AGENTS.md or ARCHITECTURE.md.

Conclusion

Hanesu promotes a systematic approach to AI-assisted coding and is particularly effective for complex software tasks that benefit from structured workflows and human oversight. Inspired by methodologies like Spec-Driven Development and Test-Driven Development, Hanesu brings a new dimension to collaborative coding practices, making it a valuable addition to any software engineering toolkit.