A portable AI-coding workflow pack, HarnessFlow, enhances AI coding assistants like Claude Code CLI, Codex CLI, and VS Code + Copilot. It transforms one-shot AI code generation into a structured process that classifies requests, orchestrates parallel analysis agents, challenges proposed solutions, validates results with QA, and learns via persistent repository memory. This aims to deliver more robust, consistent, and context-aware AI-assisted development for engineers and power users.

How It Works

This project is an instruction pack, requiring no runtime or build steps beyond file copying. It intercepts AI prompts, categorizes them into eight types, and initiates a multi-agent workflow. This involves specialized agents for analysis (Focus, Broad, Free), plan synthesis (Senior Engineer), critical review (Devil's Advocate), and validation (QA Engineer). It supports three workflow variants: general (thorough), fast (token-efficient, demonstrated 39-50% token savings), and skill (community-backed). Results are stored in repo_info/ for contextual learning.

Quick Start & Requirements

• Prerequisites: Git, Bash, and installed Claude/Codex CLIs.
• Installation: Clone HarnessFlow, then copy its contents into the target repository at .github/HarnessFlow/ using rsync or cp -r.
• Setup: Execute .github/HarnessFlow/cli_setup.sh (CLIs) or .github/HarnessFlow/setup.sh (VS Code + Copilot) from the target repo root. Setup time is not specified but involves file copying and script execution.
• Initialization: Run the respective CLI command or use Copilot Chat with "Initialize this repo." to populate repo_info/.
• Usage: Submit plain-language requests. Optionally prepend mode: fast or mode: skill, or use templates from request_template/.
• Links: Source repository: https://github.com/HangYu8123/HarnessFlow.git (assumed based on name).

Highlighted Details

• Workflow Modes: Offers general (comprehensive), fast (token-efficient, demonstrated 39-50% token savings), and skill (community-backed) variants.
• Multi-Agent Orchestration: Employs specialized agents (e.g., Senior Engineer, Devil's Advocate, QA Engineer) for systematic analysis, planning, challenging, and validation.
• Persistent Repo Memory: Leverages repo_info/ to store codebase context and interaction history for improved AI learning.
• Platform Integration: Designed for Claude Code CLI, Codex CLI, and VS Code + GitHub Copilot.

Maintenance & Community

The provided README does not detail specific maintenance practices, notable contributors, sponsorships, or community channels (e.g., Discord, Slack).

Licensing & Compatibility

No license information is specified in the README. Consequently, compatibility for commercial use or integration with closed-source projects remains undetermined.

Limitations & Caveats

HarnessFlow is an instruction pack, not a formal application, lacking a package manifest, runtime, or build system. Setup relies on Bash scripts. Subagent functionality depends on underlying CLI tool capabilities and model support. Claude-native skills are exclusive to Claude Code environments. The source repository excludes generated files like .github/ and repo_info/.