runesleo/claude-code-workflow

QuietHarness：Claude Code、Codex、Cursor 共用的轻量 AI 工作系统

QuietHarness – a lightweight reliability layer for AI‑coding agents

What it is

  • A small, portable set of configuration files and scripts that add safety checks around AI‑coding assistants such as Claude Code, Codex, or Cursor.
  • The core (1 604 bytes) defines a “working boundary”: detect unrelated changes, run tests, require confirmation before irreversible actions (e.g., credential changes, deployments), and keep a verifiable audit trail.
  • Works with any of the three supported agents; you only install the part that matches the client you use.

Why you might want it

  • Prevents the agent from overwriting files it hasn’t inspected.
  • Stops the agent from claiming a task is finished without actually running the project’s test suite.
  • Guarantees a dry‑run preview before any real changes are written to your filesystem.
  • Provides automatic backup/rollback of the agent’s configuration files, so you can revert to the previous state safely.

How it’s organized

Folder Purpose
templates/ The three minimal configuration bundles (shared core, Claude adapter, Cursor rule, Codex adapter).
scripts/ Helper scripts: inventory, install (dry‑run & apply), verification, migration.
examples/first-success/ Step‑by‑step isolated demo that shows a successful “first‑run” with the agent.
examples/leo-system/ Sanitised snapshots of the author’s real‑world task system (JSON examples, docs).
docs/ Architecture, task continuity, session close‑out, daily‑report handling, etc.
tests/ Smoke‑tests that run the installer in a temporary HOME directory.
media/ Diagrams and release graphics.

Typical workflow

  1. Clone the repo and run ./scripts/inventory.sh to see what will be touched.
  2. Create an isolated demo directory (mktemp …).
  3. Run the installer in dry‑run mode for your client, e.g.
    ./scripts/install.sh --dry-run --claude-project "$demo_dir"
  4. Review the preview, then apply with --apply.
  5. Open the chosen agent in $demo_dir and give it a simple instruction like Read TASK.md and complete the task.
  6. The agent should make the required change, run existing tests, and return concrete evidence of success without any extra prompting.
  7. If everything works, repeat the install step without the --project flag to install the core globally (writes to ~/AGENTS.md, ~/.claude/CLAUDE.md, etc.). Backups are created automatically (.bak‑ai‑workflow‑*).

Key safety mechanisms

  • Dry‑run preview – no file is written until you explicitly pass --apply.
  • Backup & rollback – original config files are saved with timestamps; the installer can restore them automatically.
  • Risk‑gated actions – any potentially destructive operation (credential changes, production deployments, publishing) requires an explicit confirmation step from the agent.
  • Isolation – the demo runs completely inside a temporary directory; the installer never contacts the internet or modifies your account settings.

Who it’s for

  • Developers who already rely on Claude Code, Codex, or Cursor for code generation and want a guard‑rail to avoid accidental overwrites or untested commits.
  • Teams switching between these agents and needing a consistent safety policy that can be moved from one tool to another without re‑writing rules.
  • Anyone looking for a reproducible “first‑success” onboarding test to verify that their AI‑assistant behaves reliably in their own codebase.

Limitations

  • It does not provide a task database, scheduler, or team‑wide orchestration platform – those must be built separately.
  • Only Unix‑like shells are supported; there is no PowerShell installer for Windows yet.
  • The core rules are static text files; extending them requires manual editing or building your own templates.
  • The system assumes you have the respective AI‑agent client installed and configured locally.

Installation & verification

# clone and inspect
git clone https://github.com/runesleo/claude-code-workflow.git
cd claude-code-workflow
./scripts/verify.sh   # runs smoke‑tests, checks backups, template size, etc.

After verification, follow the demo steps above or run a global install:

# Claude Code (user‑level)
./scripts/install.sh --dry-run --claude
./scripts/install.sh --apply   --claude

Similar commands exist for Codex (--codex) and Cursor (--cursor).

License

  • MIT License (see LICENSE).

QuietHarness is essentially a configuration‑as‑code package that makes AI‑coding assistants behave more predictably and safely, without requiring any network calls or proprietary APIs. It is a genuine, reusable tool for developers working with LLM‑based code assistants.

Related

  • Project
    Chachamaru127/claude-code-harness

    Claude Code Harness is a Go‑native plugin that orchestrates Claude‑based (or compatible) AI agents through a disciplined loop—plan, work, review, sync, release—while enforcing strong guardrails and human approvals at each step. It supports Claude Code, Codex CLI, Cursor, and Grok, generating specs, running AI‑driven implementations, independently reviewing changes, and packaging verified releases.

  • Project
    huisezhiyin/sdd-riper

    SDD‑RIPER Light is a lightweight, repo‑native framework that adds a control plane around LLM‑based coding agents. It defines a clear loop—restating goals, creating a minimal spec, awaiting human approval, executing, validating with evidence, and syncing results back—so agents can modify code safely, auditable and recoverable. Four modular “skills” (light, strict, codemap, new‑chat‑ready) cover daily work, high‑risk tasks, unfamiliar codebases, and hand‑offs.

  • Project
    shareAI-lab/learn-claude-code

    A 17‑lesson tutorial that teaches you how to build the surrounding “harness” code for an LLM‑based coding agent (Claude). It walks through adding tools, permissions, context management, memory, task scheduling, multi‑agent coordination, and goal‑driven termination, all around a simple agent‑loop that lets the model decide when to call tools.

  • Project
    doccker/cc-use-exp

    A maintainable configuration system that synchronizes coding rules, skills, and workflows across multiple AI coding assistants like Claude Code, Cursor, and Gemini CLI.

  • Dispatch
    Clean Code Cuts Token Use for AI Coding Agents – Findings from a Minimal‑Pair Study

    A study of Claude Code on 33 minimal‑pair tasks found identical pass rates on clean and messy code, but agents used 7‑8% fewer tokens and revisited files 34% less on cleaner repositories, highlighting efficiency gains from code cleanliness.