Lynx Coder

A collection of Lynx AI coding agents, skills, and actions for the Agent Smith toolkit — provides a multi-agent coding team with coordinator, planner, reviewer, and specialist agents.

✨ Features

• 🤖 16 specialized agents — Coordinator, coder, planner, reviewer, doc writer, tester, searcher, manager, and worker agents
• 🧠 Coordinator pattern — lx.yml delegates tasks to specialized agents via the run-agent tool
• 🤝 Collaborator pattern — lx-colab.yml and lx-ts-colab.yml use run-collaborator to fork independent agent instances
• 🏗️ Worker pattern — wagent.yml delegates long-running tasks to worker agents (worker-shell.yml)
• 📚 3 reusable skills — create-or-edit-code, task-planning, and task-success-evaluation for structured workflows
• 🔧 Custom actions — npm command execution with path authorization and shell sanitization
• 📝 Context fragments — Dynamic context injection via {file:...} syntax for workspace awareness

📦 Installation

npm install -g lynx-coder

🚀 Quick Start

Using the Coordinator Agent

The easiest way to use Lynx Coder is through the coordinator agent lx. It orchestrates a team of specialized agents to handle complex development tasks:

# Create a REST API with Express and TypeScript
lm lx "Create a REST API with Express and TypeScript" --workspace /workspace/my-project

The coordinator will load skills, delegate planning to lx-planner, assign implementation to lx-coder or lx-ts, request review from lx-review, and report results back.

Using a Specialist Agent Directly

You can also invoke specialist agents directly for focused tasks:

# Create a utility file
lm lx-coder "Create a new file src/utils.ts with utility functions" --workspace /workspace/my-project

# Generate documentation
lm lx-doc "Write API documentation for the project" --workspace /workspace/my-project

# Review code
lm lx-review "Review the code in src/ for potential issues" --workspace /workspace/my-project

# Plan a task
lm lx-planner "Break down building a React app with authentication into steps" --workspace /workspace/my-project

📖 Usage

Available Agents

Agent	Model	Description
lx	qwen35b	Coordinator — orchestrates the coding team
lx-ts	qwen35b	TypeScript coordinator with npm support
lx-coder	qwen80b	General-purpose coding agent (create/edit files)
lx-coder-ts	qwen80b	TypeScript-focused coding agent with npm commands
lx-planner	qwen35b	Task decomposition & planning
lx-review	qwen35b	Adversarial code reviewer (tools: rshell, readfile)
lx-doc	qwen35b	Documentation generation agent
lx-test	qwen35b	Test execution & debugging agent
lx-search	qwen35b	Web search & page reading agent
lx-manager	qwen35b	Task management & success evaluation
lx-project	qwen35b	Project-level coordination
lx-colab	qwen35b	Collaborator mode (run-collaborator)
lx-ts-colab	qwen35b	TypeScript collaborator
lxa	qwen35b	General-purpose agent
wagent	qwen4b	Test agent with worker delegation
worker-shell	qwen4b	Shell operations worker (read, write, search)

Skills

create-or-edit-code

Structured workflow for code implementation:

1. Check for execution plan (call lx-planner if missing)
2. Execute steps one at a time with verification
3. Request code review from lx-review
4. Evaluate success criteria and report

task-planning

Task decomposition workflow:

1. Analyze task and determine goals
2. Call lx-planner for execution plan
3. Define success criteria
4. Write goals to .agents/tasks/[task-name]/state.md

task-success-evaluation

Task completion assessment:

1. Read goals from .agents/tasks/[task-name]/state.md
2. Review the implemented code
3. Evaluate against success criteria
4. Provide evaluation report

Context Fragments

Agents reference these markdown fragments for dynamic context injection:

Fragment	Purpose
workspace.md	Workspace path and Linux environment instructions
ctx-helper-files.md	Project context helper file locations
agents-manager.md	List of available sub-agents (general)
agents-manager-ts.md	List of available sub-agents (TypeScript-focused)

Custom Action: run-npm-command

Execute npm commands safely within the workspace. Pipe characters (|), shell operators (&, ;), and output redirection (2>) are sanitized — this runs npm directly, not through a shell interpreter.

📁 Project Structure

lynx-coder/
├── dist/
│   ├── agents/              # 16 YAML agent definitions
│   │   ├── lx.yml           # Main coordinator (qwen35b)
│   │   ├── lx-ts.yml        # TypeScript coordinator (qwen35b)
│   │   ├── lx-coder.yml     # Coding agent (qwen80b)
│   │   ├── lx-coder-ts.yml  # TypeScript coding agent (qwen80b)
│   │   ├── lx-planner.yml   # Planning agent (qwen35b)
│   │   ├── lx-review.yml    # Code review agent (qwen35b)
│   │   ├── lx-colab.yml     # Collaborator agent (qwen35b)
│   │   ├── lx-ts-colab.yml  # TypeScript collaborator (qwen35b)
│   │   ├── wagent.yml       # Test agent with workers (qwen4b)
│   │   └── ...              # Other agents
│   ├── skills/              # 3 skill definitions
│   │   ├── create-or-edit-code/SKILL.md
│   │   ├── task-planning/SKILL.md
│   │   └── task-success-evaluation/SKILL.md
│   ├── actions/             # Custom tool actions
│   │   └── run-npm-command.js
│   ├── fragments/           # Context helper markdown files
│   │   ├── workspace.md
│   │   ├── ctx-helper-files.md
│   │   ├── agents-manager.md
│   │   └── agents-manager-ts.md
│   └── utils.js             # Path parsing utilities
├── package.json
└── README.md

📝 License

MIT

---

Part of the Agent Smith toolkit — a TypeScript framework for building local-first AI agents.