Auto-Coder Skill

A framework for orchestrating autonomous feature development across multiple Claude Code sessions.

Overview

Auto-coder operates in two phases:

1. Initializer Phase (Session 1): Analyzes a project specification and generates a comprehensive feature_list.json with test cases ordered by priority and dependency
2. Coding Phase (Sessions 2+): Implements features one-by-one, running tests, making git commits, and tracking progress until all features are complete

State Files

All auto-coder state is persisted in the .auto-coder/ directory:

See FEATURE_SCHEMA.md for the complete JSON schema.

Locating Skill Files

To quickly locate the auto-coder skill instruction files (CODER.md, INITIALIZER.md, etc.), use the skill-file-locator.py script:

python ~/.claude/plugins/cache/lynyx-claude/lynyx-agent-kit/<version>/skills/auto-coder/scripts/skill-file-locator.py

The script outputs the full path to the skill directory and a tree view of all skill files:

~/.claude/plugins/cache/lynyx-claude/lynyx-agent-kit/1.2.1/skills/auto-coder
├── scripts
│   ├── continue.sh
│   └── skill-file-locator.py
├── CODER.md
├── FEATURE_SCHEMA.md
├── INITIALIZER.md
└── SKILL.md

Usage:

• Default skill: python skill-file-locator.py (locates auto-coder skill)
• Custom skill: python skill-file-locator.py <skill-name>

Use this script at the start of each coding session to locate and read the appropriate instruction files (CODER.md for coding phase, INITIALIZER.md for initialization phase).

Phases

Phase 1: Initialization

Run with /lynyx-agent-kit:auto-coder init [spec_file]

For detailed instructions, see INITIALIZER.md.

Summary:

• Read and analyze the project specification
• Generate feature_list.json with features ordered by critical path
• Auto-generate project prefix for task IDs (present to user for approval)
• Initialize project structure and git repository
• Create initial commit
• Rename session: auto-coder: initialize {PROJECT_NAME}

Phase 2: Coding

Run with /lynyx-agent-kit:auto-coder code

For detailed instructions, see CODER.md.

Summary:

• Orient to project state (read files, git log)
• Run regression tests on high-priority passing features
• Select next incomplete feature
• Implement and test the feature
• Update feature_list.json and commit
• Rename session: auto-coder: {PROJECT_NAME} | {TASK_ID}

Session Management

Fresh Sessions vs Resume

• Fresh sessions: Recommended for starting work on each new feature
• Resume sessions: Use claude --resume <name> for interrupted work mid-feature

Session Naming

Sessions are automatically named for easy identification:

• Initializer: auto-coder: initialize {PROJECT_NAME}
• Coder: auto-coder: {PROJECT_NAME} | {TASK_ID}

Pause/Resume

1. To pause: Exit session with Ctrl+C or Ctrl+D (context preserved)
2. To resume: claude --resume "auto-coder: {PROJECT_NAME} | {TASK_ID}"
3. To start fresh: claude -p "/lynyx-agent-kit:auto-coder code" (new session)

Auto-Continuation

For fully autonomous operation, use a shell loop:

while true; do claude -p "/lynyx-agent-kit:auto-coder code" || break; sleep 3; done

Or use the helper script:

./plugins/lynyx-agent-kit/skills/auto-coder/scripts/continue.sh

Security Guidance

The following commands are recommended for use during implementation:

Allowed:

• File operations: ls, cat, head, tail, wc, grep
• Runtime: npm, node, bun, python, pytest
• Version control: git
• Process management: ps, lsof, sleep, pkill (dev processes only)

Avoid:

• System modification commands
• Network commands without clear purpose
• Commands affecting files outside the project directory