ralph-tasks
This skill should be used when the user asks to "add a task for Ralph", "list Ralph tasks", "remove Ralph task", "setup Ralph tasks", or mentions ralph-tasks, task queue for AI agent, or autonomous task management. Manages a JSON-based task queue for Ralph autonomous loop.
SKILL.md
| Name | ralph-tasks |
| Description | This skill should be used when the user asks to "add a task for Ralph", "list Ralph tasks", "remove Ralph task", "setup Ralph tasks", or mentions ralph-tasks, task queue for AI agent, or autonomous task management. Manages a JSON-based task queue for Ralph autonomous loop. |
name: ralph-tasks description: This skill should be used when the user asks to "add a task for Ralph", "list Ralph tasks", "remove Ralph task", "setup Ralph tasks", or mentions ralph-tasks, task queue for AI agent, or autonomous task management. Manages a JSON-based task queue for Ralph autonomous loop. argument-hint: "<command> [args]"
<objective> Manage a task queue for the Ralph autonomous AI coding loop. Provides commands to add, list, remove tasks, and run a continuous execution loop that processes tasks one by one until completion. </objective><quick_start>
/ralph-tasks setup # Initialize in project
/ralph-tasks add Add authentication # Add task (quotes optional)
/ralph-tasks add Add dark mode -s # Add with context search
/ralph-tasks add Fix bug --search # Same with --search
/ralph-tasks list # List all tasks
/ralph-tasks remove 1 # Remove task #1
/ralph-tasks wake-up # Show loop command
Multi-task: Add multiple tasks in one message - they're processed in parallel. </quick_start>
<commands> | Command | Description | |---------|-------------| | `setup` | Initialize .claude/ralph-tasks/ with all files | | `add <task>` | Add a task (quotes optional) | | `add <task> -s` | Add with codebase context search | | `list` | Show all tasks with status | | `remove <id>` | Remove task by ID | | `wake-up` | Show command to start loop |Aliases: add tasks, add task, list tasks, remove task
</commands>
<file_structure>
{project}/.claude/ralph-tasks/
├── tasks.json # Task queue
├── wake-up.sh # Main loop script
├── prompt.md # Agent instructions
└── progress.txt # Accumulated learnings
</file_structure>
<entry_point>
FIRST ACTION: Load steps/action-init.md
This step parses the command and routes to the appropriate action file. </entry_point>
<action_files>
| Action | File | Purpose |
|---|---|---|
| init | steps/action-init.md | Parse command and route |
| setup | steps/action-setup.md | Initialize ralph-tasks |
| add | steps/action-add.md | Add task to queue |
| add-search | steps/action-add-search.md | Add task with context |
| list | steps/action-list.md | Display tasks |
| remove | steps/action-remove.md | Remove task by ID |
| wake-up | steps/action-wake-up.md | Show loop instructions |
| </action_files> |
<state_variables>
| Variable | Type | Description |
|---|---|---|
{command} | string | Parsed command (setup, add, list, remove, wake-up) |
{task_description} | string | Task description (for add) |
{task_id} | number | Task ID (for remove) |
{search_mode} | boolean | Whether --search/-s flag is set |
{project_path} | string | Current project path |
{tasks_dir} | string | Path to .claude/ralph-tasks/ |
| </state_variables> |
<critical_rules> 🛑 NEVER run wake-up.sh automatically - user must run it themselves 🛑 NEVER modify tasks.json without proper JSON handling (use jq) ✅ ALWAYS use atomic file updates (write to tmp, then mv) ✅ ALWAYS load only the current action file (progressive disclosure) </critical_rules>
<success_criteria> ✅ Commands parsed correctly and routed to action files ✅ Each action executes its specific task ✅ Progressive loading minimizes context usage </success_criteria>