Execute multi-phase tasks through iterative verification and planner approval.
npx skills add https://github.com/alpoxdev/claude-code --skill ralph-alpoxdev
Please help me install this Agent Skill. Skill: ralph Source: https://github.com/alpoxdev/claude-code/tree/main/.claude/skills/ralph Command: npx skills add https://github.com/alpoxdev/claude-code --skill ralph-alpoxdev
The Ralph skill provides an autonomous, infinite-loop approach for solving highly complex, multi-phase tasks by enforcing evidence-based progress checks and requiring planner approval before signaling completion.
To start, invoke Ralph on a new task; the system reads the prompt, creates a session folder, launches Phase 1 agents in parallel, and proceeds through the loop to produce a final <promise> only after all phases succeed.
High-intent search queries and answers about installing and using this skill.
Multi-agent task verification is automated through an evidence-driven, self-improving loop that advances execution from implementation to completion, requiring fresh validation checks and planner approval before reporting success.
Phase-based execution divides complex tasks into four stages: implementation, verification, planner validation, and completion, launching parallel agents and performing fresh iteration checks to ensure evidence-based progress.
Iterative task tracking is set up by invoking the skill on a new task, which creates a session folder tracking progress across TASKS.md, PROCESS.md, VERIFICATION.md, and ITERATION.md files for session persistence and recovery.
Parallel execution supports implementation, design, and documentation tasks simultaneously within multi-agent orchestration, providing synchronized status updates across all active agents during the phase-driven loop.
Fresh validation is enforced by performing new checks during each loop iteration using pre-deploy verification, task list tracking, and planner response evaluation, explicitly preventing the reuse of prior validation results.
An infinite-loop task completion approach should be avoided for simple, single-phase tasks that do not require parallel multi-agent orchestration, automated verification, or explicit planner approval gates before finishing.
MadAppGangCoordinate parallel multi-agent code analyses by sharing claudemem outputs across sessions.
moveonhaCoordinate multi-phase agent loops with iterative validation and planner approval.
ZeliperDelegate task lifecycle management to the task-manager-agent using the Task tool.
GamezarDecompose complex tasks into sequential subtasks orchestrated by specialized AI agents.
