Re-run agent tasks across turns until a completion promise is detected.
npx skills add https://github.com/NeverSight/NeverC --skill ralph-neversight
Please help me install this Agent Skill. Skill: ralph Source: https://github.com/NeverSight/NeverC/tree/main/.agents/skills/ralph Command: npx skills add https://github.com/NeverSight/NeverC --skill ralph-neversight
Ralph removes the need for manual multi-turn repetition by repeatedly re-running the same agent task with a fresh context until a verifiable completion signal appears.
Use cases include long-running implementation work that often stops early, iterative refactors that need repeated verify-debug cycles, and TDD workflows where you want the agent to keep going until tests pass and a completion token is emitted.
Run /ralph "Refactor the authentication module using a test-driven workflow, and output <promise>DONE</promise> only after all tests pass" --completion-promise="DONE" and the loop will keep iterating until it sees the completion promise.
High-intent search queries and answers about installing and using this skill.
Automating multi-turn agent loops requires an AfterAgent hook that re-executes the original prompt with fresh context until a verifiable completion promise is detected, eliminating manual repetition.
A completion promise is a specific XML tag or token emitted by the agent to signal that iterative TDD cycles and refactoring tasks are fully verified and finished.
Setting up autonomous agent repetition for Codex involves running provided setup scripting to prepare the environment, as native hooks are not guaranteed on that platform.
Multi-turn agent repetition works natively with Gemini-CLI, Claude Code, and OpenCode, applying platform-aware operation to maintain stable context and auditability across long-running workflows.
Stopping runaway self-correction loops requires using a configurable maximum iteration cap to enforce safety limits, alongside a dedicated cancellation command to manually halt execution.
yellow-seedPort Dockerfile RUN and ENV steps into portable shell scripts for Codex environments.
ZpankzDetect Ralph Loop completion using four signals and activity logs.
softmgCoordinate role-based pair programming sessions with automated testing and verification.
haJ1tConstruct self-contained prompts for AI agents to execute multi-step tasks.
