Saltar al contenido
TakoAPI
Volver a skills

monitored-ralph-loop

por endogen

DevOps & Cloud
999Descargas según clawskills.sh
0Vistas
0 me gustaInicia sesión para dar me gusta a esta skill
—Aún sin valoraciones
Tu valoración

Resumen

Generate copy-paste bash scripts for Ralph Wiggum/AI agent loops (Codex, Claude Code, OpenCode, Goose)

Descripción

Qué hace esta skill

Generates copy-paste bash scripts for running AI coding agent loops using Codex, Claude Code, OpenCode, or Goose. Manages iterative planning and building through structured markdown files (PROMPT.md, AGENTS.md, IMPLEMENTATION_PLAN.md) with event-driven notifications when the agent needs human input or hits a blocker.

Combines clean per-iteration sessions, file-based memory, and notification routing in one script so long-running agent jobs don't silently fail or exceed context limits.

Cuándo usarla

  • Running an overnight Codex build session that notifies on completion
  • Generating a planning pass that breaks a feature into atomic tasks before writing code
  • Setting up parallel feature branches with independent agent sessions via git worktrees
  • Routing agent decisions back to a human when the agent hits an ambiguous requirement
  • Automating lint and test backpressure after each iteration so the codebase stays green

Flujo de trabajo de ejemplo

ENTRADA

User asks: Generate a Ralph loop for building a FastAPI CRUD app using Codex, max 20 iterations, PLANNING then BUILDING

AGENTE
  1. 1

    Creates specs/overview.md with goal, tech stack, and success criteria

  2. 2

    Generates AGENTS.md with install, test, lint, and run commands plus a Backpressure section

  3. 3

    Generates PROMPT.md in PLANNING mode instructing the agent to produce IMPLEMENTATION_PLAN.md without writing code

  4. 4

    Generates PROMPT.md variant for BUILDING mode instructing the agent to pick one task per iteration, run backpressure, and commit on success

  5. 5

    Outputs ralph.sh with RALPH_CLI=codex and max iterations set to 20

SALIDA

A ready-to-run project scaffold with ralph.sh, PROMPT.md, AGENTS.md, and specs/ that the user copies into their repo and executes

Requisitos

  • Installed and authenticated CLI for your chosen agent: Codex (requires OpenAI API key), Claude Code (requires Anthropic account), OpenCode, or Goose

Instalación

clawhub install endogen/monitored-ralph-loop

Ejecuta el comando de la CLI o pega el prompt en Claude Code, Cursor, Codex o el agente de programación que uses.

Enlaces