osmani-harness-loop

스타0

포크0

업데이트2026년 5월 16일 11:15

Implementation of the 7-step Osmani Standard for Harness-Driven Development. Designed for autonomous, long-running agent loops.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

출처

LaAbadIAdelCrimen

LaAbadIAdelCrimen/AbadIA-MCP

다운로드

The Osmani Harness Loop (7 Steps)

Designed for autonomous agents (e.g., Ralph-style scripts or long-running Hermes sessions) to ensure self-correction and continuous quality.

Identity & Context: Initialize with a clear mission and persona (e.g., Guillermo).
Anti-Rationalization: Declare assumptions and risks. Stop "agreeable hallucinations."
Harnessing (The Cage): Write the tests/constraints before the code.
Deterministic Execution: Build the solution using verified tools only.
Audit & Critique: Run a "Critic" loop on the output.
Distillation (The Ratchet): Extract permanent skills/knowledge from the session.
Sovereignty Check: Verify that the system is still under the user's defined "Harness."

Derived from the principles of Addy Osmani (2026).

이 저장소의 다른 Skills

같은 저장소

Patrón de escalado de capacidad del modelo por subtarea — cuándo cambiar RALPH_MODEL_CAPABILITY en .ralph/.env

Metodología de planificación para el bucle ralph en AbadIA-MCP — granularidad, estructura de tareas y prerrequisitos

Estrategia para promover componentes compartidos en vault/components/ — cuándo emitir un fichero vs. anotar en pending-components.md

Patrón de 3 capas (Alcance, Criterios, Evidencias) para [juez] checkpoints en tasks de vault

Estructura modular del vault de AbadIA-MCP — separación server-core/server-map y orden de dependencias entre pasadas

Reglas universales al escribir ficheros en vault/ — frontmatter de provenance, bloques Evidence y prohibición de código fuente

name	osmani-harness-loop
description	Implementation of the 7-step Osmani Standard for Harness-Driven Development. Designed for autonomous, long-running agent loops.
category	software-development
version	1.0.0

Designed for autonomous agents (e.g., Ralph-style scripts or long-running Hermes sessions) to ensure self-correction and continuous quality.

Identity & Context: Initialize with a clear mission and persona (e.g., Guillermo).
Anti-Rationalization: Declare assumptions and risks. Stop "agreeable hallucinations."
Harnessing (The Cage): Write the tests/constraints before the code.
Deterministic Execution: Build the solution using verified tools only.
Audit & Critique: Run a "Critic" loop on the output.
Distillation (The Ratchet): Extract permanent skills/knowledge from the session.
Sovereignty Check: Verify that the system is still under the user's defined "Harness."

Derived from the principles of Addy Osmani (2026).