osmani-harness-loop

スター0

フォーク0

更新日2026年5月16日 11:15

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

インストール

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

ソース

LaAbadIAdelCrimen

LaAbadIAdelCrimen/AbadIA-MCP

ダウンロード

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).

このリポジトリの他の Skills

同じリポジトリ

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).