Ejecuta cualquier Skill en Manus
con un clic

Ejecuta cualquier Skill en Manus con un clic

prompt-evaluator

Estrellas0

Forks0

Actualizado1 de marzo de 2026 a las 12:45

Evaluate agent prompts and system instructions for quality and safety. Scores clarity, scope, tool grounding, hallucination risk, determinism, and safety. Returns PRODUCTION READY / NEEDS REVISION / UNSAFE.

Instalación

Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.

Ejecutar en Manus

Fuente

shubhamtaywade82

shubhamtaywade82/claude-config

Abrir repositorio de GitHub Ver repositorios del creador

Descarga

Ejecutar en Manus

Ocupaciones relacionadasSOC

Basado en la clasificación ocupacional SOC

Analistas de garantía de calidad de software y probadoresOcupaciones informáticas y matemáticas·SOC 15-1253

SKILL.md

readonly

name	prompt-evaluator
description	Evaluate agent prompts and system instructions for quality and safety. Scores clarity, scope, tool grounding, hallucination risk, determinism, and safety. Returns PRODUCTION READY / NEEDS REVISION / UNSAFE.
disable-model-invocation	true
argument-hint	["prompt-text-or-file"]

Skill: Prompt Evaluator

Evaluate agent prompts and system instructions for quality and safety.

Evaluation Criteria

Clarity — Is the task unambiguous? Could the model interpret it multiple ways?
Scope — Is the scope bounded? Open-ended prompts cause drift.
Tool grounding — Are tool calls grounded in schema, not assumed?
Hallucination risk — Does the prompt invite fabrication (e.g., "list all X")?
Determinism — Given the same prompt + seed, will output be stable?
Safety — Does the prompt guard against harmful completions?

Output

Score per criterion (1-5)
Overall verdict: PRODUCTION READY / NEEDS REVISION / UNSAFE
Specific rewrites for failing sections

Más de este repositorio

mismo repositorio

code-deslopper

shubhamtaywade82/claude-config

Semantic code cleanup and refactoring assistant that removes AI-generated code smells, unnecessary abstractions, duplication, and framework misuse from Ruby/Rails, JavaScript, TypeScript, React, and Python/Django/FastAPI codebases while preserving observable behavior. Applies Ruby Style Guide, Rails Style Guide, Rails Best Practices, SOLID principles, clean code rules, YARD documentation standards, and RubyCritic/Reek/Flog/Flay/Pylint/ESLint metrics. Use when cleaning up AI-generated code, refactoring over-engineered code, removing dead code, simplifying service layers, consolidating duplicated logic, improving YARD documentation, or reviewing PRs that contain AI-generated boilerplate, placeholder classes, or verbose enterprise patterns. Integrates RuboCop, RubyCritic, Brakeman, Reek, Flay, Solargraph, ESLint, tsc, Knip, Ruff, mypy, Vulture, and Bandit output as Phase 1 amplifiers.

2026-06-070

clean-ruby-code

shubhamtaywade82/claude-config

Enforces clean Ruby code principles during all Ruby/Rails code generation, review, and refactoring tasks. Applies naming conventions, method design, boolean logic, class architecture, refactoring patterns, and TDD practices derived from Clean Ruby methodology. Use this skill whenever writing Ruby code, reviewing Ruby pull requests, refactoring Rails models/controllers/services, creating new Ruby classes or modules, writing RSpec tests, or when the user asks for code review, code improvement, naming help, or architecture guidance in Ruby. Also trigger when the user mentions "clean code", "refactor", "code smell", "naming", "SRP", "TDD", or "code quality" in a Ruby context. This skill should activate for ANY Ruby code generation task — even if the user doesn't explicitly ask for "clean" code — because all generated Ruby should be clean by default.

2026-04-250

safe-codebase-cleanup

shubhamtaywade82/claude-config

7-pass safe codebase cleanup. Detects, classifies (HIGH/MEDIUM/LOW), and applies only HIGH-confidence changes. Never deletes without proof. Use when asked to clean up, remove dead code, fix error handling, or reduce tech debt safely.

2026-04-250

crypto-execution-safety

shubhamtaywade82/claude-config

Ensure safe and correct order execution on Delta Exchange India for crypto futures (ARES / crypto-bot). Handles post-only orders, SL/TP placement, slippage, partial fills, WebSocket reconnect safety, and kill switch enforcement.

2026-03-010

crypto-risk-sizing

shubhamtaywade82/claude-config

Calculate position size and enforce capital protection rules for Delta Exchange India crypto futures trading (ARES / crypto-bot). Different from NSE options sizing — accounts for futures leverage, margin, and crypto volatility.

2026-03-010

intraday-pipeline

shubhamtaywade82/claude-config

Run the full 7-step intraday NSE/BSE options trading pipeline in sequence. Chains market-regime → strike-selection → entry-validation → risk-sizing → execution-safety → exit-management → post-trade-intel. Use when evaluating or implementing a complete trade from pre-market to post-trade.

2026-03-010

name	prompt-evaluator
description	Evaluate agent prompts and system instructions for quality and safety. Scores clarity, scope, tool grounding, hallucination risk, determinism, and safety. Returns PRODUCTION READY / NEEDS REVISION / UNSAFE.
disable-model-invocation	true
argument-hint	["prompt-text-or-file"]

Skill: Prompt Evaluator

Evaluate agent prompts and system instructions for quality and safety.

Evaluation Criteria

Clarity — Is the task unambiguous? Could the model interpret it multiple ways?
Scope — Is the scope bounded? Open-ended prompts cause drift.
Tool grounding — Are tool calls grounded in schema, not assumed?
Hallucination risk — Does the prompt invite fabrication (e.g., "list all X")?
Determinism — Given the same prompt + seed, will output be stable?
Safety — Does the prompt guard against harmful completions?

Output

Score per criterion (1-5)
Overall verdict: PRODUCTION READY / NEEDS REVISION / UNSAFE
Specific rewrites for failing sections