skills.rest
ivanrvpereira
ivanrvpereiraCOMMUNITY
Explore all skills by this author →
𝕏@ivanrvpereiraGitHub

oracle

Bundle repository files with a prompt for cross-model LLM debugging review.

17|Updated Feb 19, 2026
Search Tags:#code review#refactoring#code debugging#LLM routing#file bundling#repo context#cross-model review
One-click install
npx skills add https://github.com/ivanrvpereira/.agents --skill oracle-ivanrvpereira
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: oracle
Source: https://github.com/ivanrvpereira/.agents/tree/main/skills/oracle
Command: npx skills add https://github.com/ivanrvpereira/.agents --skill oracle-ivanrvpereira

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires claude, codex, git, stat, wc, find, and includes scripts (resource) and references (resource) components.

What problem does it solve?

When one LLM’s answers miss bugs, weak design choices, or risky edge cases, this skill helps you package the most relevant truth and obtain a second opinion from a different model family for more reliable debugging, refactoring guidance, and review quality.

Core Features & Use Cases

  • Second-opinion bundle: Combines a standalone task prompt with a curated set of repo files into a single context payload for cross-model review.
  • Model-family routing: Runs an Opus (Claude) review when under OpenAI-powered sessions, otherwise runs a GPT-5.3-Codex review via the Codex CLI.
  • Safer selection controls: Supports include/exclude file glob patterns, ignores common large/noisy directories (e.g., node_modules, dist, build), enforces a max bundle size, and discourages secrets by default.
  • Use cases: Debugging failing code paths, requesting refactor proposals with design checks, producing pros/cons and maintainability feedback, and reviewing diffs for correctness, simplicity, and security.

Quick Start

Bundle the relevant source files (excluding tests if desired) and ask for thorough review by running the oracle-to-gpt script with a prompt that specifies the desired output format, such as a patch or a pros/cons critique.

Frequently Asked Questions about oracle

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I get a second opinion on code review from a different LLM?▼

You can get a second opinion by bundling your prompt with curated repository files and delegating the analysis to an opposite-family LLM, such as routing from Claude to Codex or vice versa, for cross-model code review.

How does cross-model debugging work for validating refactoring suggestions?▼

Cross-model debugging works by packaging your task prompt and source files into a context payload that an opposite-family LLM analyzes to validate correctness, check edge cases, and review refactoring proposals.

Can I exclude test files and node_modules when bundling repo context for review?▼

Yes, you can exclude test files and node_modules by using include and exclude file glob patterns to curate the exact repository file set sent for cross-model review.

Does this require the Codex CLI to get a GPT code review while using Claude?▼

Yes, executing a GPT-5.3-Codex review requires the Codex CLI when you are in an OpenAI-powered session, while an Opus review is triggered under opposite model-family conditions.

What is the best way to review a diff for security and correctness across different models?▼

The best way to review a diff for security and correctness is to bundle the patch with relevant source files and route it to an opposite-family LLM for cross-model analysis and validation.

Are there file size limits or skip rules for large directories when sending repo context?▼

Yes, the Skill enforces a max bundle size and applies skip rules to automatically ignore common large or noisy directories like dist and build, ensuring only relevant context is sent.

Related Skills

Explore More Software Engineering Skills→
borealBytesborealBytes
7

oracle

Bundle a prompt with selected repository files into a context-aware payload.

Community
Intermediate
juspayjuspay

elegance

Refactor codebases with idiomatic patterns and verify via CI.

Official
Intermediate
MohammedSaudAlsahliMohammedSaudAlsahli

ccg

Orchestrates CodexCLI, AntigravityCLI/GeminiCLI parallel analysis with Claude synthesis for complex tasks.

Community
Advanced
goncalossilvagoncalossilva
24

oracle

Send curated files to an opposite AI model family for code review.

Community
Intermediate