Oracle Skill for Claude Code
A Claude Code plugin that gives Claude access to an "Oracle" - GPT-5.2 with deep reasoning capabilities for strategic technical advice.
What is Oracle?
Oracle is a read-only technical advisor powered by GPT-5.2 with maximum reasoning depth (xhigh). It can:
- Read files and explore your codebase
- Run shell commands for investigation
- Search the web for current information
- Provide strategic advice with structured responses
It cannot edit files or make changes - it's purely advisory, making it safe to consult without risk of unintended modifications.
https://github.com/user-attachments/assets/495851ad-9ddf-4672-93df-b8a37f151517
Installation
Prerequisites
-
Codex CLI installed and authenticated:
npm install -g @openai/codex
codex login
-
Access to GPT-5.2 model in your OpenAI account
Install the Plugin
# Add the marketplace
/plugin marketplace add andreasasprou/oracle-skill
# Install the plugin
/plugin install oracle@oracle-marketplace
# Restart Claude Code to load the skill
Usage
The oracle skill is automatically invoked when you ask Claude for:
- Architecture decisions
- Complex debugging help
- Security analysis
- Trade-off analysis
- Code review opinions
Or explicitly ask: "Ask the oracle about..."
Example Questions
- "Ask the oracle: Given our PostgreSQL + Redis stack, should we add Elasticsearch for full-text search or use pg_trgm? We have 10M records, 100 QPS expected."
- "Oracle, analyze the security implications of this authentication flow"
- "What does the oracle think about migrating from our custom event system to a standard library?"
Background Execution
For complex questions requiring deep reasoning, Claude will automatically run the oracle in the background and poll for results, allowing you to continue working.
Response Format
Oracle responses follow a tiered structure:
Essential (Always Present)
- Bottom Line: 1-2 sentence direct answer
- Action Plan: Numbered concrete next steps
- Effort Estimate: Quick (<1h) | Short (1-4h) | Medium (1-2d) | Large (3d+)
Expanded (When Relevant)
- Reasoning: Why this approach over alternatives
- Trade-offs: What you gain vs sacrifice
- Dependencies: Prerequisites and external factors
Edge Cases (When Applicable)
- Escalation Triggers: When to reconsider
- Alternatives: Backup options
- Gotchas: Common mistakes to avoid
Configuration
The oracle uses these Codex CLI settings:
- Model:
gpt-5.2
- Reasoning effort:
xhigh (maximum depth)
- Sandbox:
read-only (cannot write files)
- File editing: Disabled
- Web search: Enabled
License
MIT
andreasasprou
