oracle

Consults GPT-5.2 with high reasoning for deep analysis. Triggers: "ask the oracle", "consult oracle", "get a second opinion", "ask GPT", "check with GPT", "what would GPT think", "cross-check my reasoning", "external AI". Use for complex debugging, architectural decisions, security audits, or code review requiring deep reasoning. Not for routine tasks.

$ Instalar

git clone https://github.com/richardgill/nix /tmp/nix && cp -r /tmp/nix/built/ai-agents/claude/skills/oracle ~/.claude/skills/nix

// tip: Run this command in your terminal to install the skill


name: oracle description: | Consults GPT-5.2 with high reasoning for deep analysis. Triggers: "ask the oracle", "consult oracle", "get a second opinion", "ask GPT", "check with GPT", "what would GPT think", "cross-check my reasoning", "external AI". Use for complex debugging, architectural decisions, security audits, or code review requiring deep reasoning. Not for routine tasks.

Oracle

Escalate to GPT-5.2 with high reasoning effort for problems requiring deeper analysis. Use for:

  • Complex debugging with elusive bugs
  • Architectural decisions and tradeoffs
  • Security audits and vulnerability analysis
  • Code review requiring deep reasoning

Main command

Run Codex with GPT-5.2 in non-interactive mode with high reasoning:

codex exec -m gpt-5.2 --sandbox read-only -c model_reasoning_effort='"high"' "$PROMPT"

Where $PROMPT is your analysis request. Codex bundles relevant context from your repo automatically.

Reasoning effort

  • high — deep reasoning, ~3x tokens, best for complex problems (oracle default)
  • medium — balanced (codex default)
  • low — fast, minimal thinking

Options

  • --sandbox read-only — analyze without modifying files (recommended)
  • --sandbox workspace-write — allow file modifications if needed
  • -o /tmp/oracle-response.md — save response to file
  • -C <dir> — run in specific directory

Guidelines

  • Oracle is read-only — analyzes but doesn't modify code
  • Treat outputs as advisory — verify against codebase
  • Use for genuinely complex problems (not routine tasks)
  • High reasoning uses ~3x tokens but thinks deeper

Examples

# Debug intermittent auth issue
codex exec -m gpt-5.2 --sandbox read-only -c model_reasoning_effort='"high"' \
  "The auth flow fails intermittently. Check src/auth/ for race conditions."

# Architectural review
codex exec -m gpt-5.2 --sandbox read-only -c model_reasoning_effort='"high"' \
  "Review src/api/ data flow and suggest improvements for scalability."

# Security audit
codex exec -m gpt-5.2 --sandbox read-only -c model_reasoning_effort='"high"' \
  "Audit src/handlers/ for OWASP top 10 vulnerabilities."

What to ask oracle:

$ARGUMENTS