oracle

name	oracle
description	Get a second opinion by bundling a prompt + a curated file set, then asking a strong model from a different family through a separate Pi invocation.

Oracle

Use this skill when you want a second opinion from a capable model in a different family than the current session.

The oracle workflow bundles selected files into a standalone prompt, selects an alternate model with pi --list-models, and sends the bundle to pi -p --no-tools. It is read-only by default because the oracle receives the selected context directly and does not get tools.

Model selection

The ./scripts/oracle wrapper chooses the strongest known model it can find in the first usable family from this order:

Current family	Oracle preference
OpenAI / Codex	Claude / Anthropic → Gemini / Google → OpenAI / Codex
Claude / Anthropic	OpenAI / Codex → Gemini / Google → Claude / Anthropic
Gemini / Google	OpenAI / Codex → Claude / Anthropic → Gemini / Google
Unknown	Claude / Anthropic → OpenAI / Codex → Gemini / Google

It prefers models listed in Pi's enabledModels, then falls back to the best matching model shown by pi --list-models. Override only when needed with --model provider/model.

If the current model is not obvious to the script, pass it explicitly with --current provider/model. Otherwise it falls back to Pi's default model from settings.json.

Workflow

1. Pick the smallest file set that contains the truth. Avoid secrets by default.
2. Preview selected files before sending.
3. Ask the oracle with a standalone prompt that states the task, constraints, and desired output format.
4. Treat the result as advice, not authority. Reason from first principles before applying recommendations.

Commands

Run commands from this skill directory:

# Preview selected files
./scripts/oracle-bundle --dry-run -p "<task>" --file "src/**" --file "!**/*.test.*"

# Preview the full bundle
./scripts/oracle-bundle -p "<task>" --file "src/**" --file "!**/*.test.*"

# Show which oracle model would be selected
./scripts/oracle --list-models --current openai-codex/gpt-5.6-sol

# Ask the automatically selected oracle model
./scripts/oracle --current openai-codex/gpt-5.6-sol \
  -p "<task>" --file "src/**" --file "!**/*.test.*"

# Override the oracle model when the automatic choice is wrong
./scripts/oracle --model anthropic/claude-fable-5 \
  -p "<task>" --file "src/**" --file "!**/*.test.*"

Tips

• Prefer a minimal file set over the whole repo.
• If you need diffs reviewed, paste the diff into the prompt or attach the diff file via --file.
• Make the prompt completely standalone: include intent, goals, constraints, error text, and whether you want a plan, review, pros/cons, or patch guidance.
• Never include secrets (.env, tokens, key files).
• For general code review, ask for feedback on correctness, consistency with existing code, simplicity, maintainability, performance, security, and edge cases.
• Oracle can be slow while it reasons. Allow it several minutes to process.