omp packages

@ravshansbox/pi-oracle0.2.1

Second-opinion model extension for pi

GitHub Actions Published 3 days ago MIT ~252 downloads/mo

Untested on omp

This is a pi package. No one has reported whether it works on omp yet. If you try it, add your result below.

Did it work for you?

Result

One report per person. A new report replaces your old one.

Reports 0

No reports yet.

Readme

pi-oracle

Second-opinion model extension for pi.

Install

pi install npm:@ravshansbox/pi-oracle

Add -l to install it in project settings.

Usage

Oracle tool

The LLM can call oracle with a prompt containing its question. Set includeContext: true to also share recent conversation history; it defaults to false. The tool chooses a second model from the current provider:

Current provider Oracle model
anthropic openai-codex/gpt-5.6-sol
openai-codex anthropic/claude-opus-5

The current thinking level is clamped to the Oracle model's capabilities. With includeContext: false, only the tool prompt is sent to the other provider. With it enabled, recent conversation text, tool calls, and tool results are included; hidden thinking, images, and previous Oracle opinions are excluded. Responses are limited to 4,096 output tokens, and nested-call usage is included in Pi's session totals.

Oracle command

After an assistant answers, run:

/oracle

Add an optional review instruction after the command:

/oracle what do you think? answer short

Oracle opens a searchable picker containing the authenticated models available in pi, excluding the model that produced the latest answer. Type to filter by model ID, provider, or display name. Press Tab to cycle through the selected model's supported thinking levels, then press Enter to run the review.

The current Pi thinking level is used by default and clamped to the selected model's capabilities. The selected model is remembered per reviewed model. The next time /oracle reviews that model, the previous Oracle model is preselected.

Oracle honours explicit output constraints from the latest user message and optional command instruction. Requests for a short or brief answer suppress the default detailed sections.

Oracle sends recent conversation text, tool calls, and tool results to the selected model's provider. Hidden thinking is excluded. Images are not forwarded, so image-dependent answers may receive an incomplete review. Nested-call usage appears in the expanded Oracle card but is not included in pi's session totals.

Configuration

Pairings for /oracle are stored as a flat map in <Pi agent directory>/oracle.json (~/.pi/agent/oracle.json by default):

{
  "anthropic/claude-opus-5": "openai-codex/gpt-5.6-sol",
  "openai-codex/gpt-5.6-sol": "anthropic/claude-opus-5"
}

Each direction is independent and updates when a different Oracle model is selected.

Development

npm install
npm run check