pi-runner

Installation
SKILL.md

Pi Runner

Use shared/references/model-preview.md to choose the branch first. Deterministic checks need no model selection. Actual dispatch uses one concrete approved snapshot; nested calls consume it without re-resolution or extra workers.

Execute prompts through Pi (pi) in non-interactive mode. Strict request controls use RPC with a startup handshake; legacy calls retain print mode. Pin the gateway and model per invocation with --provider and --model, or select a central named seat with --seat. Credentials come from the gateway environment or Pi auth store. An exact model selection does not constrain inference provider fallback; carry that separate policy through --provider-routing.

Native routing

Read shared/references/host-model-execution.md before choosing an external route. If the current host can dispatch the exact selected provider model with the required effort, isolation, and receipt policy in a native subagent or task thread, use that route. A host skill descriptor does not dispatch a native model.

Named seats

Named seats and their exact model IDs live in shared/model-routing.json. --seat reads those pins for a direct selection and labels the envelope. Workflow dispatch uses explicit --provider, --model, effort, and policy from the approved snapshot. It does not resolve a seat pin again. An explicit --model must match the approved plan when one exists. Change maintained pins only in the central file. A missing CLI or key reports status: seat_unavailable; there is no fallback to another seat.

Installs
36
GitHub Stars
2
First Seen
Aug 16, 2026