The Frontier SDK turns governance patterns into versioned, executable checks. Plugins make that engine natural to use inside a coding harness without moving the scoring rules into prompts.
SDK
Frontier SDK
A model- and harness-neutral SDK for deterministic repository scoring, Git-bound evidence packets, and optional signed AAR receipts. Versioned code owns the scoring semantics so results remain comparable across model and host updates.
The SDK owns deterministic scoring, evidence commitments, and receipt verification. Each adapter owns discovery, permission-aware bootstrap, and the conventions of its host. A host update can change how the workflow is invoked without silently changing what a score means.
Codex
Frontier Audit
Inspects for the pinned SDK, asks before installing executable code, verifies the approved integrity chain, then runs the SDK-owned audit instead of merely returning commands.
Release candidate · public source · marketplace publication pending
The planned Claude Code adapter will use the same SDK, scorer lock, evidence format, and trust disclosures. No separate prompt-defined scoring semantics and no public marketplace claim yet.
Adapter in preparation · no public marketplace listing yet