Repository navigation
FAQ
No. The public preview is a portable Agent Skill and optional local tools that guide how an existing AI coding host selects relevant context, skills and verification. The host still performs and authorizes execution.
No such general claim is supported by the public release. Early single-task observations cannot establish repeatable, attributable savings without verified activation, matched budgets and independently checked results.
Yes, offline. The public harness repeats three deterministic tasks in fresh worktrees across four iterations and two identical scripted adapters. It checks the repetition and reporting mechanism, not AI inference. See Benchmarks and Evidence.
Yes. Contribute licensed or synthetic tasks, exact acceptance checks, full per-trial results (including failures) and sanitized receipts through Issue #5 or the evaluation issue template.
The portable skill does not. Local commands operate on explicitly supplied files; optional hooks are disabled unless you enable them in the original host. Review the host's own privacy policy independently.
No. Provider-native tools and permissions are authoritative. Metadata discovery is not permission.
Yes, and it is the recommended first experiment. Run the installer's dry-run and review any changes. See Getting Started.
It is a public developer preview under the MIT license. Compatibility varies by host/version; performance and security isolation are not guaranteed. See the roadmap.