Omp TypeSafe — npmjs.com

Run omp plugin install github:siddicky/omp-typesafe and export TYPESAFE_API_KEY. From a clone, use git clone, bun install --production, and omp plugin link "$PWD". Newly added extension modules need a full omp restart; /reload-plugins is not enough.

The extension still loads, logs a warning, shows a notice at session start, and stays inactive. It never wedges the agent, so a missing key degrades to no review rather than blocking work.

Adversarial, the default, argues the other side that the last action or claim is wrong, unverified, or incomplete. Advisory mirrors omp's helpful-reviewer framing: worth-checking steps, simpler alternatives, missed updates, ambiguities, and unconsidered requirements.

No. Review notes are advice and never block a tool call. Two other pieces can intervene: the ambiguity gate can refuse a plan submission in plan mode until the user has been asked, and the optional stop gate can ask the agent to keep going.

Everything the reviewer judges is sent to TypeSafe's API, including parts of your transcript and diffs. The documentation advises reading its "What leaves your machine" section before enabling it on a sensitive repository.

Tool results for edit, write, apply_patch, ast_edit, bash, eval, notebook, debug, and task; assistant messages of at least minMessageChars, 200 by default; and turn-end transcript deltas. Each battery ends with a shared severity score and a defect_class choice.

The extension runs on @typesafe-ai/sdk, about 230 KB, which is what bun install --production installs. A plain bun install also pulls the omp host package and the TypeScript tooling used for type checking: about 1 GB the extension never loads.