It was a deterministic verifier for AI coding loops: checkpoint the tree, let the agent edit, diff the findings, roll back on a regression. Development stopped at v0.1.8, a preview that never cleared the bar for real use. It comes back as something else.
Precision on repositories the tool had never seen did not reach the point where an automatic rollback is defensible — a false positive there throws away correct work, which is worse than no guard. Only JavaScript, TypeScript and Python were parsed. The roadmap past that gate was years of solo work, in a field where the agents themselves now ship checkpoints and reviews.
The same author is building getdev.ai, a gallery of shipped work. A code verifier does not serve that product. A way to publish to it from the terminal does.
The binary, the release pipeline and every install channel are kept. When getdev.ai has members asking for it, this repository is revived as the gallery's CLI: getdev ship publishes the repository you are standing in as a ship on your profile — stack read from the manifest, screenshot taken, live URL checked — and an MCP server lets a coding agent do the same at the end of a session.
snap/back, init, doctor, update and the MCP transport carry over. The analysis engine and its rule packs are retired. The source at the pause is tagged v0.1-verifier; the reasoning is in docs/DECISIONS.md §19.
The install URLs are a published contract and keep resolving, so getdev.ai/install.sh, npm i -g getdev, brew install getdev-ai/tap/getdev and the rest still hand you v0.1.8. That build prints its own warning on every run. Do not verify real work with it, do not gate a pipeline on it, and do not run guard unattended. Uninstalling is the reasonable move until the publishing client ships.