Shipmoor blog
Notes from the loop between agent and merge.
Engineering posts on the detectors that catch AI failure modes, product notes on the Shipmoor surface, and perspectives on what changes when agents write most of the diff.
Shipmoor 0.8.0: the authoring loop closes Claim Check's last gap, and Test Evidence stops trusting a green checkmark
Shipmoor 0.8.0 ships the Claim Check design end to end: an acceptance set with a real approval step, a fix packet that names the exact test to write, and a tag-bound authoring loop that graduates a judgment call to the deterministic floor for good. Test Evidence joins as a companion producer that checks whether your tests actually passed, whether the suite quietly shrank, and whether a kept test was gutted, before Claim Check ever trusts a bare exit code.
Read post-
Shipmoor 0.7.0: Claim Check gives a binding verdict on whether a change did what the author claimed
Shipmoor 0.7.0 adds Claim Check. You write an acceptance set of atomic obligations, the deterministic floor decides READY versus BLOCKED, and a bring-your-own intent fidelity judge adjudicates the judgment ACs the floor cannot check. A binding verdict and a VSA attestation, local, no source upload. The 0.7.0 release also splits the entitlement model: claim_check and code_review replace the old cli_pro and intent_scan.
Read -
Shipmoor 0.6.0: a code review that runs your agent, not ours
Shipmoor 0.6.0 adds Code Review, an advisory AI review of your change that runs your own coding agent, locally, with one command. It finds the soft problems the deterministic scan was never built to catch, it never blocks your build, and your source never leaves your machine. The scan gate you already trust is unchanged.
Read -
How an AI lab team runs Shipmoor as a CI gate on a brownfield Python monorepo
A field report: an AI lab team adopted the free Shipmoor Community CLI as a diff-scoped CI gate across eight services in a complex Python monorepo, alongside Trivy, with SARIF and stable exit codes, and what running on real agent-written code taught us.
Read -
Shipmoor Community CLI v0.2.1: a scan output you can read in five seconds
v0.2.1 redesigns the scan output around the moment a reviewer reads it. A one-line verdict, a one-line context, findings grouped by file with blockers first, inline evidence and fixes, and a footer that tells you exactly what to type next. Plus a progressive explain view and color that means something.
Read -
After the agent, before the review: where AI code tooling misses the moment
A category argument for pre-merge integrity checks on agent-written code, and the eight-criterion proof-of-usefulness methodology we used to test our rules against three production-shape open-source projects.
Read -
Introducing Shipmoor Community CLI: verify agent work before review
A free, local, offline command-line tool that scans Python, TypeScript/JavaScript, and Go for the defects AI coding agents actually leave behind. Built around the moment between 'the agent finished' and 'I'm asking a human to review this.' Includes live runs against Flask, Zod, and Cobra.
Read -
Why AI code integrity is the new category
Linters catch syntax. SAST catches known CVEs. Neither catches the failure modes specific to AI-generated code, and that gap is where the next category lives.
Read