Shipmoor is a code-integrity layer for the agent era. It runs after your coding agent finishes and before a human reviews the change, and it catches the failure modes agents introduce: hallucinated dependencies, calls to APIs that don’t exist, handlers that return success without doing the work. It can also check whether the change actually did what it was asked to do.
Analysis runs locally. By default, source code, diffs, and repository content are not uploaded. Shipmoor contacts its services for authentication, entitlements, command allowances, and features you explicitly enable.
The problem Shipmoor solves
A coding agent produces a diff that looks plausible. It lints clean, it type-checks, the tests it wrote pass, and it still did the wrong thing: wired the wrong event, edited a sibling service, or “handled” a case by quietly swallowing it. Reviewing agent output at volume is exhausting, so confident-looking changes get rubber-stamped. The gap between what the agent said it did and what the diff actually does is where defects now hide.
Shipmoor closes that gap at the moment before merge, while the change is still in the developer’s shell.
One binary, authorized to your account
You install a single shipmoor 0.10.0 binary and authorize the device to your account:
- First 30 days: full Pro access, with no card required.
- Free: unlimited Scan, Harness, and Skills, plus five monthly invocations each of Claim Check, Review, Test Evidence, and Blast Radius.
- Pro: $19/month or $190/year, with the four metered command families unlimited plus Pro IDE, advanced configuration, repair, and history capabilities.
- Team / Enterprise: coming soon. Org scope, shared baselines, managed CI.
See Plans & tiers for what each unlocks.
What you can do with Shipmoor
| You want to… | Start here |
|---|---|
| Install, authorize, and Scan an agent’s change | Shipmoor CLI · Scanning your code |
| Check whether the change did what the task asked | Claim Check |
| Run the Shipmoor loop inside Claude, Codex, Cursor, or Aider | Agent Skills |
| Have the agent self-correct as it edits | Agent Harness |
| Wire the harness into Claude Code or Cursor | With Claude Code · With Cursor |
| See findings as diagnostics in your editor | IDE Extension |
| Run the supported self-managed Scan in CI | CI & Automation |
| Manage your license and billing | Account & Billing |
Coming soon: Team and Enterprise managed CI, shared policy, organization controls, and governance capabilities are not currently shipped. Current self-managed CI supports Scan only with
SHIPMOOR_TOKEN.
What makes Shipmoor different
Deterministic evidence decides; an LLM only advises. Where competitors put a probabilistic model in the critical path (“AI reviews your AI”), Shipmoor’s stance is the opposite and is enforced in code: a model can never block your merge. Only falsifiable, deterministic checks can, and only if you opt in. The optional LLM second opinion runs in your own agent (BYO-Judge), so Shipmoor hosts and calls no model.
Honest about what it can’t check. When Shipmoor has no probe for something, it says “couldn’t check” rather than implying a pass. That honesty is the point; it’s what makes the checks it does make trustworthy.
Next steps
- Quickstart: install and run your first scan in five minutes.
- Installation: channels,
PATH, and upgrades. - Core concepts: the scan, the claim check, severity, and tiers.