Keep your linters
Cotterpin runs them as part of its checks. They catch the cheap problems fast.
How it compares
AI code review vs verification is the difference between reading a change and running it. Review tools, linters, CI and free tools such as Backstage all have a place. This page compares the kinds of check a team usually has with what Cotterpin adds, so you can see what each one can and can't catch, and what a subscription buys over the free tools. Compared by category, as of October 2026.
| Linters and type checks | The agent's own tests | CI on the pull request | AI code review bots | Cotterpin | |
|---|---|---|---|---|---|
| When it runs | On save or commit | When the agent runs them | After the push | On the pull request | In the agent session, before the commit |
| Where it runs | Your machine | Your machine | A build server | Usually the vendor's cloud | Your machine |
| Runs your code | No | Yes, the tests the agent chose | Yes, the suite you configured | No — it reads the diff | Yes — a real build and the tests the change reaches |
| Finds code nothing calls | Sometimes, as unused code | No | No | Sometimes, if it notices | Yes — it checks the work is wired up |
| Finds tests that prove nothing | No | No — they're the agent's own | No | Sometimes | Yes — for example tests that mock what they test |
| Attacks the fix (races, retries) | No | Only if the agent wrote that test | Only if a test covers it | No | Yes — an adversarial refuter |
| Stops risky agent actions | No | No | No | No | Yes — destructive ones wait for your approval |
| Your code leaves your machine | No | No | To your build server | Usually, to the vendor | No — only your own model provider sees it |
A general comparison of each kind of tool, not of any one product: individual tools differ, and many teams use several.
Cotterpin runs them as part of its checks. They catch the cheap problems fast.
Cotterpin shifts checking left, before the commit, so what reaches CI and your reviewers has already been verified.
Cotterpin answers the question the others can't: does the work your agent called “done” actually work?
Good free tools already exist, and many teams run them. Here is what each one is for, and what Cotterpin does that they don't. Compared by category, as of October 2026.
| Backstage | pre-commit and linters | Your agent tool's own hooks | Cotterpin | |
|---|---|---|---|---|
| What it is for | A developer portal: a catalog of your software, its owners and its docs | Running the checks you configured when you commit | Running a command of yours before or after an agent's action | Checking that an agent's “done” is true before it reaches a commit |
| Who writes the rules | No rules about agents; plugins add features | You choose and configure each check | You write and maintain every one | It ships with maintained rules and language packs, and your team can add its own |
| Builds and tests what the agent changed | No | Only the checks you configured | Only if you write that hook | Yes — a real build and the tests the change reaches |
| Sends unfinished work back to the agent | No | No — it fails the commit | If your hook does it | Yes, with the evidence and what to fix (Pro) |
| Stops risky agent actions | No | No | If your hook does it | Yes — destructive ones wait for your approval |
| Cost | Free; you run and host it | Free | Included with the agent tool | Watch free; Pro $100 a year; Team $30 per seat per month |
Cotterpin doesn't replace them. It runs your linters as part of its checks, and from 0.3.0 it can feed your own Backstage: cotterpin backstage export writes catalog entities and a TechDocs site from your repository's knowledge files into your repository, so your portal shows them. Nothing is sent to Cotterpin.
You could build some of this yourself with free tools and your own hooks. What a subscription buys is that it acts, and that someone else writes and maintains the rules.
Gates that block a false “done”, work sent back to your agent with evidence and a fix, an adversarial refuter that attacks the fix, isolated tasks that end in a branch, commit and pull request, and receipts bound to the exact inputs they checked. New language packs the day they ship.
Everything in Pro, with single sign-on (Okta, Entra ID, Google Workspace or GitHub) and seat management. Your team's own rules become a team pack in your repository, reviewed in your pull requests: coming in 0.3.0. Priority email support.
Pro costs $100 a year. Each false “done” it stops is time you don't spend finding a broken change after it has been merged. Work it out for your own team on the time and money saved page.
Cotterpin runs alongside your agent. After every session it tells you what it would have caught, and where. It also reports audit findings for your existing code. It never blocks and never changes your code. Install it on as many machines as you like.
You find out, on your own code and your own agent, how often “done” isn't true before you pay anything. If the report stays empty, you've lost nothing. If it doesn't, you have the evidence to decide.
Every install is free, with no card, no sign-up and no time limit. It runs as Watch, reporting what the gates would have enforced, and the free upgrade and vulnerability-fix routes run in full. Pro or Team turns the gates on. New language packs reach Watch 60 days after Pro.
On the free Watch plan, Cotterpin reports what it would have caught, session by session.