Your AI said it's done.
It's lying. We caught it. π»
nohup ghostverifies every PR against what the ticket actually asked for β and a lying model can't fake a pass.
free forever for open source
Every agent does this. Every single one.
$ β
AI coding agents are confident liars.
Not malicious β they believethe "done." Then you merge it, and prod is on fire at 2am. π»
of AI βdoneβs donβt survive review
illustrative
tests actually ran
very tired you
How the haunting works
Install the app on a repo
One click. No CI YAML, no runner to babysit. The ghost moves in.
Open a PR
The ghost reads the ticket and the diff, then adversarially checks the claim against the actual code.
Get a verdict that gates the merge
A pass/fail comment with scored criteria β and a commit status that can block the merge.
SpecReview β pass
hover / tap to see a fail
SpecReview β fail
- [x]Auth flow implemented per ticket
- [x]Session invalidation on logout
- [ ]Ticket required tests β none ran
Why the ghost is different
It canβt be bribed
The verdict is derived in code from evidence, not trusted from the model. A lying agent canβt fake a pass.
It keeps receipts
Every verdict lands in the ledger β claim, evidence, reasoning, cost. Boo-lean logic you can audit.
It gets smarter
Calibrated over time: βwhen we say FAIL, a revert follows 84% of the time.β (Illustrative.)
Runs on YOUR machines
Agents execute on hardware you control. $0 metered tokens, your keys, your data.
Glass box
Every verdict, every dollar, every reasoning step β visible. No spooky action at a distance.
Words from the haunted
(these are jokes. mostly.)
βIt caught my agent claiming a migration βapplied cleanlyβ against a database that did not exist.β
βI said it was done. It was, in fact, not done. Rude, but fair.β
βIf the ghost had been installed, I would still be merged.β
βNo cap, just commits. Prod hasnβt been on fire in 3 whole weeks.β
βFinally something that reads the ticket. Unlike some of us.β
Pricing
Flat pricing. No per-review tolls β we're not animals.
Free
$0foreverthe ghost works for exposure
- βUnlimited public repos
- β1 private repo
- βVerdict comments + merge gate
- βCasper, the local OSS sibling
Pro
$20/mofor the haunted solo dev
- βUnlimited private repos
- βRuns on your own machines
- βBudget caps per run
- βPriority verdict queue
Team
$40/mohaunt the whole org
- βEverything in Pro
- βOrg-wide dashboard
- βCalibration + ledger analytics
- βPolicy-based merge gates
Ready to stop getting ghosted?
or leave your email and we'll gently haunt you