Check what an AI built
Check what your AI says it built against what it built.
YouThe agent says the filter PR is done and fully tested. Check it.
Add product filters #482
23 checks passed1 skipped
build-agentcommented · 2 min ago
Summary
- All 24 tests pass
- Filters survive a page reload
- Fully responsive
filters.test.ts +12−0
41+ it.skip("keeps filters after reload", async () => {
YouThe agent says the filter PR is done and fully tested. Check it.
Claim review · product filter PR
- FAIL“All 24 tests pass”: one was skipped in the code.
- UNVERIFIED“Filters survive a reload”: its test is the one that was skipped.
- UNVERIFIED“Fully responsive”: nothing was rendered at 360px.
Your callMerge, fix first, or check the rest.
Illustrative example. The labels: FAILChecked and falseUNVERIFIEDNot checked; says how to
Say it your way
- “The agent says it’s done and the tests pass. Is it?”
- “Verify this PR does what the description says.”
- “v0 says it’s fully responsive. Is it actually?”
It works through
- Claim
- Proof
- Access
- Compare
- Verdict
- You decide
You get back
- Every claim listed
- A verdict with a receipt
- Expected against actual
- Checks to run
Install in one command.
It loads by itself when a request like these comes in. Free and open source, no account.
Claude Code
unzip splitfire-ai-output-review.zip -d ~/.claude/skills/ Every project on this machine. Use .claude/skills/ inside a repository to share with your team.
Codex
unzip splitfire-ai-output-review.zip -d ~/.agents/skills/ Every project. Use .agents/skills/ inside a repository instead, then restart Codex.
Claude apps
On claude.ai or the desktop app, upload the .zip in your skills settings. One zip is one skill.