Sentinel

Showcase

Test the demo PRs.

One Fail with real secret hits, two AI, two human, one mixed. AI advisories can fire on a Pass — they never stop the merge. Only blocking secrets Fail.

Fail · secrets leak01

Secrets.Sandbox intentional leak

Expect Fail · multiple blocking findings (AWS, GitHub PAT, DB URL…)

Public demo PR packed with fake-but-shaped secrets. The gate must Fail — rules block, Muse only explains.

Open on GitHub
AI · Claude agent02

Anthropic skills agent PR

Expect Pass · AI advisories fire (never blocks)

Agent-authored work. Verdict stays Pass — expand AI signals. Advisories are evidence, not a merge stop.

Open on GitHub
AI · Provenance03

Provenance emulator PR

Expect Pass · AI advisories fire (never blocks)

Second agent-leaning PR. Same story: signals show, Pass stays green.

Open on GitHub
Human · curl04

curl pull request

Expect Pass · silent AI panel

Ordinary human-authored change. Secrets gate runs; AI authorship should stay quiet.

Open on GitHub
Human · eslint05

eslint pull request

Expect Pass · silent AI panel

Second human baseline. Use after an AI case to show the contrast.

Open on GitHub
Mixed · AI checkbox06

eslint PR with AI-generated box

Expect Pass · mixed AI disclosure signals

Human PR that still carries AI disclosure. Demo line: signals, not proof — still Pass unless secrets hit.

Open on GitHub

Live scan

Pick a case to test

Start with the Fail case to see blocking secrets, then AI/human for Pass contrast.

Select Fail, AI, human, or mixed above to start a scan.