Problem explored: Human-written PRs provide no agent reasoning to draw from — yet reviewers still need structure to investigate efficiently. Currently: open diff, get overwhelmed, close diff, come back tomorrow. This demo shows two AI-driven approaches that work without any author cooperation: inferring what decisions were likely made from diff shape, and generating the skeptical questions a thorough reviewer should ask.

Open Human-written
Migrate auth to shared middleware layer
+291 / −56 lines 8 files changed Commit message: "fix auth stuff"
No agent brief available. This PR was written by a human with a single commit message: "fix auth stuff". The AI analyzed the diff shape and reconstructed 4 likely decisions. Verify each is accurate — correcting wrong inferences is faster than explaining decisions from scratch.
Verified: 0 / 4
🔍
Inferred · Architecture
Auth was consolidated from per-route to middleware layer
92% confidence
Detected: authMiddleware newly registered on app + simultaneous removal of requireAuth from 23 route definitions. This pattern indicates middleware consolidation.

If this is accurate: any route not in the public allowlist is now protected by default. New routes are secure unless explicitly excluded.
🔍
Inferred · Config
Public routes defined via explicit allowlist
88% confidence
Detected: new PUBLIC_ROUTES array exported from config/routes.ts, imported and checked inside the middleware. This is an allowlist pattern — routes not in this array require auth.

Question this raises: is /health, /docs, and /auth/register the complete list of public routes? Are there any that were missed?
🔍
Inferred · Cleanup
23 @requireAuth decorators removed across route files
99% confidence
Detected: identical deletion pattern across 6 route files — requireAuth removed from every handler. High-confidence mechanical cleanup. @adminOnly and @roleRequired decorators appear to be intentionally preserved.

Question this raises: were all 23 removals checked against the PUBLIC_ROUTES allowlist to verify nothing became accidentally public?
⚠️
Inferred · Infrastructure
Session storage changed from in-memory Map to Redis
64% confidence · Needs verification
Detected: Map<string, Session> replaced with Redis client in lib/session.ts. This may be intentional infrastructure improvement or could be an unplanned change bundled into this PR. Lower confidence because the commit message ("fix auth stuff") provides no signal.

This needs explicit verification: was this change planned? Is Redis deployed in this environment? What happens if Redis is unavailable at startup?
Open Human-written
Migrate auth to shared middleware layer
+291 / −56 lines 8 files changed Commit message: "fix auth stuff"
config/routes.ts
+12−0
lib/session.ts
+38−14
middleware/auth.ts
+47−0
routes/admin.ts
+2−12
routes/api.ts
+3−10
routes/users.ts
+3−12
server.ts
+8−4
tests/auth.test.ts
+178−4
Where do you start? 8 files. 347 lines. Commit message: "fix auth stuff".

Was the Redis change intentional? Were all decorator removals safe? Is /health?check=1 still reachable?

Switch to "Decision Surface" to see the same PR with AI-inferred structure.