qa/evidence/01-login.png / 02-after-login.png / 03-mission.png — screenshots of the live site.
qa/evidence/_*.mjs — repros so anyone can re-run.
Action needed (cluster access required):
Identify which deployment actually serves canvas.flow-master.ai (the live CSP signature doesn't match any nginx.conf in this repo or flowmaster-ops).
Fix proxy_pass to demo from inside that pod.
Verify NetworkPolicy permits egress to demo.
Resolve the /api/v1/auth/* 404s — either the frontend prefix is wrong or demo isn't the auth backend canvas should target.
Real probes against the live `canvas.flow-master.ai`. Findings:
- Every `/api/*` through canvas's nginx returns **502 Bad Gateway** (proxy is broken).
- `demo.flow-master.ai` upstream is alive (200 on root, 401 on EA2 endpoints) — so the proxy target itself is reachable from the public internet.
- Frontend Login.tsx calls auth routes that return **404 even on direct probes to demo**: `/api/v1/auth/dev-login-config`, `/api/v1/auth/microsoft/login`.
- The user's "live console shows non-stop 200 GETs" hypothesis is wrong — they are 502s on `/api/*` calls, not 200s.
- Zero EA2 traffic actually reaches the backend right now.
Artifacts:
- `qa/evidence/REPORT.md` — full writeup with root cause + four cluster-side fixes needed.
- `qa/evidence/endpoint-probe.json` — raw status of each endpoint.
- `qa/evidence/network-report.json` — Playwright network trace.
- `qa/evidence/01-login.png` / `02-after-login.png` / `03-mission.png` — screenshots of the live site.
- `qa/evidence/_*.mjs` — repros so anyone can re-run.
Action needed (cluster access required):
1. Identify which deployment actually serves `canvas.flow-master.ai` (the live CSP signature doesn't match any nginx.conf in this repo or `flowmaster-ops`).
2. Fix `proxy_pass` to demo from inside that pod.
3. Verify NetworkPolicy permits egress to demo.
4. Resolve the `/api/v1/auth/*` 404s — either the frontend prefix is wrong or demo isn't the auth backend canvas should target.
Captured against https://canvas.flow-master.ai with browser-style
probes + Playwright dogfood + raw fetches:
- Every /api/* through canvas's nginx returns 502 (proxy broken).
- demo.flow-master.ai upstream is alive (200/401 direct).
- Frontend calls auth routes that return 404 even direct on demo:
/api/v1/auth/dev-login-config, /api/v1/auth/microsoft/login.
- User's 'GETs all 200' hypothesis was wrong — they're 502s.
- Zero EA2 traffic reaches the backend.
Repros in qa/evidence/_*.mjs. Screenshots + JSON dumps + full
report in qa/evidence/.
shad
merged commit d68cb5b34e into main2026-06-14 21:21:44 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Real probes against the live
canvas.flow-master.ai. Findings:/api/*through canvas's nginx returns 502 Bad Gateway (proxy is broken).demo.flow-master.aiupstream is alive (200 on root, 401 on EA2 endpoints) — so the proxy target itself is reachable from the public internet./api/v1/auth/dev-login-config,/api/v1/auth/microsoft/login./api/*calls, not 200s.Artifacts:
qa/evidence/REPORT.md— full writeup with root cause + four cluster-side fixes needed.qa/evidence/endpoint-probe.json— raw status of each endpoint.qa/evidence/network-report.json— Playwright network trace.qa/evidence/01-login.png/02-after-login.png/03-mission.png— screenshots of the live site.qa/evidence/_*.mjs— repros so anyone can re-run.Action needed (cluster access required):
canvas.flow-master.ai(the live CSP signature doesn't match any nginx.conf in this repo orflowmaster-ops).proxy_passto demo from inside that pod./api/v1/auth/*404s — either the frontend prefix is wrong or demo isn't the auth backend canvas should target.