You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Live testing with a real credential (2026-07-20): the authorizer-probe validation is inverted against the deployed service. A valid key now gets HTTP 400 on the fake-task probe (aws-infra#186 added a task-id format guard after the trick was designed, so __validate__ never reaches the handler) → classified error, not saved. A garbage key gets HTTP 403 (HTTP API lambda-authorizer denials emit 403, not the assumed 401) → classified valid → saved and shown Connected. Every unit test passed because the stub service pinned the assumed semantics.
Approach
Delete the probe hack: validate against the now-deployed GET /solves/whoami (aws-infra#185/#188) — 200 with a submitter = valid (identity captured, lighting "Connected as " at connect time — the designed upgrade path), 401/403 = invalid, network/5xx = unreachable. One server-side change in the fork's connections module covers both the panel and the re-pointed setCloudKey command.
Scope
In: the fork's Company Compute validate step + tests/fixtures; identity captured on connect.
Against a stub whoami: 200 {submitter} → connected with identity persisted in status; 403 and 401 → invalid (nothing saved); network failure/5xx → unreachable (nothing saved).
No request ever targets a fake task id; the whoami path is the only validation call.
Connected status renders "Connected as " immediately after key entry (identity from the validate response).
Existing suites green; fixtures updated to the real gateway semantics (403-on-denial), with a comment citing the live finding.
Source
Found live with the internal test key (maps to raghav-internal). aws-infra#186 (the guard), #185/#188 (whoami). Part of the #159 family — post-ship hotfix.
Important
Problem
Live testing with a real credential (2026-07-20): the authorizer-probe validation is inverted against the deployed service. A valid key now gets HTTP 400 on the fake-task probe (aws-infra#186 added a task-id format guard after the trick was designed, so
__validate__never reaches the handler) → classified error, not saved. A garbage key gets HTTP 403 (HTTP API lambda-authorizer denials emit 403, not the assumed 401) → classified valid → saved and shown Connected. Every unit test passed because the stub service pinned the assumed semantics.Approach
Delete the probe hack: validate against the now-deployed
GET /solves/whoami(aws-infra#185/#188) — 200 with a submitter = valid (identity captured, lighting "Connected as " at connect time — the designed upgrade path), 401/403 = invalid, network/5xx = unreachable. One server-side change in the fork's connections module covers both the panel and the re-pointed setCloudKey command.Scope
Acceptance Criteria
Source
Found live with the internal test key (maps to raghav-internal). aws-infra#186 (the guard), #185/#188 (whoami). Part of the #159 family — post-ship hotfix.