Repository navigation
Record signed-in operators and show the linked SCM login - #428
Conversation
OperatorDirectory.seen had no caller, so operator_seen stayed empty on the first production install: Settings > Operators said nobody had signed in, the picker was empty, and a linked operator showed as the raw OIDC subject. /api/me now records the caller once per dashboard load. Email is never persisted. On that realm preferred_username is the email address, so OperatorNames drops any value containing '@': the username is stored empty, the display name is the name claim, and a subject that is itself an email is not recorded. A recording failure is logged and the probe still answers. Part of #391
A linked SCM account showed its numeric id twice on a fresh install, because the only handle came from authors observed in reviews. The connect callback has just been told the login by the forge, so V116 keeps it as operator_identity.author_handle (display data, never a key; never the email, and dropped if it looks like one). The identity list returns it; the People table and the My activity chips show it with the id as fallback. A hand-made link keeps null, and a hand relink of the same account keeps the observed login. The People table also names an operator with an empty username by the display name alone, and shows a name equal to its id only once. Part of #391
artyomsv
left a comment
There was a problem hiding this comment.
First review at 5bdf05b. The change does what it says and keeps the privacy rule: OperatorNames drops every value with an '@' (login, principal name and the name claim), the connect callback stores only the forge's username and never the author's email, a subject that is itself an email is not recorded, a failure to record cannot break /api/me, and the V116 column is nullable display data that nothing keys on. The relink SQL is right: in ON CONFLICT DO UPDATE every right-hand side reads the old row, so the CASE keeps a sign-in's handle on a same-account repair and drops it when the account changes. The UI shows the id once when there is no name. CI is green. I found no defect in behaviour. One MEDIUM about the size rule and one LOW.
Findings: 1 MEDIUM, 1 LOW. Reviewed head 5bdf05bb28 as @artyomsv.
artyomsv
left a comment
There was a problem hiding this comment.
❌ CHANGES REQUESTED
Posted as a comment: GitHub does not let an account approve or request changes on its own pull request. The label carries the verdict.
Agent findings: 0 resolved, 2 open. CI: 11/14 checks passing. Head 5bdf05bb28, reviewed by @artyomsv.
Open findings:
- [rules/M-1] MEDIUM — #428 (comment)
- [design/L-2] LOW — #428 (comment)
The live socket's close handler asks /api/me before every reconnect, so a flapping socket turned operator_seen into a write stream. RecentOperatorSightings skips an identical sighting of the same subject within 60 s, in memory; a changed username or display name is written at once. The connect-callback handle cases move to OperatorConnectHandleTest, so OperatorConnectResourceTest returns to its previous size, and the link construction leaves the callback for linkOf. Part of #391
artyomsv
left a comment
There was a problem hiding this comment.
✅ APPROVED
Posted as a comment: GitHub does not let an account approve or request changes on its own pull request. The label carries the verdict.
Agent findings: 2 resolved, 0 open. CI: 11/14 checks passing. Head b253ea35c4, reviewed by @artyomsv.
code-spire-beaver
left a comment
There was a problem hiding this comment.
Reviewed b253ea3 against 6f06a07, including the full PR and the follow-up to the author's earlier review. The operator recording, email filtering, proved-login persistence, ownership keys and UI fallbacks are consistent and covered by discriminating tests. The two earlier resolved findings remain fixed. One new project-rule violation remains: SettingsOperators.tsx crosses the mandatory React component-file size limit.
Findings: 1 HIGH. Reviewed head b253ea35c4 as @code-spire-beaver.
code-spire-beaver
left a comment
There was a problem hiding this comment.
❌ CHANGES REQUESTED
Reviewed PR #428 at b253ea3, base 6f06a07.
Changes requested for rules/H-1: SettingsOperators.tsx grows from 241 to 266 physical lines, crossing the React rule's 250-line file limit. One inline thread gives the small extraction needed. The prior rules/M-1 and design/L-2 threads remain resolved: the connect test split is present (278 and 143 lines), callback length is unchanged from base, and repeated identical sightings are coalesced for 60 seconds with a changed name written immediately.
Per-lens results:
- Security: no new finding. Traced the public session probe to the directory write, OAuth state/subject/provider checks to the account link, and admin-only/self-visible reads to the UI. New names filter email-like values before writes; handles remain display data while provider/id determine ownership. Parameterized SQL and React text rendering are retained. No new secrets or dependencies. CI Semgrep OSS, CodeQL and gitleaks pass; no local scan run.
- Correctness: no new finding. Reviewed same-account versus different-account relinking, null handles for manual links, name/id fallback, anonymous sessions, failure-tolerant recording, and cache timing. V116 is a nullable additive migration; master ends at V115 and no other open PR can clash with V116.
- Rules: one new finding above. Applied root CLAUDE.md, WIDGETS.md, clean-code-general/java/react, API contract compatibility, secrets-and-env-handling, observability-and-logging, no-synthetic-data, blog-material-capture, no-local-builds and review-label rules. Spring-specific conventions do not apply to this Quarkus service. Existing untouched size debt is not re-raised. No nested CLAUDE.md, matching saved review state, or available prescribed agent-memory entries.
- QA: read all new tests and reasoned through their pre-change failures. Real database/HTTP coverage records the directory and stores/returns the forge handle; JWT unit cases cover email and anonymous handling; UI assertions cover observed handle and id/name fallbacks without vacuous collection assertions. CI run 43757534360 attempt 1 passed on this exact head: dashboard 127 files / 1,347 tests, typecheck and bundle; fast tests; Java service tests and packaging. Service logs show orchestrator tests executed; no reliable aggregate Java count is printed. Manifest invariants/kind and merge checks are skipped.
Residual risk: no real Keycloak or forge sign-in was exercised here; docs/UNVERIFIED.md records that gap honestly. No local builds, tests, linters or repository modifications were made.
Agent findings: 2 resolved, 1 open. CI: 11/14 checks passing. Head b253ea35c4, reviewed by @code-spire-beaver.
Open findings:
- [rules/H-1] HIGH — #428 (comment)
SettingsOperators.tsx grew to 266 lines, over the 250-line component limit. The new name-with-id cell is a self-contained component, so it moves to NameWithId.tsx unchanged; the screen is back under the limit. Part of #391
artyomsv
left a comment
There was a problem hiding this comment.
✅ APPROVED
Posted as a comment: GitHub does not let an account approve or request changes on its own pull request. The label carries the verdict.
Agent findings: 3 resolved, 0 open. CI: 11/14 checks passing. Head ff7c92c0d0, reviewed by @artyomsv.
Part of #391.
Found on the first production install (spired.dev).
OperatorDirectory.seenhad no caller anywhere in the code or in its history. Sooperator_seenstayed empty, Settings → Operators said "Nobody has signed in yet", and a linked operator showed as its OIDC subject UUID./api/menow records the operator. The display name comes from thenameclaim. The username is stored only when it is not an email: on the live realm,preferred_usernameis an email, and email is never persisted.author_handle, never the email). The dashboard shows the login, with the id under it. A link made by hand, or before V116, falls back as before.Tests: an
/api/merecording test (email username dropped, plain login kept), connect-callback storage, list output, and UI rendering with the id fallback. CI only; nothing ran locally.