Repository navigation
feat(site): add AI Gateway keys management page - #25817
Conversation
61253f1 to
c6f9888
Compare
e929815 to
2da5a54
Compare
c6f9888 to
f6d48a7
Compare
c4c363c to
e953487
Compare
99083fe to
7d516d4
Compare
ac72698 to
fc22aeb
Compare
2c718a0 to
e4bf7e6
Compare
fc22aeb to
ca99dc0
Compare
e4bf7e6 to
abe5a17
Compare
ca99dc0 to
5303d30
Compare
abe5a17 to
a57cc01
Compare
841b471 to
f1cb484
Compare
05c5d96 to
b9eb8cb
Compare
f1cb484 to
6d42bab
Compare
b9eb8cb to
677456f
Compare
6d42bab to
8b22178
Compare
|
/coder-agents-review |
|
Chat: Review in progress | View chat deep-review v0.7.1 | Round 3 | Last posted: Round 3, 12 findings (2 P2, 8 P3, 1 Nit, 1 Note), APPROVE. Review Finding inventoryFindings
Round logRound 1Panel. 1 P2, 5 P3, 1 Note, 1 Nit. Netero: no findings. Reviewed against cfb03f5..590996b. Round 2Panel. CRF-1 through CRF-8 all addressed. 1 P2, 3 P3 new. CRF-9 is fix-chain regression from CRF-7. Reviewed against cfb03f5..b212d0f. Round 3Panel. CRF-9 through CRF-12 all addressed. No new findings. All 12 findings resolved. Reviewed against cfb03f5..eada64a. About deep-reviewCRF = Coder Review Finding (P0-P4, Nit, Note)
|
There was a problem hiding this comment.
Clean, well-structured CRUD page that reuses the right shared components. The page/view/dialog decomposition, query cache invalidation, permission gating, and paywall pattern all follow the established AI providers page conventions. The Storybook coverage is strong: the interaction stories use real assertions on real transitions, not render-only placeholders. The delete confirmation text is a highlight: "Any AI Gateway replica using it will no longer be able to authenticate" tells the operator the consequence, not the mechanism.
"tells the operator the consequence: not what the system does ('deletes the key'), but what breaks in their infrastructure. This is the standard every delete confirmation should meet." (Leorio)
Severity summary: 1 P2, 5 P3, 1 Note, 1 Nit.
The PR description is empty (only the template comment). For a 563-line, 11-file feature addition with CRUD, paywall gating, and permission checks, the reasoning and design choices belong in the PR description.
site/src/modules/dashboard/Navbar/Navbar.tsx:30
Note [CRF-7] canViewAISettings only checks viewAnyAIProvider. A user with viewAIGatewayKeys but not viewAnyAIProvider cannot discover the gateway keys page through the navigation (the page itself works via direct URL). This is unlikely to affect real users since these permissions are typically granted together, but worth noting if the permission model evolves.
(Zoro P3, Pariston Note)
🤖
🤖 This review was automatically generated with Coder Agents.
|
/coder-agents-review |
There was a problem hiding this comment.
All 8 R1 findings verified fixed. The error surfacing, validation schema, key-reveal guard, error/empty state separation, and story coverage are all correct. The PR description is now filled in. Nice turnaround.
"This is the kind of assertion suite that proves the fix: it checks for the presence of what should appear AND the absence of what shouldn't." (Bisky, on the LoadError story)
Severity summary: 1 P2, 3 P3 new this round.
The P2 (CRF-9) is a fix-chain regression from the CRF-7 fix. Expanding canViewAISettings to include viewAIGatewayKeys made the "AI" dropdown link visible to users who lack viewAnyAIProvider, but clicking it lands on the Providers index route, which shows a full-screen permission modal with no escape to the sidebar. The gateway keys page is unreachable through the UI for this user class.
🤖 This review was automatically generated with Coder Agents.
8f195e2 to
eada64a
Compare
|
/coder-agents-review |
There was a problem hiding this comment.
All 12 findings from rounds 1 and 2 verified fixed. No new findings from the R3 panel (Bisky, Hisoka, Mafuuu, Knov). Netero clean.
The R2 fix for CRF-9 (permission-wall dead end) is well-designed: AISettingsIndexPage checks permissions in priority order and redirects gateway-keys-only users with replace. The sidebar correctly gates each link on the individual permission. The CRF-12 simplification (single onOpenChange guard) eliminated the dual-mechanism split cleanly.
"The code presented one face and showed the same face on inspection. Nothing hidden." (Hisoka)
Test density climbed from 18.8% (R1) to 34.6% (R3) across the three rounds. Story coverage now includes happy paths, both error shapes, client-side validation, escape guard, state-reset lifecycle, and error/empty mutual exclusion. The contract chain (permissions, routing, API, dialog state, query cache) is consistent end to end.
🤖 This review was automatically generated with Coder Agents.
jakehwll
left a comment
There was a problem hiding this comment.
State was annoying me as stated but LGTM otherwise 🥳
eada64a to
51867bd
Compare
51867bd to
6462221
Compare
Co-authored-by: Jake Howell <jacob@coder.com>
…Dialog.tsx Co-authored-by: Jake Howell <jacob@coder.com>
…Dialog.tsx Co-authored-by: Jake Howell <jacob@coder.com>
…Dialog.tsx Co-authored-by: Jake Howell <jacob@coder.com>
… updated story books

Adds an
AI Gateway Keyspage underAdmin settings > AI > AI Gateway Keysfor key management of keys used by standalone AI Gateway replicas to authenticate intocoderd.The page is shown to users with
viewAIGatewayKeyspermission and a Premium license with AI Gateway enabled.Adds Storybook coverage.