🤖 fix: trust only Coder's own 404 when deleting AI Gateway keys

[ThomasK33]

K10a of #156: a gate before activation. When the controller deletes an AI Gateway key, only Coder's own 404 now proves that the key is gone. The controller is still unregistered. K10 (activation) follows on top of this branch.

What changes

1. Only Coder's own 404 counts as gone. The sweep, the retire step and the deletion path (deleteOwnKey and deleteKeys) used to accept any 404 except Route not found.. An HTML 404 from a proxy, or any other 404, counted as "the key is gone". The controller then dropped a retiring key from status, or released the finalizer, while the key could still work. These calls now use isCoderResourceNotFound (from 🤖 fix: check a CoderAIProvider's provider ID before Coder calls, and trust only Coder's own delete 404 #325). It accepts only Coder's Resource not found or you do not have access to this resource answer, which DELETE /api/v2/ai-gateway/keys/{key} sends for a missing key (enterprise/coderd/aigatewaykeys.go:178-180 in v2.37.3). Any other 404 is CoderUnavailable and retries. Route not found. is still the NotSupported stall, as before. coderRouteMissing is gone.
2. The bounded pass names what it left (K9 notes item p). While the control plane is being deleted, a failed pass now logs the IDs of the keys that its own LIST showed and that it did not delete (listedKeyIDs), next to the key IDs from status.
3. A test for K9 notes item (t). After a kill-switch re-arm, a mismatched create answer blocks every later create in that generation, also across a failed LIST and the settle window.

Tests

• TestCoderAIGatewayKeyProxyNotFoundDeletesNothing: an HTML 404 on a sweep DELETE keeps the orphan, and the next reconcile deletes it. An HTML 404 on a retire DELETE keeps the key in retiringKeys, and the next reconcile deletes it.
• TestCoderAIGatewayKeyDeleteRetries has a new "proxy 404" case: the finalizer stays, with CoderUnavailable and a 60 s requeue, and the next reconcile releases it.
• TestCoderAIGatewayKeyDeleteWithControlPlaneGoing: in the failing namespace-deletion case, the release log names an orphan that only the pass's LIST knows.
• TestCoderAIGatewayKeyRearmKeepsAnswerMismatchGate (item t).

Mutations that the tests catch:

• isCoderNotFound (any 404) in deleteOwnKey;
• isCoderNotFound in deleteKeys;
• the release log without the pass's leftover IDs.

Size: 130 lines added and 34 removed, of which about 95 are tests.

Local gates on the head:

• make verify-vendor, make build, make lint, make test, make test-integration.
• go tool govulncheck ./....
• go test -race ./internal/controller/....
• make codegen, make build-installer and make docs-reference leave no drift.

Refs #156

---

Generated with xum • Model: anthropic:claude-opus-5-5 • Thinking: high • Cost: $136.24

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-14T07:39:13.041935Z	6a14d5b	PR opened
🔒 Security Review	✅ Completed 2026-10-14T07:41:46.870224Z	6a14d5b	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[ThomasK33]

@codex security review

[chatgpt-codex-connector]

🛡️ Codex Security Review

Security review completed. No security issues were found in this pull request.

Reviewed commit: 6a14d5bb52

View security finding report

Only the user who started this review can view the report in Codex.

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

• You comment "@codex security review"
• A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.