No security concerns identified.

Best possible solution:

Land the provider-local guard with the regression test, classify the changelog bullet under Unreleased ### Fixes, and require either real behavior proof from a redacted setup or an explicit maintainer proof override before merge.

Do we have a high-confidence way to reproduce the issue?

Yes. Source inspection gives a high-confidence path: with multiple enabled subscriptions, a custom or mocked prompter returning an id outside the list makes current main assign undefined and then read tenantId.

Is this the best way to solve the issue?

Yes for the code fix. The explicit guard is the narrowest maintainable provider-local fix and is safer than falling back to another Azure subscription; the changelog placement and real behavior proof still need cleanup.

Full review comments:

• [P3] Move the fix entry under Fixes — CHANGELOG.md:9
  This PR is a user-facing provider auth fix, but the added changelog line is placed under ### Changes while the active ### Fixes section exists. Move it to ### Fixes so the release notes classify the fix correctly.
  Confidence: 0.82

Overall correctness: patch is correct
Overall confidence: 0.86

What I checked:

• Current main has the unchecked dereference: entraIdAuthMethod.run still assigns selectedSub = subs.find((sub) => sub.id === selectedId)! and immediately reads selectedSub.tenantId, so a stale or custom selected id can become an undefined dereference. (extensions/microsoft-foundry/auth.ts:101, 7468e071ddc3)
• PR patch adds the provider-local guard: The PR head replaces the non-null assertion with a match variable, throws Selected subscription not found: ${selectedId} when absent, and only then assigns selectedSub. (extensions/microsoft-foundry/auth.ts:101, 208c1082edca)
• PR patch adds focused mock regression coverage: The new test mocks multiple enabled Azure subscriptions and makes the prompter return missing-subscription, then expects the descriptive error instead of the old TypeError path. (extensions/microsoft-foundry/index.test.ts:274, 208c1082edca)
• Changelog entry is under Changes: The branch adds the Microsoft Foundry auth fix bullet under active ### Changes, while the active ### Fixes section exists later in the same Unreleased block. (CHANGELOG.md:9, 208c1082edca)
• Real behavior proof gate fails: The exact-head Real behavior proof check reports that external PRs need a PR-body Real behavior proof section with after-fix evidence from a real setup; tests, mocks, lint, typechecks, and CI are supplemental only. (.github:7, 208c1082edca)
• Maintainer comment reports test and build validation only: BradGroux reports adding the changelog attribution and regression coverage, then running the focused Microsoft Foundry test, pnpm build, and pnpm check; this is useful validation but not after-fix real behavior proof. (208c1082edca)

Likely related people:

• haxudev: Auth-path provenance points to the merged Microsoft Foundry provider PR, which introduced Entra ID subscription and resource discovery in this plugin. (role: introduced behavior; confidence: high; commits: a16dd967da51; files: extensions/microsoft-foundry/auth.ts, extensions/microsoft-foundry/onboard.ts, extensions/microsoft-foundry/provider.ts)
• BradGroux: BradGroux force-pushed the prepared head, added the changelog/test follow-up according to the maintainer comment, and is the committer on the current PR head. (role: recent PR maintainer; confidence: medium; commits: 208c1082edca; files: CHANGELOG.md, extensions/microsoft-foundry/auth.ts, extensions/microsoft-foundry/index.test.ts)

Remaining risk / open question:

• No after-fix real OpenClaw/Azure behavior proof is present in the PR body, and the exact-head proof check fails.
• Exact-head CI also has unrelated-looking auto-reply/core failures outside the touched Microsoft Foundry files, so merge readiness still needs normal maintainer check or rerun.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 7468e071ddc3.