refactor(ai): first-class Microsoft Foundry v1 protocols

[atulmgupta]

Microsoft Foundry v1 only

The Azure provider now supports only the modern Foundry OpenAI v1 surface. Classic Azure OpenAI deployment routes, inference routes, API-version/flavor/override fields, and model-name heuristics are removed—not hidden.

• New API protocol selector: Auto (default), Chat Completions, or Responses. Explicit selections never fall back. Auto uses one bounded structured-error Chat → Responses fallback.
• Validation and Helix share the same model identity and protocol. Embeddings always use the same v1 endpoint with their separately configured deployment name.
• Provider key remains azure, preserving credentials and feature-provider identity. Existing configurations without api_protocol migrate previously effective deployment names into visible model fields; obsolete fields are dropped on settings save. Modern configs ignore obsolete overrides.
• Resource roots normalize to /openai/v1. Other API paths/query parameters are rejected, not silently rerouted.
• Probe budget remains explicit (1,024 tokens / 30 seconds), caller budgets are honored, both operation errors survive, Responses uses store:false and rejects failed/incomplete/refused/empty results.
• Responses streaming remains buffered completion-to-chunks, clearly documented in UI and docs; Chat Completions uses native SSE.

Verification for e44ebf9

Verified in an isolated worktree because unrelated Alert Packs edits were concurrently in progress in the shared checkout:

• All provider packages + settings validation + Helix dispatch: Go -race PASS; targeted golangci-lint PASS.
• npx tsc --noEmit PASS; 3 frontend test files / 49 tests PASS; targeted ESLint PASS.
• Both settings component audits: 0 violations; both hook paths match backend routes. Panels retained, line counts 469→403 and 584→565.
• Generated-artifact freshness PASS.
• docker compose -p foundryverify build: all five service images Built.
• Regression coverage includes all three protocols, explicit no-fallback errors, arbitrary aliases, native SSE/tool replay, buffered Responses tool replay, migration/credential identity, request budgets and embeddings.

No Azure endpoint credentials were available: this is not live Azure verification. CI is being monitored; no merge requested.

Official references:
https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/responses
https://learn.microsoft.com/en-us/azure/foundry/openai/api-version-lifecycle

[Copilot]

🟡 Changes recommended

Fix retry response handling, deployment selection, fallback error reporting, and outdated package documentation.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Restores Azure Foundry chat routing by distinguishing v1 URLs from classic endpoints and adding fallback behavior.

Changes:

• Restricts v1 detection to /openai/v1 paths.
• Adds classic-route fallback for failed v1 chat requests.
• Updates routing and fallback regression tests.

File summaries

File	Summary
internal/ai/provider/azure/azure.go	Routing and fallback implementation; response handling, deployment selection, fallback errors, and package documentation require updates.
internal/ai/provider/azure/azure_test.go	Covers host-only routing and fallback behavior.

Review details

Suppressed comments (3)

internal/ai/provider/azure/azure.go:185

• resp.Body.Close() is deferred until this Chat call returns, but the fallback starts another request on the same http.Client before that return. With a transport configured with MaxConnsPerHost: 1, the first response keeps the only connection occupied and the fallback can block until the context expires; even without that limit, each in-flight 404 consumes an extra connection. Close the failed response before starting the retry (and keep the defer only for the successful-response path).

		if resp.StatusCode == http.StatusNotFound && a.usesOpenAIV1() {
			if fallback, ferr := a.withoutOpenAIV1().Chat(ctx, req); ferr == nil {

internal/ai/provider/azure/azure.go:185

• This retry does not cover saved v1 configurations where cfg.Deployment is stale: chatEndpoint selects the v1 body model through chatDeployment, which prefers cfg.Deployment over cfg.Model, and this fallback then uses that same deployment in the classic URL. The settings form intentionally merges hidden deployment fields when switching flavors, so both requests can target the old deployment and still return 404. Use a v1-specific identity that prefers req.Model/cfg.Model (with an explicit fallback only when appropriate), while retaining cfg.Deployment for classic OpenAI routing.

		if resp.StatusCode == http.StatusNotFound && a.usesOpenAIV1() {
			if fallback, ferr := a.withoutOpenAIV1().Chat(ctx, req); ferr == nil {

internal/ai/provider/azure/azure.go:188

• When the classic retry also fails, ferr is discarded and the caller receives only the original v1 404. That masks whether the fallback failed because of authentication, API version, or another routing problem; surface the fallback error while retaining the initial error for diagnosis.

			if fallback, ferr := a.withoutOpenAIV1().Chat(ctx, req); ferr == nil {
				return fallback, nil
			}
		}

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

One or more custom setup steps configured for this repository failed during this Copilot code review run:

Verify TypeScript build

Setup steps run before each review. If the review above is missing context, or no review was posted at all, the failing step above may be the cause. See the workflow run for failure details, fix your setup steps configuration, and re-request a review.

Note

You can configure setup steps for Copilot code review separately from Copilot cloud agent with a copilot-code-review.yml file. Read the docs for details.

[Copilot]

🔵 Needs a closer look

Address the two moderate issues in azure.go.

Review details

Suppressed comments (2)

internal/ai/provider/azure/azure.go:429

• The package-level documentation above still says that the v1 surface is detected from a services.ai.azure.com hostname, but this change makes the path the only detection signal. That now contradicts the implementation and the new behavior for host-only Foundry URLs; please update the package comment in the same change so future callers do not rely on the removed auto-detection rule.

// OpenAI-compatible v1 surface. Only the path is authoritative —
// hostname *.services.ai.azure.com also hosts classic
// /openai/deployments/{name}/chat/completions?api-version= which
// Helix used successfully before auto-detect (#118) forced v1.

internal/ai/provider/azure/azure.go:187

• When the classic retry fails, ferr is discarded and the original v1 404 is returned. That hides the actual fallback failure (for example, a 401 or 500), so validation and callers can misclassify a bad key or upstream outage as not_found; return the fallback error (or otherwise preserve it as the primary error) instead of silently ignoring it.

			if fallback, ferr := a.withoutOpenAIV1().Chat(ctx, req); ferr == nil {
				return fallback, nil
			}

• Files reviewed: 2/2 changed files
• Comments generated: 0 new
• Review effort level: Lite

[Copilot]

One or more custom setup steps configured for this repository failed during this Copilot code review run:

Install gopls (Go LSP)

Setup steps run before each review. If the review above is missing context, or no review was posted at all, the failing step above may be the cause. See the workflow run for failure details, fix your setup steps configuration, and re-request a review.

Note

You can configure setup steps for Copilot code review separately from Copilot cloud agent with a copilot-code-review.yml file. Read the docs for details.

[Copilot]

🟡 Changes recommended

Moderate issues remain in model capability detection, validation token limits, and package documentation.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (2)

internal/ai/provider/azure/azure.go:705

• The cloud validation path always sends MaxTokens: 1 (see handleValidateCloud), so this branch emits max_completion_tokens: 1 for GPT-5/o-series requests. The new comment explicitly says Azure rejects an omitted or 1-token reasoning cap, which means Validate connection still fails for the models this change targets; omit the probe cap or normalize that specific probe to a valid completion budget before encoding.

		// Plural tool_calls — what dispatch.go now copies from
		// resp.ToolCalls into the assistant message before
		// appending to history. This is the round-trip path that
		// makes multi-iteration tool dispatching work.
		for _, mc := range m.ToolCalls {
			tc := azureWireToolCall{ID: mc.ID, Type: "function"}
			tc.Function.Name = mc.Name
			tc.Function.Arguments = string(mc.Arguments)
			wm.ToolCalls = append(wm.ToolCalls, tc)
		}

internal/ai/provider/azure/azure.go:429

• The package comment at lines 5-7 still says any services.ai.azure.com hostname auto-selects v1. This function now deliberately ignores the hostname, so the public package documentation contradicts the implementation and can lead future changes back to the removed behavior; update the top-level comment in this change.

	}
	switch a.cfg.Flavor {
	case provider.AzureFlavorFoundry:
		return a.buildURL("embeddings")

• Files reviewed: 7/7 changed files
• Comments generated: 1
• Review effort level: Lite

[Copilot]

One or more custom setup steps configured for this repository failed during this Copilot code review run:

Verify TypeScript build

Setup steps run before each review. If the review above is missing context, or no review was posted at all, the failing step above may be the cause. See the workflow run for failure details, fix your setup steps configuration, and re-request a review.

Note

You can configure setup steps for Copilot code review separately from Copilot cloud agent with a copilot-code-review.yml file. Read the docs for details.

[Copilot]

🟡 Changes recommended

Unresolved critical and moderate findings affect Azure routing, streaming, fallback, token limits, model selection, and completion status.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (3)

internal/ai/provider/azure/azure.go:740

• The adjacent comment says Azure rejects max_completion_tokens=1, but this classic GPT-5 branch only defaults when n <= 0. internal/api/aisettingsvalidate/handler.go sends MaxTokens: 1, so host-only classic GPT-5 validation still emits max_completion_tokens: 1 and can return 400 instead of validating. Apply the same lower-bound handling here or change the probe contract.

		if n <= 0 {
			// gpt-5 / o-series spend hidden reasoning tokens
			// against this cap. Azure 400s with "max_tokens or
			// model output limit was reached" when the field is
			// omitted or set to 1 (the settings probe).
			n = defaultMaxCompletionTokens

internal/ai/provider/azure/azure.go:8

• The package comment now says a gpt-5.6+/gpt-6 model selects the OpenAI v1/Responses surface, but routing is path-based (and the model only changes the token-cap field). A host-only classic GPT-5 deployment follows chat completions, so this documentation contradicts the path-only detection described by the change and can lead maintainers to configure the wrong endpoint.

// A third URL shape — Azure AI Foundry OpenAI v1 — is used when the
// path contains /openai/v1, or when the model is gpt-5.6+ / gpt-6
// (official Responses API). See:
// https://learn.microsoft.com/en-us/azure/foundry/openai/how-to/responses

internal/ai/provider/azure/azure.go:196

• The documented fallback from a v1 404 re-enters classic Chat Completions with the original Helix tool list, but the new Responses implementation explicitly notes that gpt-5.6+ rejects function tools unless reasoning_effort=none; azureChatRequest has no such field or value. For the target gpt-5.6-sol deployment, a failed v1 probe will therefore fail again instead of restoring chat routing. Add the required reasoning setting to this classic fallback or restrict it to models that support the existing Chat Completions payload.

		if resp.StatusCode == http.StatusNotFound && a.usesOpenAIV1() {
			if fallback, ferr := a.withoutOpenAIV1().Chat(ctx, req); ferr == nil {
				return fallback, nil
			}

• Files reviewed: 7/7 changed files
• Comments generated: 5
• Review effort level: Lite

[Copilot]

One or more custom setup steps configured for this repository failed during this Copilot code review run:

Verify TypeScript build

Setup steps run before each review. If the review above is missing context, or no review was posted at all, the failing step above may be the cause. See the workflow run for failure details, fix your setup steps configuration, and re-request a review.

Note

You can configure setup steps for Copilot code review separately from Copilot cloud agent with a copilot-code-review.yml file. Read the docs for details.

[Copilot]

One or more custom setup steps configured for this repository failed during this Copilot code review run:

Verify TypeScript build

Setup steps run before each review. If the review above is missing context, or no review was posted at all, the failing step above may be the cause. See the workflow run for failure details, fix your setup steps configuration, and re-request a review.

Note

You can configure setup steps for Copilot code review separately from Copilot cloud agent with a copilot-code-review.yml file. Read the docs for details.

[Copilot]

Copilot review overview

🟡 Changes recommended

Unresolved critical and moderate issues remain in Azure routing, authentication, fallback behavior, and response handling.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 2 · 5 · 1

Open (8)

• Avoid injecting auth headers on non-v1 URLs · New
• Correct the expected Authorization header in the test · New
• Prevent retries after failed streaming negotiation · New
• Preserve temperature during Responses fallback · New
• On the Foundry flavor, chatDeployment prefers cfg.Deployment over cfg.Model. The settings…
• Because Capabilities advertises Streaming=true, dispatch always selects Stream, but this…
• ProviderConfig keeps the model identifier separate from the optional deployment name…
• Keep explanatory text outside the complete adapter table · New

Resolved since last review (3)

• AzureFlavorFoundry is still defined as the classic model-in-body /chat/completions?api-version=…
• When the Responses API returns status: incomplete with `incomplete_details.reason:…
• ChatRequest.MaxTokens documents zero as the adapter default, while the validation handler…

[Copilot]

Copilot review overview

🔵 Needs a closer look

The mixed Foundry and Alert Packs scope plus unresolved API-contract and localization issues require human review and changes.

Review effort: Lite
Findings: 5 · 8 · 4

Open (17)

• Preserve GitHub Pages base path in documentation links
• Azure embedding model setting is ignored by production RAG
• Azure RAG ignores configured embedding deployment
• Bound upstream response details in client-facing errors
• Wire Azure embedding deployment into the production RAG path
• Apply channel filtering to the API-polling evaluator
• Remove unrelated Alert Packs changes from this refactor
• Rule reuse key omits trigger mode
• Preserve temperature for Azure Chat Completions requests
• Align zero MaxTokens default with shared cost reservation
• Preserve temperature during Responses fallback
• On the Foundry flavor, chatDeployment prefers cfg.Deployment over cfg.Model. The settings…
• Because Capabilities advertises Streaming=true, dispatch always selects Stream, but this…
• Remove unrelated Alert Packs changes or expand PR scope
• Public OpenAPI spec omits Alert Packs endpoints
• Alert Packs changes are out of scope for the Foundry v1 PR
• Correct validation comment for remaining Azure probe fields

[atulmgupta]

Final verification for 6debba4: CI and Frontend quality gates both succeeded. PR checks: 64 passed, 4 skipped by their existing conditions, none failed or pending. All five Docker jobs started after the generated/backend/frontend gates completed; this ordering was checked against job timestamps. Local API/web Docker builds and responsive inline-edit/Helix tests also passed. No merge performed.

[atulmgupta]

Settings readability verified in 82aa2f7. The stacked-card production fix was already present in 31db0e7; this follow-up adds stronger regression coverage rather than changing working layout code.

Tested widths: 320, 390, 768, 1024, 1280, 1440, 1920 and 2560px, each in light/dark themes at 100% and 135% text scale.

Assertions cover all three shortcut descriptions (at least 180px wide), buttons below descriptions with 44px minimum height and unobscured click targets, separate category description/save-hint lines, no overview text spilling outside cards, no horizontal page overflow, and preserved unsaved edits across categories. Individual card screenshots are captured as browser-test artifacts.

32 passed (1.9m)
E2E preview proxy escape check: 0 errors

TypeScript and scoped ESLint passed. SettingsPage and SettingsActionCard audits each reported zero violations; Settings hook routes matched the backend. No production components, sections, or API behavior were changed in this follow-up. CI is being monitored; no merge or deployment performed.

[Copilot]

One or more custom setup steps configured for this repository failed during this Copilot code review run:

Verify TypeScript build

Setup steps run before each review. If the review above is missing context, or no review was posted at all, the failing step above may be the cause. See the workflow run for failure details, fix your setup steps configuration, and re-request a review.

Note

You can configure setup steps for Copilot code review separately from Copilot cloud agent with a copilot-code-review.yml file. Read the docs for details.

[Copilot]

Copilot review overview

🔵 Needs a closer look

The served OpenAPI contract is missing the new Alert Packs and AI draft routes, and the PR combines unrelated Alert Packs work with the Foundry refactor.

Review effort: Lite
Findings: 5 · 8 · 4

Open (17)

• Preserve GitHub Pages base path in documentation links
• Azure embedding model setting is ignored by production RAG
• Azure RAG ignores configured embedding deployment
• Bound upstream response details in client-facing errors
• Wire Azure embedding deployment into the production RAG path
• Apply channel filtering to the API-polling evaluator
• Remove unrelated Alert Packs changes from this refactor
• Rule reuse key omits trigger mode
• Preserve temperature for Azure Chat Completions requests
• Align zero MaxTokens default with shared cost reservation
• Preserve temperature during Responses fallback
• On the Foundry flavor, chatDeployment prefers cfg.Deployment over cfg.Model. The settings…
• Because Capabilities advertises Streaming=true, dispatch always selects Stream, but this…
• Remove unrelated Alert Packs changes or expand PR scope
• Public OpenAPI spec omits Alert Packs endpoints
• Alert Packs changes are out of scope for the Foundry v1 PR
• Correct validation comment for remaining Azure probe fields

[atulmgupta]

Follow-up correction pushed in 7261999 after the stronger CI matrix caught a real platform-specific issue: Linux system-font metrics made the Typography value overflow at 135% text scale. Settings summary cards now permit shrinking and word wrapping; the strict overflow assertions remain unchanged.

The visual diff was isolated to synthetic long-copy mutation of the live native footer. The fixture previously excluded explicit role=contentinfo but missed native footer elements; it now excludes both. Inspected and regenerated the two long-content baselines, then reran comparisons without update mode. Screenshot thresholds remain unchanged.

Final local output:

32 passed (2.0m)
Test Files  2 passed (2)
Tests       42 passed (42)
9 skipped
11 passed (40.1s)
E2E preview proxy escape check: 0 errors
Image teslasync-web Built
Settings page lines: 316 -> 317
Sections: 5 -> 5

TypeScript, scoped ESLint and the Settings audit also pass. Awaiting the new-head CI results; no merge/deployment.

[Copilot]

One or more custom setup steps configured for this repository failed during this Copilot code review run:

Verify TypeScript build

Setup steps run before each review. If the review above is missing context, or no review was posted at all, the failing step above may be the cause. See the workflow run for failure details, fix your setup steps configuration, and re-request a review.

Note

You can configure setup steps for Copilot code review separately from Copilot cloud agent with a copilot-code-review.yml file. Read the docs for details.

[Copilot]

Copilot review overview

🔵 Needs a closer look

Add the missing public OpenAPI contracts and clarify or separate the Alert Packs scope.

Review effort: Lite
Findings: 5 · 8 · 4

Open (17)

• Preserve GitHub Pages base path in documentation links
• Azure embedding model setting is ignored by production RAG
• Azure RAG ignores configured embedding deployment
• Bound upstream response details in client-facing errors
• Wire Azure embedding deployment into the production RAG path
• Apply channel filtering to the API-polling evaluator
• Remove unrelated Alert Packs changes from this refactor
• Rule reuse key omits trigger mode
• Preserve temperature for Azure Chat Completions requests
• Align zero MaxTokens default with shared cost reservation
• Preserve temperature during Responses fallback
• On the Foundry flavor, chatDeployment prefers cfg.Deployment over cfg.Model. The settings…
• Because Capabilities advertises Streaming=true, dispatch always selects Stream, but this…
• Remove unrelated Alert Packs changes or expand PR scope
• Public OpenAPI spec omits Alert Packs endpoints
• Alert Packs changes are out of scope for the Foundry v1 PR
• Correct validation comment for remaining Azure probe fields

[atulmgupta]

Reworked the pack preview in f3c43b6 using the flat editable-grid pattern discussed with the user (KendoReact as an interaction reference, no new UI dependency).

• One property per column: Operator, Value, Cooldown, Alert behavior, Channels, Notification message, Include title, and Defaults reset. No nested mini-forms or repeated labels inside desktop cells.
• Header labels align with their cells; editors use consistent 44px heights. Compact rows remain under the regression's 100px limit.
• Five pack-default controls occupy one aligned row. Removed the ambiguous Master: Once option; each rule displays its effective Once per condition / Repeat while active value.
• Editing still overrides only that field. Reset delivery / Apply defaults to all preserve messages, operators, thresholds and channels.
• Per-row Helix review/apply, mobile cards, pagination, dirty guards and explicit installation remain intact.

Final local output:

Test Files 56 passed (56)
Tests      673 passed (673)
22 passed (1.8m)
E2E preview proxy escape check: 0 errors
Image teslasync-web Built
Pack component lines: 618 -> 641
GlassPanel/ChartContainer sections: 3 -> 3

TypeScript, full frontend lint, scoped violations audit and hook-route verification passed. Browser coverage now includes 320, 390, 768, 1024, 1280, 1440, 1920 and 2560px in both themes, asserting default-control alignment, cell alignment, flat column structure and compact row height. Fresh desktop/mobile screenshots were inspected.

CI is being monitored. No merge or deployment performed.

[Copilot]

One or more custom setup steps configured for this repository failed during this Copilot code review run:

Verify TypeScript build

Setup steps run before each review. If the review above is missing context, or no review was posted at all, the failing step above may be the cause. See the workflow run for failure details, fix your setup steps configuration, and re-request a review.

Note

You can configure setup steps for Copilot code review separately from Copilot cloud agent with a copilot-code-review.yml file. Read the docs for details.

[Copilot]

Copilot review overview

🔵 Needs a closer look

Unresolved API contract, Azure authentication and embedding, test coverage, localization, and scope issues remain.

Review effort: Lite
Findings: 5 · 8 · 4

Open (17)

• Preserve GitHub Pages base path in documentation links
• Azure embedding model setting is ignored by production RAG
• Azure RAG ignores configured embedding deployment
• Bound upstream response details in client-facing errors
• Wire Azure embedding deployment into the production RAG path
• Apply channel filtering to the API-polling evaluator
• Remove unrelated Alert Packs changes from this refactor
• Rule reuse key omits trigger mode
• Preserve temperature for Azure Chat Completions requests
• Align zero MaxTokens default with shared cost reservation
• Preserve temperature during Responses fallback
• On the Foundry flavor, chatDeployment prefers cfg.Deployment over cfg.Model. The settings…
• Because Capabilities advertises Streaming=true, dispatch always selects Stream, but this…
• Remove unrelated Alert Packs changes or expand PR scope
• Public OpenAPI spec omits Alert Packs endpoints
• Alert Packs changes are out of scope for the Foundry v1 PR
• Correct validation comment for remaining Azure probe fields

Previously missed (1)

In code that hasn't changed since last review

Document new Alert Packs API routes in OpenAPI

docs/​public/​openapi.yaml:1373

This PR adds the pack catalog, installation/removal, and Helix draft routes in internal/api/router.go, but the public OpenAPI document has no /alerts/packs, /alerts/pack-installations, or /ai/alerts/packs/draft paths (only the bulk-delete/channel field additions are documented here). API consumers and generated clients therefore cannot discover the new feature; add the new operations and their request/response schemas alongside the bulk-delete entry.

[atulmgupta]

Final verification for f3c43b6: CI and Frontend quality gates succeeded, including the Linux responsive grid/alignment cases and all Windows visual shards. No failed or pending PR checks remain. Local frontend lint, 673 notification tests, 22 browser cases and the web Docker build passed. The Settings readability follow-up also remains green. PR is unmerged; no deployment performed.

[atulmgupta]

Implemented in 4534a79.

• Added Default channels to Pack defaults; untouched rules inherit, while explicit all/none/fixed-channel overrides survive default changes.
• Individual reset and Apply defaults to all rules now reset channels too. The existing API receives resolved per-rule channel IDs, never an unsupported root field.
• Shared the Studio cooldown default (15 minutes) and behavior options (Re-alert until resolved, Notify on event). Preserved packs' initial Notify on event selection and the ordinary new-rule editor's explicit-choice placeholder.
• Six top controls align in one row on wide desktops and wrap into aligned rows on narrower displays. Updated docs and inspected desktop/tablet/mobile screenshots.

Actual local output:

TypeScript exit: 0 (no diagnostics if exit 0)
Test Files  56 passed (56)
     Tests  678 passed (678)

Final pack-only run, including 10 new channel-contract cases:
Test Files  2 passed (2)
     Tests  35 passed (35)

22 passed (1.9m)
E2E preview proxy escape check: 0 errors

ALL CHECKS PASSED — 0 total violations
ALL EXTRACTED HOOK PATHS MATCH REGISTERED ROUTES
Pack component lines: 641 -> 641; panels: 3 -> 3
Studio page lines: 2413 -> 2413; panels: 8 -> 8
Image teslasync-web Built

Full frontend lint and final helper lint exited 0. Browser coverage includes 320–2560px, both themes, six-control alignment, per-rule channel overrides, bulk reset, and submitted channel routing. CI is pending; no merge or deployment.

[Copilot]

One or more custom setup steps configured for this repository failed during this Copilot code review run:

Verify TypeScript build

Setup steps run before each review. If the review above is missing context, or no review was posted at all, the failing step above may be the cause. See the workflow run for failure details, fix your setup steps configuration, and re-request a review.

Note

You can configure setup steps for Copilot code review separately from Copilot cloud agent with a copilot-code-review.yml file. Read the docs for details.

[Copilot]

Copilot review overview

🟡 Changes recommended

Unresolved critical, moderate, and nit findings remain, including the broken atomic-install fixture.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 6 · 8 · 4

Open (18)

• Atomic-install fixture uses an invalid pack version · New
• Preserve GitHub Pages base path in documentation links
• Azure embedding model setting is ignored by production RAG
• Azure RAG ignores configured embedding deployment
• Bound upstream response details in client-facing errors
• Wire Azure embedding deployment into the production RAG path
• Apply channel filtering to the API-polling evaluator
• Remove unrelated Alert Packs changes from this refactor
• Rule reuse key omits trigger mode
• Preserve temperature for Azure Chat Completions requests
• Align zero MaxTokens default with shared cost reservation
• Preserve temperature during Responses fallback
• On the Foundry flavor, chatDeployment prefers cfg.Deployment over cfg.Model. The settings…
• Because Capabilities advertises Streaming=true, dispatch always selects Stream, but this…
• Remove unrelated Alert Packs changes or expand PR scope
• Public OpenAPI spec omits Alert Packs endpoints
• Alert Packs changes are out of scope for the Foundry v1 PR
• Correct validation comment for remaining Azure probe fields

Previously missed (1)

In code that hasn't changed since last review

PR scope mixes Foundry changes with unrelated Alert Packs work

internal/​alertpacks/​doc.go:1

This PR is described as Microsoft Foundry v1-only, but it also introduces an unrelated Alert Packs domain/API/UI/migration surface (plus settings and CI changes). Please split these changes or update the PR scope/description before approval; reviewing them as one change materially increases the merge risk.

[atulmgupta]

Final verification for 4534a79: 64 checks passed, 4 conditionally skipped, no failed or pending checks. Main CI: https://github.com/ev-dev-labs/teslasync/actions/runs/35428386641 — frontend quality/browser gates: https://github.com/ev-dev-labs/teslasync/actions/runs/35428386642. Worktree is clean. No merge or deployment performed.