chore: integrate upstream v1.0.0 into fork

[pokabookinflab]

Summary

Integrate upstream envoyproxy/ai-gateway v1.0.0 (46 commits since v0.7.0) into our fork's main, plus one fork-only regression fix uncovered during dev deploy.

Upstream v1.0.0 highlights

• v1beta1 control-plane API declared stable
• Feat: OpenAI → GCP embedContent translation for gemini-embedding-2 (feat: add openai to gcp embedding translations for gemini embedding 2 envoyproxy/ai-gateway#2114)
• Feat: completion tokens details for Anthropic models (feat: add completion tokens details for anthropic models envoyproxy/ai-gateway#2199)
• Bug fixes: Bedrock translator coalescing (fix: use correct slice index for Bedrock tool result coalescing envoyproxy/ai-gateway#2237), Anthropic streaming usage double-counting (fix: double counting for the usage in stream mode for anthropic models envoyproxy/ai-gateway#2239), GCP camelCase JSON tags (fix: use camelCase JSON tags for GCP Vertex AI request fields envoyproxy/ai-gateway#2225), MCP secret leak (fix: ensure that plaintext API key secret ref is not exposed in MCP backend HTTPRoute envoyproxy/ai-gateway#2134), RBAC empty-namespace (fix: prevent empty namespace from breaking RBAC name rendering in helm chart envoyproxy/ai-gateway#2226), upstream ext_proc filter placement (extproc: skip CONTINUE_AND_REPLACE when no body change is needed envoyproxy/ai-gateway#2170, fix: insert AI Gateway extproc after the buffer filter envoyproxy/ai-gateway#2227)
• Security policy docs (docs: add security policy envoyproxy/ai-gateway#2269), QuotaPolicy capability guide (docs: add QuotaPolicy capability guide envoyproxy/ai-gateway#1947)

Merge conflict resolution

2 files:

• internal/apischema/gcp/gcp.go: fork's CachedContentRequest and upstream's new EmbedContent* structs both live at file-tail — kept both.
• internal/translator/openai_gcpvertexai_test.go: upstream removed the content-length entry from wantHeaderMut across 10 test cases (consistent with extproc: skip CONTINUE_AND_REPLACE when no body change is needed envoyproxy/ai-gateway#2170 skipping CONTINUE_AND_REPLACE when no body change needed) — took upstream's version.

Fork-only regression fix — 60fcd8cb

After the merge, POST /v1beta/cachedContents started 504-timing out (response_flags=UT). Root cause: v1.0.0 merge dropped a fork-authored guard in router.ProcessRequestBody that stored the pre-rewrite :path into x-ai-eg-original-path only when unset. Without the guard, the second write clobbered the header with the rewritten path (/cachedContents), and the upstream filter's processorForPath lookup no longer matched any registered prefix — the filter silently produced no response, and Envoy hit the 60 s timeout.

Restored the guard. Only paths that go through rewriteDeveloperAPIPath (cachedContents Developer-API form) were affected; generateContent was unaffected because its :path rewrite happens in the upstream stage.

Verified on dev

• go build ./... && go vet ./... ✓
• All affected unit tests pass (apischema/anthropic, apischema/openai, translator, extproc, endpointspec, mainlib, backendauth).
• Fork-only feature tests pass: cachedContents translator (Vertex form + Developer API form), expandToVertexFullModelPath, rewriteVertexModelName, ImageEdits translator, server prefix fallback router, mainlib path extractors for cachedContents + region.
• Deployed to dev cluster: inflab-60fcd8cb-dev. Verified end-to-end via POST /v1beta/cachedContents (created cache successfully), POST /v1beta/models/{m}:generateContent (cachedContentTokenCount reported).

Test plan

• CI passes
• Reviewer sanity-checks the two conflict resolutions
• After merge, prod deploy uses inflab-<merge-sha>-prod tag

🤖 Generated with Claude Code

Summary by CodeRabbit

• 새 기능

  • Gemini 임베딩 모델의 embedContent 및 멀티모달 입력을 지원합니다.
  • MCP 요청 본문 크기 제한과 초과 시 413 응답을 제공합니다.
  • Quota Policy 사용 가이드와 트래픽 관리 문서를 추가했습니다.
  • 컨트롤러가 Envoy Gateway 네임스페이스를 별도로 설정할 수 있습니다.

• 개선 및 버그 수정

  • MCP 인증 정보 주입과 오래된 자격 증명 정리를 개선했습니다.
  • 잘못된 게이트웨이 참조와 누락된 세션 ID를 명확히 처리합니다.
  • 민감한 인증 정보가 로그에 노출되지 않도록 마스킹합니다.
  • pprof 엔드포인트를 로컬호스트에서만 접근할 수 있도록 제한했습니다.

• 문서

  • 보안 취약점 제보 절차와 API 안정성·지원 정책을 문서화했습니다.

[coderabbitai]

📝 Walkthrough

Walkthrough

이번 변경은 임베딩 요청·Vertex AI 경로, 컨트롤러 네임스페이스와 MCP 자격 증명 처리, 프록시 보호, 로그 마스킹, 쿼터·보안 문서, 릴리스·CI 구성을 갱신합니다.

Changes

임베딩 및 런타임

Layer / File(s)	Summary
임베딩 계약과 Vertex AI 경로 internal/apischema/..., internal/translator/openai_gcpvertexai_embeddings.go, internal/translator/*embedding*_test.go	임베딩 요청이 completion/chat 판별 유니온을 사용하며, Vertex AI 모델별 predict·embedContent 요청과 응답 변환을 지원합니다.
번역·추적 동작 internal/translator/..., internal/tracing/..., tests/data-plane/...	시스템·도구 메시지, reasoning 토큰, 빈 응답과 입력 검증, 임베딩 추적 속성 및 사용량 기대값이 갱신됩니다.

컨트롤러 및 MCP

Layer / File(s)	Summary
Gateway 네임스페이스와 동기화 cmd/controller/..., internal/controller/...	Envoy Gateway 네임스페이스를 플래그에서 컨트롤러로 전달하고, Gateway 리소스를 두 네임스페이스에서 검색하며 누락 오류를 처리합니다.
MCP 자격 증명 주입 internal/controller/mcp_route.go, internal/internalapi/internalapi.go, internal/controller/mcp_route_test.go	API 키를 URL, 헤더 또는 관리 시크릿 기반 credential injection으로 구성하고, 오래된 시크릿을 삭제합니다.
프록시·extproc 보호 internal/mcpproxy/*, internal/extproc/*, internal/pprof/pprof.go	POST 바디 크기 제한, 세션 ID 검증, 조건부 바디 교체, localhost 전용 pprof 바인딩을 적용합니다.

문서 및 릴리스

Layer / File(s)	Summary
쿼터·보안 문서 RELEASES.md, SECURITY.md, api/..., site/docs/..., manifests/...	QuotaPolicy 동작과 제약, API 안정성, 보안 취약점 제보 절차를 문서화합니다.
릴리스 사이트와 Helm 구성 site/..., manifests/charts/...	사이트 버전·호환성·릴리스 콘텐츠와 Helm 문자열 출력 및 네임스페이스 인자를 갱신합니다.

빌드 유지보수

Layer / File(s)	Summary
CI와 모듈 버전 .github/workflows/*, go.mod, tools/go.mod, .gitignore	고정된 GitHub Actions 커밋, Go 및 모듈 버전, worktree 무시 규칙을 갱신합니다.

Estimated code review effort: 5 (Critical) | ~120 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Client
  participant EmbeddingTranslator
  participant VertexAI
  participant OpenAIResponse
  Client->>EmbeddingTranslator: 임베딩 요청
  EmbeddingTranslator->>VertexAI: predict 또는 embedContent
  VertexAI-->>EmbeddingTranslator: 임베딩·토큰 사용량
  EmbeddingTranslator->>OpenAIResponse: OpenAI 응답 변환

Loading

Poem

토끼가 당근처럼 코드를 갉아,
임베딩 길을 새로 닦았네.
시크릿은 꼭꼭 숨기고,
쿼터는 똑똑히 세고,
CI도 깡충 업데이트했네! 🐇

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feature/upstream-v1.0.0-integration

[coderabbitai]

Actionable comments posted: 12

🧹 Nitpick comments (6)

site/versioned_docs/version-0.7/_vars.json (1)

2-5: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

aigwGitRef가 여전히 main을 참조합니다.

aigwVersion은 이제 릴리스된 0.7.0으로 고정되었지만, aigwGitRef는 main(이동하는 브랜치)을 그대로 유지하고 있습니다. versioned docs는 해당 릴리스 시점의 소스코드를 링크해야 하므로, main을 참조하면 v0.7.0 릴리스 이후 main에 반영된 변경 사항(코드가 다를 수 있음)을 잘못 링크할 위험이 있습니다. v0.7.0 같은 고정 태그로 설정하는 것이 안전합니다.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@site/versioned_docs/version-0.7/_vars.json` around lines 2 - 5, Update the
aigwGitRef configuration in _vars.json from the moving main branch to the fixed
v0.7.0 release tag, keeping it aligned with aigwVersion.

internal/endpointspec/endpointspec_test.go (1)

151-179: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

파싱된 임베딩 유니온 페이로드도 검증하세요.

현재는 모델과 parsed != nil만 확인하므로 input 또는 messages가 OfCompletion/OfChat에 보존되지 않아도 통과합니다. 각각의 유니온 존재 여부와 입력 텍스트를 assertion으로 확인하세요.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/endpointspec/endpointspec_test.go` around lines 151 - 179,
Strengthen the “success with input” and “success with messages” tests after
spec.ParseBody by asserting the parsed embedding union is present in the
expected OfCompletion or OfChat variant and that it retains the original input
text or message content. Keep the existing model, stream, mutation, and error
assertions unchanged.

internal/translator/openai_awsbedrock_embeddings_test.go (1)

34-156: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

OfCompletion 누락 거부 경로를 테스트하세요.

현재 테이블은 모두 OfCompletion을 설정합니다. 구현의 Line 44-46 분기가 제거·변경되어도 감지되지 않으므로, OfCompletion이 없는 요청이 ErrInvalidRequestBody를 반환하는 케이스를 추가하세요.

제안된 테스트 케이스

+		{
+			name: "missing completion input rejected",
+			input: openai.EmbeddingRequest{
+				EmbeddingBaseRequest: openai.EmbeddingBaseRequest{
+					Model: "amazon.titan-embed-text-v2:0",
+				},
+			},
+			wantErr: true,
+		},

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/translator/openai_awsbedrock_embeddings_test.go` around lines 34 -
156, Extend the table-driven tests around the existing embedding request cases
with a request whose OfCompletion field is nil, and assert that it returns
ErrInvalidRequestBody. Keep the case focused on the missing OfCompletion
validation path and verify the specific error rather than only checking that an
error occurred.

internal/mcpproxy/handlers.go (2)

305-311: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

세션 검증 순서 수정에 대한 테스트 커버리지 부재.

doNotForwardResponseToBackends 체크 전에 s == nil 검사를 먼저 수행하도록 순서가 바뀌었는데(스펙 준수를 위한 올바른 수정), handlers_test.go에는 세션 ID 없이 *jsonrpc.Response를 보내 400을 기대하는 테스트가 보이지 않습니다. 회귀 방지를 위해 테스트 추가를 권장합니다.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/mcpproxy/handlers.go` around lines 305 - 311, 현재 세션 검증 순서가 회귀되지 않도록
해당 핸들러 테스트에 세션 ID 없이 *jsonrpc.Response를 전송하는 케이스를 추가하세요.
doNotForwardResponseToBackends 처리와 관계없이 s == nil일 때 HTTP 400과 적절한 “missing
session ID” 오류 응답이 반환되는지 검증하고, 기존 유효 세션 테스트는 유지하세요.

---

279-294: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

413 응답에 MCPErrorInternal 사용은 오분류.

요청 바디 크기 초과는 클라이언트 오류인데 errType을 metrics.MCPErrorInternal("internal_error")로 설정하고 있습니다. 메트릭/로그에서 클라이언트 유발 오류가 서버 내부 오류로 잘못 집계될 수 있습니다. metrics.MCPErrorInvalidParam 등 클라이언트 오류 계열로 변경을 고려하세요.

🛠️ 제안

 		var maxBytesErr *http.MaxBytesError
 		if errors.As(err, &maxBytesErr) {
-			errType = metrics.MCPErrorInternal
+			errType = metrics.MCPErrorInvalidParam
 			onErrorResponse(w, http.StatusRequestEntityTooLarge, "request body too large")
 			return
 		}

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/mcpproxy/handlers.go` around lines 279 - 294, Update the
max-request-body error branch in the handler around http.MaxBytesReader so
requests exceeding the limit set errType to the appropriate client-error metric,
such as metrics.MCPErrorInvalidParam, instead of metrics.MCPErrorInternal;
preserve the existing 413 response and handling for other body-read errors.

internal/translator/anthropic_helper_test.go (1)

1164-1295: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

ReasoningTokens 검증 누락.

이 PR의 핵심 변경인 SetReasoningTokens(message_delta의 output_tokens_details.thinking_tokens 기반)가 이 스트리밍 테스트 스위트에서 전혀 검증되지 않습니다. 테스트 케이스에 output_tokens_details를 포함한 시나리오와 tokenUsage.ReasoningTokens() assertion을 추가하는 것을 권장합니다.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/translator/anthropic_helper_test.go` around lines 1164 - 1295,
Extend TestAnthropicStreamParser_StreamingTokenUsage with a scenario containing
message_delta.usage.output_tokens_details.thinking_tokens, add an expected
reasoning-token value to the table, and assert tokenUsage.ReasoningTokens() is
set and matches it after streaming completes. Ensure the fixture exercises the
SetReasoningTokens path while preserving existing token assertions.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.gitignore:
- Line 55: Update the .worktrees/ pattern in the root .gitignore to /.worktrees/
so only the repository-root worktree directory is ignored, while same-named
directories in nested paths remain unaffected.

In `@api/v1alpha1/quota_policy.go`:
- Around line 15-18: Update the top-level quota policy description near
QuotaRule/ShadowMode to state that 429 responses apply only to non-shadow
quotas, while ShadowMode quotas continue allowing requests after exhaustion.
Regenerate the affected CRD/API documentation so the corrected behavior is
reflected there as well.

In `@internal/controller/gateway.go`:
- Around line 1151-1187: Deduplicate the namespace list used by the pod,
deployment, and daemonset collection loop before iterating. Ensure gw.Namespace
is queried only once when it equals c.envoyGatewayNamespace, and avoid adding it
again when c.envoyGatewayNamespace is empty, while preserving distinctNamespaces
validation and existing list behavior.

In `@internal/controller/mcp_route.go`:
- Around line 596-604: Update the query-parameter URL construction in the
HTTPRoute rule generation flow around readAPIKey to URL-escape both
apiKey.QueryParam and apiKeyLiteral before composing fullPathPtr. Add the
required net/url import and preserve the existing error handling and parameter
structure.

In `@internal/tracing/openinference/anthropic/messages.go`:
- Around line 273-277: Apply the same non-negative and maximum-index validation
used for start events to ContentBlockDelta and ContentBlockStop before any slice
access in the message handling flow. Update
internal/tracing/openinference/anthropic/messages.go at lines 273-277 and the
corresponding delta/stop paths; add regression coverage in
internal/tracing/openinference/anthropic/messages_test.go at lines 213-268
confirming that negative delta and stop indices after a valid start are ignored
without panicking.

In `@internal/translator/anthropic_awsbedrock.go`:
- Around line 184-186: Update the system prompt construction around systemTexts
and body.System so existing top-level system content is preserved when messages
also contain role "system" entries. Merge the existing body.System text with the
promoted systemText blocks, retaining both contents and their ordering, instead
of unconditionally replacing body.System.

In `@internal/translator/anthropic_helper.go`:
- Around line 432-435: Keep the nil check in the tool-call validation flow
before any dereference of toolCall.ID, including the later use in the
surrounding translator logic. When toolCall.ID is nil, return the existing
internalapi.ErrInvalidRequestBody-wrapped error with the tool-call index and
required id field context, preventing a panic.

In `@internal/translator/openai_awsbedrock.go`:
- Around line 417-419: Update the tool-call ID validation in the translator flow
around toolCall.ID so it rejects nil, empty, and values that do not match
Bedrock’s allowed toolUseId pattern. Return the existing
internalapi.ErrInvalidRequestBody-wrapped error with the tool-call index for
every invalid ID, before forwarding it upstream.

In `@internal/translator/openai_gcpvertexai_embeddings_test.go`:
- Around line 476-483: Update the wantBodyNotContain assertions in the data URI
test case to use the correctly cased serialized key "fileUri" instead of
"fileURI", preserving the existing checks for "instances" and "parameters".

In `@internal/translator/openai_gcpvertexai_embeddings.go`:
- Around line 148-153: Update the global title assignment in the instances loop
so openAIReq.Title is applied only when the request task type is
RETRIEVAL_DOCUMENT. Preserve the existing behavior of leaving titles unset for
all other task types.
- Around line 165-171: Update isEmbedContentModel to recognize
gemini-embedding-001 by prefix rather than exact equality, so path- or
suffix-qualified forms such as `@default` and /models/... are routed to the
predict endpoint while other Gemini models still use embedContent.

In `@SECURITY.md`:
- Around line 69-81: Update the support-policy wording in SECURITY.md to align
with RELEASES.md and the documented v0.7.x support status: explicitly include
v0.7.x as a supported pre-v1 release or define the pre-v1 support policy
separately, and ensure the stated “two most recent minor releases” rule does not
incorrectly apply before v1.0.0.

---

Nitpick comments:
In `@internal/endpointspec/endpointspec_test.go`:
- Around line 151-179: Strengthen the “success with input” and “success with
messages” tests after spec.ParseBody by asserting the parsed embedding union is
present in the expected OfCompletion or OfChat variant and that it retains the
original input text or message content. Keep the existing model, stream,
mutation, and error assertions unchanged.

In `@internal/mcpproxy/handlers.go`:
- Around line 305-311: 현재 세션 검증 순서가 회귀되지 않도록 해당 핸들러 테스트에 세션 ID 없이
*jsonrpc.Response를 전송하는 케이스를 추가하세요. doNotForwardResponseToBackends 처리와 관계없이 s ==
nil일 때 HTTP 400과 적절한 “missing session ID” 오류 응답이 반환되는지 검증하고, 기존 유효 세션 테스트는
유지하세요.
- Around line 279-294: Update the max-request-body error branch in the handler
around http.MaxBytesReader so requests exceeding the limit set errType to the
appropriate client-error metric, such as metrics.MCPErrorInvalidParam, instead
of metrics.MCPErrorInternal; preserve the existing 413 response and handling for
other body-read errors.

In `@internal/translator/anthropic_helper_test.go`:
- Around line 1164-1295: Extend TestAnthropicStreamParser_StreamingTokenUsage
with a scenario containing
message_delta.usage.output_tokens_details.thinking_tokens, add an expected
reasoning-token value to the table, and assert tokenUsage.ReasoningTokens() is
set and matches it after streaming completes. Ensure the fixture exercises the
SetReasoningTokens path while preserving existing token assertions.

In `@internal/translator/openai_awsbedrock_embeddings_test.go`:
- Around line 34-156: Extend the table-driven tests around the existing
embedding request cases with a request whose OfCompletion field is nil, and
assert that it returns ErrInvalidRequestBody. Keep the case focused on the
missing OfCompletion validation path and verify the specific error rather than
only checking that an error occurred.

In `@site/versioned_docs/version-0.7/_vars.json`:
- Around line 2-5: Update the aigwGitRef configuration in _vars.json from the
moving main branch to the fixed v0.7.0 release tag, keeping it aligned with
aigwVersion.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: ea6c1fdb-c66e-4bb8-8cdf-9594eeaf7c48

📥 Commits

Reviewing files that changed from the base of the PR and between 7983c22 and 60fcd8c.

⛔ Files ignored due to path filters (4)

• go.sum is excluded by !**/*.sum
• site/package-lock.json is excluded by !**/package-lock.json
• site/static/img/adopters/stacklok.png is excluded by !**/*.png
• tools/go.sum is excluded by !**/*.sum

📒 Files selected for processing (86)

• .github/workflows/build_and_test.yaml
• .github/workflows/codeql.yaml
• .github/workflows/docker_build_job.yaml
• .github/workflows/release.yaml
• .gitignore
• RELEASES.md
• SECURITY.md
• api/v1alpha1/quota_policy.go
• api/v1beta1/mcp_route.go
• cmd/aigw/translate.go
• cmd/controller/main.go
• cmd/controller/main_test.go
• go.mod
• internal/apischema/gcp/gcp.go
• internal/apischema/openai/openai.go
• internal/bodymutator/body_mutator.go
• internal/controller/ai_gateway_route.go
• internal/controller/ai_gateway_route_test.go
• internal/controller/controller.go
• internal/controller/gateway.go
• internal/controller/gateway_test.go
• internal/controller/mcp_route.go
• internal/controller/mcp_route_test.go
• internal/endpointspec/endpointspec_test.go
• internal/extensionserver/extensionserver_test.go
• internal/extensionserver/inferencepool.go
• internal/extensionserver/post_translate_modify.go
• internal/extensionserver/post_translate_modify_test.go
• internal/extproc/processor_impl.go
• internal/extproc/processor_impl_test.go
• internal/filterapi/filterconfig.go
• internal/filterapi/filterconfig_test.go
• internal/internalapi/internalapi.go
• internal/mcpproxy/config.go
• internal/mcpproxy/handlers.go
• internal/mcpproxy/handlers_test.go
• internal/mcpproxy/mcpproxy.go
• internal/mcpproxy/mcpproxy_test.go
• internal/pprof/pprof.go
• internal/tracing/openinference/anthropic/messages.go
• internal/tracing/openinference/anthropic/messages_test.go
• internal/tracing/openinference/openai/embeddings_test.go
• internal/tracing/openinference/openai/request_attrs.go
• internal/tracing/tracing_test.go
• internal/translator/anthropic_awsbedrock.go
• internal/translator/anthropic_awsbedrock_test.go
• internal/translator/anthropic_helper.go
• internal/translator/anthropic_helper_test.go
• internal/translator/anthropic_openai_test.go
• internal/translator/gemini_helper.go
• internal/translator/openai_awsanthropic_test.go
• internal/translator/openai_awsbedrock.go
• internal/translator/openai_awsbedrock_embeddings.go
• internal/translator/openai_awsbedrock_embeddings_test.go
• internal/translator/openai_awsbedrock_test.go
• internal/translator/openai_gcpanthropic_test.go
• internal/translator/openai_gcpvertexai_embeddings.go
• internal/translator/openai_gcpvertexai_embeddings_test.go
• internal/translator/openai_gcpvertexai_test.go
• internal/translator/openai_openai.go
• internal/translator/openai_openai_test.go
• manifests/charts/ai-gateway-crds-helm/templates/aigateway.envoyproxy.io_mcproutes.yaml
• manifests/charts/ai-gateway-crds-helm/templates/aigateway.envoyproxy.io_quotapolicies.yaml
• manifests/charts/ai-gateway-helm/templates/_helpers.tpl
• manifests/charts/ai-gateway-helm/templates/deployment.yaml
• manifests/charts/ai-gateway-helm/templates/envoy_gateway_cluster_role_for_inference_pool.yaml
• manifests/charts/ai-gateway-helm/templates/serviceaccount.yaml
• site/blog/authors.yml
• site/docs/_vars.json
• site/docs/api/api.mdx
• site/docs/capabilities/index.md
• site/docs/capabilities/traffic/index.md
• site/docs/capabilities/traffic/quota-policy.md
• site/docs/capabilities/traffic/usage-based-ratelimiting.md
• site/docs/compatibility.md
• site/docusaurus.config.ts
• site/package.json
• site/src/components/HomepageFeatures/index.tsx
• site/src/data/adopters/adopters.json
• site/src/data/talks.json
• site/versioned_docs/version-0.7/_vars.json
• site/versioned_docs/version-0.7/compatibility.md
• tests/data-plane/testupstream_test.go
• tests/internal/testopenai/embeddings_requests.go
• tests/internal/testopenai/openai_test.go
• tools/go.mod

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
set -euo pipefail

# Inspect the root .gitignore around the reported line.
echo "== .gitignore context =="
sed -n '45,65p' .gitignore | cat -n

# Check gitignore matching behavior for the current pattern.
tmpfile="$(mktemp)"
cat > "$tmpfile" <<'EOF'
.worktrees/
foo/.worktrees/
bar/baz/.worktrees/
EOF

echo
echo "== git check-ignore with current rules =="
git check-ignore -v --stdin < "$tmpfile" || true

rm -f "$tmpfile"

Repository: inflearn/ai-gateway

Length of output: 549

---

.gitignore:55에서 패턴을 루트로 고정하세요.

루트 .gitignore의 .worktrees/는 하위 경로의 동일한 디렉터리명도 무시합니다. 저장소 루트의 worktree 디렉터리만 제외하려면 /.worktrees/로 바꾸세요.

권장 수정

- .worktrees/
+ /.worktrees/

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	.worktrees/
	/.worktrees/

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In @.gitignore at line 55, Update the .worktrees/ pattern in the root .gitignore
to /.worktrees/ so only the repository-root worktree directory is ignored, while
same-named directories in nested paths remain unaffected.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

ShadowMode 예외를 최상위 429 설명에 반영하세요.

QuotaRule.ShadowMode는 quota 초과 시에도 요청을 항상 성공시킨다고 명시하지만, 최상위 설명은 무조건 429를 반환하는 것처럼 읽힙니다. 비-shadow quota에만 적용된다는 조건을 명시하고 생성된 CRD/API 문서도 함께 갱신해야 합니다.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@api/v1alpha1/quota_policy.go` around lines 15 - 18, Update the top-level
quota policy description near QuotaRule/ShadowMode to state that 429 responses
apply only to non-shadow quotas, while ShadowMode quotas continue allowing
requests after exhaustion. Regenerate the affected CRD/API documentation so the
corrected behavior is reflected there as well.

[coderabbitai]

🩺 Stability & Availability | 🔴 Critical | ⚡ Quick win

gw.Namespace == c.envoyGatewayNamespace인 경우 동일 네임스페이스를 두 번 조회해 reconcile이 영구 실패합니다.

Gateway 리소스가 Envoy Gateway 네임스페이스(기본 envoy-gateway-system)에 생성되면 루프가 같은 네임스페이스를 두 번 순회합니다. 그 결과 pods/deployments/daemonSets에 동일 객체가 중복 append되고, distinctNamespaces에도 같은 값이 두 번 들어가 len > 1이 성립하여 found gateway-labeled objects in multiple namespaces: [envoy-gateway-system envoy-gateway-system] 에러로 매 reconcile이 실패합니다. c.envoyGatewayNamespace가 빈 문자열인 경우(모든 네임스페이스 조회)에도 동일하게 gw.Namespace 객체가 중복 수집됩니다.

순회 대상 네임스페이스를 중복 제거해 주세요.

🐛 중복 네임스페이스 제거 제안

 	var distinctNamespaces []string
-	for _, ns := range []string{gw.Namespace, c.envoyGatewayNamespace} {
+	namespaces := []string{gw.Namespace}
+	if c.envoyGatewayNamespace != "" && c.envoyGatewayNamespace != gw.Namespace {
+		namespaces = append(namespaces, c.envoyGatewayNamespace)
+	}
+	for _, ns := range namespaces {
 		var ps *corev1.PodList

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/controller/gateway.go` around lines 1151 - 1187, Deduplicate the
namespace list used by the pod, deployment, and daemonset collection loop before
iterating. Ensure gw.Namespace is queried only once when it equals
c.envoyGatewayNamespace, and avoid adding it again when c.envoyGatewayNamespace
is empty, while preserving distinctNamespaces validation and existing list
behavior.

[coderabbitai]

🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

쿼리 파라미터 API 키를 URL 인코딩 없이 경로에 삽입합니다.

apiKeyLiteral이 &, #, =, 공백 등을 포함하면 ReplaceFullPath 값이 왜곡되어 파라미터가 잘리거나 추가 쿼리 파라미터로 해석됩니다. 키 이름과 값 모두 이스케이프하는 것이 안전합니다.

🐛 URL 이스케이프 적용 제안

-			fullPathPtr = fmt.Sprintf("%s?%s=%s", fullPathPtr, *apiKey.QueryParam, apiKeyLiteral)
+			fullPathPtr = fmt.Sprintf("%s?%s=%s", fullPathPtr,
+				url.QueryEscape(*apiKey.QueryParam), url.QueryEscape(apiKeyLiteral))

net/url 임포트가 필요합니다.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/controller/mcp_route.go` around lines 596 - 604, Update the
query-parameter URL construction in the HTTPRoute rule generation flow around
readAPIKey to URL-escape both apiKey.QueryParam and apiKeyLiteral before
composing fullPathPtr. Add the required net/url import and preserve the existing
error handling and parameter structure.

[coderabbitai]

🩺 Stability & Availability | 🟠 Major | ⚡ Quick win

SSE 인덱스 검증이 start 이벤트에만 적용됩니다. ContentBlockDelta와 ContentBlockStop은 음수 index를 검사하지 않아, 정상 start 뒤의 -1 이벤트가 음수 슬라이스 접근 panic을 일으킬 수 있습니다.

• internal/tracing/openinference/anthropic/messages.go#L273-L277: start, delta, stop 모두에 공통 범위 검증을 적용하세요.
• internal/tracing/openinference/anthropic/messages_test.go#L213-L268: 유효한 start 뒤 음수 delta/stop을 보내도 panic 없이 무시되는 회귀 테스트를 추가하세요.

📍 Affects 2 files

• internal/tracing/openinference/anthropic/messages.go#L273-L277 (this comment)
• internal/tracing/openinference/anthropic/messages_test.go#L213-L268

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/tracing/openinference/anthropic/messages.go` around lines 273 - 277,
Apply the same non-negative and maximum-index validation used for start events
to ContentBlockDelta and ContentBlockStop before any slice access in the message
handling flow. Update internal/tracing/openinference/anthropic/messages.go at
lines 273-277 and the corresponding delta/stop paths; add regression coverage in
internal/tracing/openinference/anthropic/messages_test.go at lines 213-268
confirming that negative delta and stop indices after a valid start are ignored
without panicking.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

빈 문자열과 형식이 잘못된 tool-call ID도 거부하세요.

Line 417은 nil만 검사합니다. "" 또는 Bedrock 허용 패턴 밖의 값은 그대로 toolUseId로 전송되어 업스트림 400을 유발합니다. Bedrock의 허용 패턴까지 검증해 invalid-request-body 오류로 반환하세요.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/translator/openai_awsbedrock.go` around lines 417 - 419, Update the
tool-call ID validation in the translator flow around toolCall.ID so it rejects
nil, empty, and values that do not match Bedrock’s allowed toolUseId pattern.
Return the existing internalapi.ErrInvalidRequestBody-wrapped error with the
tool-call index for every invalid ID, before forwarding it upstream.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

"fileURI" 단정은 대소문자가 달라 항상 통과합니다.

실제 직렬화 키는 Line 449에서 확인되듯 fileUri입니다. data URI 케이스에서 파일 URI가 새지 않는지 실제로 검증하려면 케이싱을 맞춰야 합니다.

💚 제안 수정

-			wantBodyNotContain: []string{`"instances"`, `"parameters"`, `"fileURI"`},
+			wantBodyNotContain: []string{`"instances"`, `"parameters"`, `"fileUri"`},

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	wantBodyContains: []string{
	`"content"`,
	`"parts"`,
	`"inlineData"`,
	`"mimeType":"image/png"`,
	},
	wantBodyNotContain: []string{`"instances"`, `"parameters"`, `"fileURI"`},
	},
	wantBodyContains: []string{
	`"content"`,
	`"parts"`,
	`"inlineData"`,
	`"mimeType":"image/png"`,
	},
	wantBodyNotContain: []string{`"instances"`, `"parameters"`, `"fileUri"`},
	},

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/translator/openai_gcpvertexai_embeddings_test.go` around lines 476 -
483, Update the wantBodyNotContain assertions in the data URI test case to use
the correctly cased serialized key "fileUri" instead of "fileURI", preserving
the existing checks for "instances" and "parameters".

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

전역 Title은 task_type=RETRIEVAL_DOCUMENT 조건을 무시합니다.

같은 파일의 per-item 경로(Line 61-65, 74-76)는 RETRIEVAL_DOCUMENT일 때만 title을 설정하는데, 전역 Title은 task type과 무관하게 모든 instance에 주입됩니다. Vertex AI는 다른 task type과 함께 온 title을 거부하므로 400이 발생할 수 있습니다.

🐛 제안 수정

 		// Apply global title to all instances if specified.
 		if openAIReq.Title != "" {
 			for _, instance := range instances {
-				instance.Title = openAIReq.Title
+				if instance.TaskType == openai.EmbeddingTaskTypeRetrievalDocument {
+					instance.Title = openAIReq.Title
+				}
 			}
 		}

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	// Apply global title to all instances if specified.
	if openAIReq.Title != "" {
	for _, instance := range instances {
	instance.Title = openAIReq.Title
	}
	}
	// Apply global title to all instances if specified.
	if openAIReq.Title != "" {
	for _, instance := range instances {
	if instance.TaskType == openai.EmbeddingTaskTypeRetrievalDocument {
	instance.Title = openAIReq.Title
	}
	}
	}

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/translator/openai_gcpvertexai_embeddings.go` around lines 148 - 153,
Update the global title assignment in the instances loop so openAIReq.Title is
applied only when the request task type is RETRIEVAL_DOCUMENT. Preserve the
existing behavior of leaving titles unset for all other task types.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

모델명 판정이 부분 문자열 기반이라 버전/경로 접미사에 취약합니다.

gemini-embedding-001은 완전 일치만 예외 처리되므로 gemini-embedding-001@default, .../models/gemini-embedding-001 같이 접미사·경로가 붙은 형태(모델 오버라이드로 지정 가능)는 embedContent로 잘못 라우팅되어 404/400을 유발합니다. 접두사 기반 판정이 더 안전합니다.

🛡️ 제안 수정

 func isEmbedContentModel(model string) bool {
-	return strings.Contains(model, "gemini") && model != "gemini-embedding-001"
+	if !strings.Contains(model, "gemini") {
+		return false
+	}
+	// gemini-embedding-001 (and its versioned/qualified variants) still use predict.
+	base := model[strings.LastIndex(model, "/")+1:]
+	return !strings.HasPrefix(base, "gemini-embedding-001")
 }

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	// isEmbedContentModel returns true if the model should use the embedContent endpoint
	// instead of the predict endpoint.
	// Reference: https://github.com/googleapis/go-genai/blob/v1.54.0/transformer.go#L565
	// Check and update this function when new Gemini embedding model versions are released.
	func isEmbedContentModel(model string) bool {
	return strings.Contains(model, "gemini") && model != "gemini-embedding-001"
	}
	// isEmbedContentModel returns true if the model should use the embedContent endpoint
	// instead of the predict endpoint.
	// Reference: https://github.com/googleapis/go-genai/blob/v1.54.0/transformer.go#L565
	// Check and update this function when new Gemini embedding model versions are released.
	func isEmbedContentModel(model string) bool {
	if !strings.Contains(model, "gemini") {
	return false
	}
	// gemini-embedding-001 (and its versioned/qualified variants) still use predict.
	base := model[strings.LastIndex(model, "/")+1:]
	return !strings.HasPrefix(base, "gemini-embedding-001")
	}

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/translator/openai_gcpvertexai_embeddings.go` around lines 165 - 171,
Update isEmbedContentModel to recognize gemini-embedding-001 by prefix rather
than exact equality, so path- or suffix-qualified forms such as `@default` and
/models/... are routed to the predict endpoint while other Gemini models still
use embedContent.

[coderabbitai]

🔒 Security & Privacy | 🟠 Major | ⚡ Quick win

지원 버전 범위가 저장소의 다른 정책과 충돌합니다.

이 문서는 현재 및 이전 v1.x만 지원한다고 하지만, site/versioned_docs/version-0.7/compatibility.md Line 14는 v0.7.x를 Supported로 표시하고 RELEASES.md Line 16-17은 두 릴리스 이후 EOL을 정의합니다. v1.0.0에는 이전 v1.x 라인도 없습니다. 실제 보안 백포트 대상인 v0.7.x를 명시하거나, pre-v1 지원 정책을 별도로 정의하세요.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@SECURITY.md` around lines 69 - 81, Update the support-policy wording in
SECURITY.md to align with RELEASES.md and the documented v0.7.x support status:
explicitly include v0.7.x as a supported pre-v1 release or define the pre-v1
support policy separately, and ensure the stated “two most recent minor
releases” rule does not incorrectly apply before v1.0.0.