From 3c538c53913bcf17b0fca939a2ff9c7c742577e2 Mon Sep 17 00:00:00 2001 From: mlsmaycon Date: Thu, 23 Jul 2026 03:40:39 +0000 Subject: [PATCH] [e2e] exercise Kimi via base-URL /anthropic prefix on a bare upstream The documented Claude Code / Kimi CLI setup keeps the provider's default https://api.moonshot.ai upstream and carries the /anthropic shape prefix in the agent's base URL, riding through the endpoint to Moonshot. Add ChatPrefixed to the harness and run the Kimi scenario that way. --- e2e/agentnetwork/chat_test.go | 42 ++++++++++++++++-------------- e2e/agentnetwork/guardrail_test.go | 2 +- e2e/harness/client.go | 13 ++++++++- 3 files changed, 36 insertions(+), 21 deletions(-) diff --git a/e2e/agentnetwork/chat_test.go b/e2e/agentnetwork/chat_test.go index c38d297ae..ad145b37e 100644 --- a/e2e/agentnetwork/chat_test.go +++ b/e2e/agentnetwork/chat_test.go @@ -20,14 +20,15 @@ import ( // covers whatever credentials are present (source ~/.llm-keys locally / set the // Actions secrets in CI). type providerCase struct { - name string - catalogID string - upstream string - apiKey string - model string // body model (chat/messages) or path model@version (vertex) - kind string // harness.WireChat, harness.WireMessages, or harness.WireVertex - project string // vertex only: GCP project for the rawPredict path - region string // vertex only: GCP region for the rawPredict path + name string + catalogID string + upstream string + apiKey string + model string // body model (chat/messages) or path model@version (vertex) + kind string // harness.WireChat, harness.WireMessages, or harness.WireVertex + project string // vertex only: GCP project for the rawPredict path + region string // vertex only: GCP region for the rawPredict path + pathPrefix string // base-URL path prefix the agent carries (e.g. "/anthropic" for Kimi) } // availableProviders builds the matrix from the provider env vars that are set. @@ -43,16 +44,19 @@ func availableProviders() []providerCase { // Kimi (Moonshot AI) serves two body shapes from the same key: OpenAI // Chat Completions on the bare host (/v1/...) and the Anthropic // Messages API under the /anthropic path prefix (the endpoint - // Moonshot's Claude Code guide uses). The platform serves this - // account exactly ONE model — kimi-k3 (kimi-k2-thinking and even - // kimi-latest return resource_not_found_error on both surfaces) — so - // two concurrent provider records would claim the same model and - // route ambiguously. Run the Anthropic shape, the flagship Claude - // Code path; the OpenAI wire shape is covered live by the other - // chat-shaped matrix providers, and Kimi-over-chat passed with - // kimi-k3 before the single-model constraint surfaced (run #73 on - // the kimi feature branch). - ps = append(ps, providerCase{name: "kimi", catalogID: "kimi_api", upstream: "https://api.moonshot.ai/anthropic", apiKey: k, model: "kimi-k3", kind: harness.WireMessages}) + // Moonshot's Claude Code guide uses). The provider keeps the bare + // default upstream and the AGENT carries the /anthropic prefix in + // its base URL — exactly the documented Claude Code / Kimi CLI + // setup (ANTHROPIC_BASE_URL=https:///anthropic) — so one + // provider serves both shapes and the prefix rides through to + // Moonshot. Run the Anthropic shape, the flagship Claude Code path; + // the OpenAI wire shape is covered live by the other chat-shaped + // matrix providers, and Kimi-over-chat passed with kimi-k3 before + // the single-model constraint surfaced (run #73 on the kimi feature + // branch). The platform serves this account exactly ONE model — + // kimi-k3 (kimi-k2-thinking and even kimi-latest return + // resource_not_found_error on both surfaces). + ps = append(ps, providerCase{name: "kimi", catalogID: "kimi_api", upstream: "https://api.moonshot.ai", apiKey: k, model: "kimi-k3", kind: harness.WireMessages, pathPrefix: "/anthropic"}) } if k, u := os.Getenv("VERCEL_TOKEN"), os.Getenv("VERCEL_URL"); k != "" && u != "" { ps = append(ps, providerCase{name: "vercel", catalogID: "vercel_ai_gateway", upstream: u, apiKey: k, model: "openai/gpt-4o-mini", kind: harness.WireChat}) @@ -268,7 +272,7 @@ func TestProvidersMatrix(t *testing.T) { case harness.WireBedrock: c, b, cerr = cl.Bedrock(ctx, settings.Endpoint, proxyIP, pc.model, "Reply with exactly: pong", sessionID) default: - c, b, cerr = cl.Chat(ctx, settings.Endpoint, proxyIP, pc.kind, pc.model, "Reply with exactly: pong", sessionID) + c, b, cerr = cl.ChatPrefixed(ctx, settings.Endpoint, proxyIP, pc.pathPrefix, pc.kind, pc.model, "Reply with exactly: pong", sessionID) } if cerr == nil { code, body = c, b diff --git a/e2e/agentnetwork/guardrail_test.go b/e2e/agentnetwork/guardrail_test.go index e2b5d74d8..4b005cb15 100644 --- a/e2e/agentnetwork/guardrail_test.go +++ b/e2e/agentnetwork/guardrail_test.go @@ -74,7 +74,7 @@ func sendModel(ctx context.Context, t *testing.T, cl *harness.Client, endpoint, case harness.WireVertex: code, _, err = cl.Vertex(ctx, endpoint, proxyIP, pc.project, pc.region, model, "Reply with exactly: pong", "") default: - code, _, err = cl.Chat(ctx, endpoint, proxyIP, pc.kind, model, "Reply with exactly: pong", "") + code, _, err = cl.ChatPrefixed(ctx, endpoint, proxyIP, pc.pathPrefix, pc.kind, model, "Reply with exactly: pong", "") } require.NoError(t, err, "request must reach the proxy for %s", pc.name) return code diff --git a/e2e/harness/client.go b/e2e/harness/client.go index 19210349f..a352d7378 100644 --- a/e2e/harness/client.go +++ b/e2e/harness/client.go @@ -206,6 +206,17 @@ const ( // the wire shape: WireChat (OpenAI) or WireMessages (Anthropic). A non-empty // sessionID is sent as the universal x-session-id header the proxy records. func (cl *Client) Chat(ctx context.Context, endpoint, proxyIP, kind, model, prompt, sessionID string) (int, string, error) { + return cl.ChatPrefixed(ctx, endpoint, proxyIP, "", kind, model, prompt, sessionID) +} + +// ChatPrefixed is Chat with a base-URL path prefix prepended to the wire +// path, mirroring agents whose base URL carries a shape-selecting prefix that +// rides through to the upstream — e.g. Claude Code against a Kimi provider +// sets ANTHROPIC_BASE_URL=https:///anthropic so the proxy forwards +// /anthropic/v1/messages to Moonshot's Anthropic surface while the provider's +// upstream URL stays the bare https://api.moonshot.ai. Empty prefix is plain +// Chat. +func (cl *Client) ChatPrefixed(ctx context.Context, endpoint, proxyIP, pathPrefix, kind, model, prompt, sessionID string) (int, string, error) { var path, body string var headers []string switch kind { @@ -217,7 +228,7 @@ func (cl *Client) Chat(ctx context.Context, endpoint, proxyIP, kind, model, prom path = "/v1/chat/completions" body = fmt.Sprintf(`{"model":%q,"messages":[{"role":"user","content":%q}]}`, model, prompt) } - return cl.post(ctx, endpoint, proxyIP, path, body, withSessionID(headers, sessionID)) + return cl.post(ctx, endpoint, proxyIP, pathPrefix+path, body, withSessionID(headers, sessionID)) } // Vertex issues an Anthropic-on-Vertex rawPredict POST over the tunnel. Unlike