fix(otel): replace gen_ai.provider.name with gen_ai.system + engine-to-system mapping

actions/setup/js/send_otlp_span.cjs

@@ -20,6 +20,23 @@ const { getErrorMessage } = require("./error_helpers.cjs");
  * - No third-party dependencies: uses only Node built-ins + native fetch.
  */
 
+// ---------------------------------------------------------------------------
+// OTel GenAI engine-to-system mapping
+// ---------------------------------------------------------------------------
+
+/**
+ * Maps gh-aw internal engine IDs to the OTel GenAI semantic-convention
+ * `gen_ai.system` values expected by Grafana, Datadog, Honeycomb, and Sentry.
+ * Unknown engines fall back to the engine ID as-is.
+ * @type {Record<string, string>}
+ */
+const ENGINE_TO_SYSTEM_MAP = {
+  copilot: "github_models",
+  claude: "anthropic",
+  codex: "openai",
+  gemini: "google_vertex_ai",
+};
+
 // ---------------------------------------------------------------------------
 // Low-level helpers
 // ---------------------------------------------------------------------------
@@ -913,9 +930,14 @@ async function sendJobConclusionSpan(spanName, options = {}) {
     // All gh-aw agent executions are chat-style LLM completions.
     agentAttributes.push(buildAttr("gen_ai.operation.name", "chat"));
     if (model) agentAttributes.push(buildAttr("gen_ai.request.model", model));
-    // Emit gen_ai.provider.name when engineId is available; it may be omitted when
-    // engine metadata is unavailable, so this span does not guarantee full GenAI spec compliance.
-    if (engineId) agentAttributes.push(buildAttr("gen_ai.provider.name", engineId));
+    // gen_ai.system is the OTel GenAI standard attribute for the LLM system/provider.
+    // Map the gh-aw internal engine ID to the standardized value so backends can apply
+    // native GenAI dashboard detection. The original engine ID is preserved in gh-aw.engine.
+    if (engineId) {
+      const genAiSystem = ENGINE_TO_SYSTEM_MAP[engineId] || engineId;
+      agentAttributes.push(buildAttr("gen_ai.system", genAiSystem));
+      agentAttributes.push(buildAttr("gh-aw.engine", engineId));
+    }
     // gen_ai.workflow.name identifies the agentic workflow, matching the OTel spec example
     // use-cases (e.g. "multi_agent_rag", "customer_support_pipeline").
     if (workflowName) agentAttributes.push(buildAttr("gen_ai.workflow.name", workflowName));

actions/setup/js/send_otlp_span.test.cjs

@@ -1887,7 +1887,7 @@ describe("sendJobConclusionSpan", () => {
     expect(agentSpan.kind).toBe(3); // SPAN_KIND_CLIENT
   });
 
-  it("includes gen_ai.request.model, gen_ai.provider.name, gen_ai.operation.name and gen_ai.workflow.name on the agent span from aw_info.json", async () => {
+  it("includes gen_ai.request.model, gen_ai.system, gh-aw.engine, gen_ai.operation.name and gen_ai.workflow.name on the agent span from aw_info.json", async () => {
     const mockFetch = vi.fn().mockResolvedValue({ ok: true, status: 200, statusText: "OK" });
     vi.stubGlobal("fetch", mockFetch);
 
@@ -1915,11 +1915,12 @@ describe("sendJobConclusionSpan", () => {
     const attrs = Object.fromEntries(agentSpan.attributes.map(a => [a.key, a.value.stringValue ?? a.value.intValue]));
     expect(attrs["gen_ai.operation.name"]).toBe("chat");
     expect(attrs["gen_ai.request.model"]).toBe("claude-3-5-sonnet-20241022");
-    expect(attrs["gen_ai.provider.name"]).toBe("claude");
+    expect(attrs["gen_ai.system"]).toBe("anthropic");
+    expect(attrs["gh-aw.engine"]).toBe("claude");
     expect(attrs["gen_ai.workflow.name"]).toBe("otel-advisor");
   });
 
-  it("omits gen_ai.request.model, gen_ai.provider.name and gen_ai.workflow.name from the agent span when model, engine_id and workflow_name are absent", async () => {
+  it("omits gen_ai.request.model, gen_ai.system, gh-aw.engine and gen_ai.workflow.name from the agent span when model, engine_id and workflow_name are absent", async () => {
     const mockFetch = vi.fn().mockResolvedValue({ ok: true, status: 200, statusText: "OK" });
     vi.stubGlobal("fetch", mockFetch);
 
@@ -1945,7 +1946,8 @@ describe("sendJobConclusionSpan", () => {
     expect(attrs["gen_ai.operation.name"]).toBe("chat");
     const keys = agentSpan.attributes.map(a => a.key);
     expect(keys).not.toContain("gen_ai.request.model");
-    expect(keys).not.toContain("gen_ai.provider.name");
+    expect(keys).not.toContain("gen_ai.system");
+    expect(keys).not.toContain("gh-aw.engine");
     expect(keys).not.toContain("gen_ai.workflow.name");
   });