Emit OTEL error signals for cancelled conclusions and success-with-errors runs

actions/setup/js/action_otlp.test.cjs

@@ -336,7 +336,7 @@ describe("action_conclusion_otlp run()", () => {
     delete process.env.GH_AW_AGENT_CONCLUSION;
   });
 
-  it("records cancelled conclusion as STATUS_CODE_OK when GH_AW_AGENT_CONCLUSION is 'cancelled'", async () => {
+  it("records cancelled conclusion as STATUS_CODE_ERROR when GH_AW_AGENT_CONCLUSION is 'cancelled'", async () => {
     process.env.OTEL_EXPORTER_OTLP_ENDPOINT = "http://localhost:14317";
     process.env.GH_AW_AGENT_CONCLUSION = "cancelled";
     let capturedBody;
@@ -349,7 +349,8 @@ describe("action_conclusion_otlp run()", () => {
 
     const payload = JSON.parse(capturedBody);
     const span = payload?.resourceSpans?.[0]?.scopeSpans?.[0]?.spans?.[0];
-    expect(span?.status?.code).toBe(1); // STATUS_CODE_OK (cancelled is not an error)
+    expect(span?.status?.code).toBe(2); // STATUS_CODE_ERROR
+    expect(span?.status?.message).toBe("agent cancelled");
     const conclusionAttr = span?.attributes?.find(a => a.key === "gh-aw.agent.conclusion");
     expect(conclusionAttr?.value?.stringValue).toBe("cancelled");
     fetchSpy.mockRestore();

actions/setup/js/send_otlp_span.cjs

@@ -700,11 +700,18 @@ async function sendJobConclusionSpan(spanName, options = {}) {
   // Values: "success", "failure", "timed_out", "cancelled", "skipped".
   const agentConclusion = process.env.GH_AW_AGENT_CONCLUSION || "";
 
-  // Mark the span as an error when the agent job failed or timed out.
+  // Mark the span as an error when the agent job failed, timed out, or was cancelled.
   const isAgentFailure = agentConclusion === "failure" || agentConclusion === "timed_out";
+  const isAgentCancelled = agentConclusion === "cancelled";
+  const isAgentNonOK = isAgentFailure || isAgentCancelled;
   // STATUS_CODE_ERROR = 2, STATUS_CODE_OK = 1
-  const statusCode = isAgentFailure ? 2 : 1;
-  let statusMessage = isAgentFailure ? `agent ${agentConclusion}` : undefined;
+  const statusCode = isAgentNonOK ? 2 : 1;
+  let statusMessage;
+  if (isAgentFailure) {
+    statusMessage = `agent ${agentConclusion}`;
+  } else if (isAgentCancelled) {
+    statusMessage = "agent cancelled";
+  }
 
   // Always read agent_output.json so output metrics are available on all outcomes.
   const agentOutput = readJSONIfExists("/tmp/gh-aw/agent_output.json") || {};
@@ -748,7 +755,7 @@ async function sendJobConclusionSpan(spanName, options = {}) {
   if (agentConclusion) {
     attributes.push(buildAttr("gh-aw.agent.conclusion", agentConclusion));
   }
-  if (isAgentFailure && errorMessages.length > 0) {
+  if (errorMessages.length > 0) {
     attributes.push(buildAttr("gh-aw.error.count", outputErrors.length));
     attributes.push(buildAttr("gh-aw.error.messages", errorMessages.join(" | ")));
   }
@@ -804,7 +811,7 @@ async function sendJobConclusionSpan(spanName, options = {}) {
   // making individual errors queryable and classifiable in backends like
   // Grafana Tempo, Honeycomb, and Datadog.
   const buildSpanEvents = eventTimeMs => {
-    if (!isAgentFailure) {
+    if (outputErrors.length === 0) {
       return [];
     }
     const errorTimeNano = toNanoString(eventTimeMs);

actions/setup/js/send_otlp_span.test.cjs

@@ -2123,6 +2123,34 @@ describe("sendJobConclusionSpan", () => {
       expect(errorMessages.value.stringValue).toBe("Rate limit exceeded | Tool call failed");
     });
 
+    it("adds gh-aw.error attributes when agent_output.json has errors on success", async () => {
+      const mockFetch = vi.fn().mockResolvedValue({ ok: true, status: 200, statusText: "OK" });
+      vi.stubGlobal("fetch", mockFetch);
+
+      process.env.OTEL_EXPORTER_OTLP_ENDPOINT = "https://traces.example.com";
+      process.env.GH_AW_AGENT_CONCLUSION = "success";
+
+      readFileSpy.mockImplementation(filePath => {
+        if (filePath === "/tmp/gh-aw/agent_output.json") {
+          return JSON.stringify({ errors: [{ message: "partial failure one" }, { message: "partial failure two" }] });
+        }
+        throw Object.assign(new Error("ENOENT"), { code: "ENOENT" });
+      });
+
+      await sendJobConclusionSpan("gh-aw.job.conclusion");
+
+      const body = JSON.parse(mockFetch.mock.calls[0][1].body);
+      const span = body.resourceSpans[0].scopeSpans[0].spans[0];
+      const attrs = span.attributes;
+      expect(span.status.code).toBe(1);
+      expect(attrs).toContainEqual({ key: "gh-aw.error.count", value: { intValue: 2 } });
+      expect(attrs).toContainEqual({ key: "gh-aw.error.messages", value: { stringValue: "partial failure one | partial failure two" } });
+      expect(span.events).toHaveLength(2);
+      expect(span.events[0].name).toBe("exception");
+      expect(span.events[0].attributes).toContainEqual({ key: "exception.type", value: { stringValue: "gh-aw.AgentError" } });
+      expect(span.events[0].attributes).toContainEqual({ key: "exception.message", value: { stringValue: "partial failure one" } });
+    });
+
     it("enriches statusMessage with the first error message on failure", async () => {
       const mockFetch = vi.fn().mockResolvedValue({ ok: true, status: 200, statusText: "OK" });
       vi.stubGlobal("fetch", mockFetch);
@@ -2165,6 +2193,21 @@ describe("sendJobConclusionSpan", () => {
       expect(span.status.message).toBe("agent timed_out: Execution exceeded 30 minute limit");
     });
 
+    it("marks cancelled conclusion spans as errors", async () => {
+      const mockFetch = vi.fn().mockResolvedValue({ ok: true, status: 200, statusText: "OK" });
+      vi.stubGlobal("fetch", mockFetch);
+
+      process.env.OTEL_EXPORTER_OTLP_ENDPOINT = "https://traces.example.com";
+      process.env.GH_AW_AGENT_CONCLUSION = "cancelled";
+
+      await sendJobConclusionSpan("gh-aw.job.conclusion");
+
+      const body = JSON.parse(mockFetch.mock.calls[0][1].body);
+      const span = body.resourceSpans[0].scopeSpans[0].spans[0];
+      expect(span.status.code).toBe(2);
+      expect(span.status.message).toBe("agent cancelled");
+    });
+
     it("caps error messages at 5 entries", async () => {
       const mockFetch = vi.fn().mockResolvedValue({ ok: true, status: 200, statusText: "OK" });
       vi.stubGlobal("fetch", mockFetch);