fix: make set_issue_field non-fatal and fix smoke-codex field discovery

.github/workflows/smoke-codex.md

@@ -120,12 +120,16 @@ features:
 10. **Set Issue Field Testing**:
     - After creating the smoke-test issue, use `set_issue_field` exactly once on that new issue
     - Reference the created issue using the `temporary_id` declared on the `create_issue` output: set `issue_number: '#aw_smoke_issue'` in the `set_issue_field` message
-    - Discover available issue fields and choose one compatible field/value pair:
+    - **Before choosing a field**, run this bash command to discover available fields (do NOT guess field names):
+      ```
+      OWNER="${GITHUB_REPOSITORY%%/*}" && REPO="${GITHUB_REPOSITORY##*/}" && gh api graphql -f query='query($owner:String!,$repo:String!){repository(owner:$owner,name:$repo){issueFields(first:20){nodes{__typename ...on IssueFieldText{id name}...on IssueFieldNumber{id name}...on IssueFieldDate{id name}...on IssueFieldSingleSelect{id name options{id name}}}}}}' -f owner="$OWNER" -f repo="$REPO"
+      ```
+    - From the returned list, choose ONE field/value pair:
+      - date field → today's date in `YYYY-MM-DD` format (prefer this type)
       - text field → short text value
       - number field → numeric value
-      - date field → `YYYY-MM-DD`
-      - single-select field → an existing option name
-    - If no editable issue fields are available, report this test as skipped with reason
+      - single-select field → an existing option name exactly as returned
+    - If the discovery command fails or returns no fields, report this test as skipped with reason
 
 ## Output
 

actions/setup/js/safe_output_handler_manager.cjs

@@ -667,15 +667,23 @@ function isFailedProcessingResult(result) {
   return Boolean(result?.success === false && !result?.deferred && !result?.skipped && !result?.cancelled);
 }
 
-/** Types whose failures are surfaced as warnings rather than failing the safe_outputs job. */
-const REPORT_ONLY_FAILURE_TYPES = new Set(["assign_to_agent", "upload_artifact"]);
+/**
+ * Types whose failures are surfaced as warnings rather than failing the safe_outputs job.
+ * - assign_to_agent: agent assignment can fail after other safe outputs already succeeded.
+ * - upload_artifact: artifact uploads are best-effort and non-critical.
+ * - set_issue_field: project-board fields can drift (renamed/removed); a bad field write
+ *   should not sink an otherwise-successful safe-outputs batch.
+ */
+const REPORT_ONLY_FAILURE_TYPES = new Set(["assign_to_agent", "upload_artifact", "set_issue_field"]);
 
 /**
  * Determine whether a failed result should be reported without failing the safe_outputs job.
  * Agent assignment can fail after other safe outputs already succeeded, so those failures
  * are surfaced through dedicated outputs and summaries instead of failing the entire job.
  * Artifact uploads are best-effort and non-critical: a failed upload should not fail an
  * otherwise-successful run.
+ * Issue field writes target project-board fields that can drift over time; a single bad
+ * field write should not sink an otherwise-successful safe-outputs batch.
  *
  * @param {{type?: string, success?: boolean, deferred?: boolean, skipped?: boolean, cancelled?: boolean}|null|undefined} result
  * @returns {boolean}
@@ -1608,7 +1616,7 @@ async function main() {
     core.info(`Successful: ${successCount}`);
     core.info(`Failed: ${failureCount}`);
     if (reportOnlyFailureCount > 0) {
-      core.info(`Reported assignment failures: ${reportOnlyFailureCount}`);
+      core.info(`Non-fatal failures (reported only): ${reportOnlyFailureCount}`);
     }
     if (cancelledCount > 0) {
       core.info(`Cancelled (code push failed): ${cancelledCount}`);

actions/setup/js/safe_output_handler_manager.test.cjs

@@ -186,6 +186,43 @@ describe("Safe Output Handler Manager", () => {
       expect(reportOnlyFailures).toEqual([{ type: "upload_artifact", success: false, error: "artifact twirp CreateArtifact failed (400)" }]);
       expect(fatalFailures).toEqual([{ type: "create_issue", success: false, error: "Validation failed" }]);
     });
+
+    it("treats failed set_issue_field results as report-only", () => {
+      expect(
+        isReportOnlyFailureResult({
+          type: "set_issue_field",
+          success: false,
+        })
+      ).toBe(true);
+    });
+
+    it("does not treat skipped or cancelled set_issue_field results as report-only", () => {
+      expect(
+        isReportOnlyFailureResult({
+          type: "set_issue_field",
+          success: false,
+          skipped: true,
+        })
+      ).toBe(false);
+      expect(
+        isReportOnlyFailureResult({
+          type: "set_issue_field",
+          success: false,
+          cancelled: true,
+        })
+      ).toBe(false);
+    });
+
+    it("partitions set_issue_field failures as report-only, not fatal", () => {
+      const { fatalFailures, reportOnlyFailures } = partitionFailureResults([
+        { type: "set_issue_field", success: false, error: 'Issue field "Status" not found.' },
+        { type: "create_issue", success: false, error: "Validation failed" },
+        { type: "create_discussion", success: true },
+      ]);
+
+      expect(reportOnlyFailures).toEqual([{ type: "set_issue_field", success: false, error: 'Issue field "Status" not found.' }]);
+      expect(fatalFailures).toEqual([{ type: "create_issue", success: false, error: "Validation failed" }]);
+    });
   });
 
   describe("loadHandlers", () => {