feat: add structured output support via --json-schema argument (#687)

* feat: add structured output support

Add support for Agent SDK structured outputs.

New input: json_schema
Output: structured_output (JSON string)
Access: fromJSON(steps.id.outputs.structured_output).field

Docs: https://docs.claude.com/en/docs/agent-sdk/structured-outputs

* rm unused

* refactor: simplify structured outputs to use claude_args

Remove json_schema input in favor of passing --json-schema flag directly
in claude_args. This simplifies the interface by treating structured outputs
like other CLI flags (--model, --max-turns, etc.) instead of as a special
input that gets injected.

Users now specify: claude_args: '--json-schema {...}'
Instead of separate: json_schema: {...}

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* chore: remove unused json-schema util and revert version

- Remove src/utils/json-schema.ts (no longer used after refactor)
- Revert Claude Code version from 2.0.45 back to 2.0.42

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude <noreply@anthropic.com>

base-action/action.yml

@@ -75,6 +75,9 @@ outputs:
   execution_file:
     description: "Path to the JSON file containing Claude Code execution log"
     value: ${{ steps.run_claude.outputs.execution_file }}
+  structured_output:
+    description: "JSON string containing all structured output fields when --json-schema is provided in claude_args (use fromJSON() or jq to parse)"
+    value: ${{ steps.run_claude.outputs.structured_output }}
 
 runs:
   using: "composite"

base-action/src/run-claude.ts

@@ -1,7 +1,7 @@
 import * as core from "@actions/core";
 import { exec } from "child_process";
 import { promisify } from "util";
-import { unlink, writeFile, stat } from "fs/promises";
+import { unlink, writeFile, stat, readFile } from "fs/promises";
 import { createWriteStream } from "fs";
 import { spawn } from "child_process";
 import { parse as parseShellArgs } from "shell-quote";
@@ -122,9 +122,54 @@ export function prepareRunConfig(
   };
 }
 
+/**
+ * Parses structured_output from execution file and sets GitHub Action outputs
+ * Only runs if --json-schema was explicitly provided in claude_args
+ * Exported for testing
+ */
+export async function parseAndSetStructuredOutputs(
+  executionFile: string,
+): Promise<void> {
+  try {
+    const content = await readFile(executionFile, "utf-8");
+    const messages = JSON.parse(content) as {
+      type: string;
+      structured_output?: Record<string, unknown>;
+    }[];
+
+    // Search backwards - result is typically last or second-to-last message
+    const result = messages.findLast(
+      (m) => m.type === "result" && m.structured_output,
+    );
+
+    if (!result?.structured_output) {
+      throw new Error(
+        `--json-schema was provided but Claude did not return structured_output.\n` +
+          `Found ${messages.length} messages. Result exists: ${!!result}\n`,
+      );
+    }
+
+    // Set the complete structured output as a single JSON string
+    // This works around GitHub Actions limitation that composite actions can't have dynamic outputs
+    const structuredOutputJson = JSON.stringify(result.structured_output);
+    core.setOutput("structured_output", structuredOutputJson);
+    core.info(
+      `Set structured_output with ${Object.keys(result.structured_output).length} field(s)`,
+    );
+  } catch (error) {
+    if (error instanceof Error) {
+      throw error; // Preserve original error and stack trace
+    }
+    throw new Error(`Failed to parse structured outputs: ${error}`);
+  }
+}
+
 export async function runClaude(promptPath: string, options: ClaudeOptions) {
   const config = prepareRunConfig(promptPath, options);
 
+  // Detect if --json-schema is present in claude args
+  const hasJsonSchema = options.claudeArgs?.includes("--json-schema") ?? false;
+
   // Create a named pipe
   try {
     await unlink(PIPE_PATH);
@@ -308,8 +353,23 @@ export async function runClaude(promptPath: string, options: ClaudeOptions) {
       core.warning(`Failed to process output for execution metrics: ${e}`);
     }
 
-    core.setOutput("conclusion", "success");
     core.setOutput("execution_file", EXECUTION_FILE);
+
+    // Parse and set structured outputs only if user provided --json-schema in claude_args
+    if (hasJsonSchema) {
+      try {
+        await parseAndSetStructuredOutputs(EXECUTION_FILE);
+      } catch (error) {
+        const errorMessage =
+          error instanceof Error ? error.message : String(error);
+        core.setFailed(errorMessage);
+        core.setOutput("conclusion", "failure");
+        process.exit(1);
+      }
+    }
+
+    // Set conclusion to success if we reached here
+    core.setOutput("conclusion", "success");
   } else {
     core.setOutput("conclusion", "failure");
 

base-action/test/run-claude.test.ts

@@ -78,5 +78,19 @@ describe("prepareRunConfig", () => {
         "stream-json",
       ]);
     });
+
+    test("should include json-schema flag when provided", () => {
+      const options: ClaudeOptions = {
+        claudeArgs:
+          '--json-schema \'{"type":"object","properties":{"result":{"type":"boolean"}}}\'',
+      };
+
+      const prepared = prepareRunConfig("/tmp/test-prompt.txt", options);
+
+      expect(prepared.claudeArgs).toContain("--json-schema");
+      expect(prepared.claudeArgs).toContain(
+        '{"type":"object","properties":{"result":{"type":"boolean"}}}',
+      );
+    });
   });
 });

base-action/test/structured-output.test.ts

@@ -0,0 +1,158 @@
+#!/usr/bin/env bun
+
+import { describe, test, expect, afterEach, beforeEach, spyOn } from "bun:test";
+import { writeFile, unlink } from "fs/promises";
+import { tmpdir } from "os";
+import { join } from "path";
+import { parseAndSetStructuredOutputs } from "../src/run-claude";
+import * as core from "@actions/core";
+
+// Mock execution file path
+const TEST_EXECUTION_FILE = join(tmpdir(), "test-execution-output.json");
+
+// Helper to create mock execution file with structured output
+async function createMockExecutionFile(
+  structuredOutput?: Record<string, unknown>,
+  includeResult: boolean = true,
+): Promise<void> {
+  const messages: any[] = [
+    { type: "system", subtype: "init" },
+    { type: "turn", content: "test" },
+  ];
+
+  if (includeResult) {
+    messages.push({
+      type: "result",
+      cost_usd: 0.01,
+      duration_ms: 1000,
+      structured_output: structuredOutput,
+    });
+  }
+
+  await writeFile(TEST_EXECUTION_FILE, JSON.stringify(messages));
+}
+
+// Spy on core functions
+let setOutputSpy: any;
+let infoSpy: any;
+
+beforeEach(() => {
+  setOutputSpy = spyOn(core, "setOutput").mockImplementation(() => {});
+  infoSpy = spyOn(core, "info").mockImplementation(() => {});
+});
+
+describe("parseAndSetStructuredOutputs", () => {
+  afterEach(async () => {
+    setOutputSpy?.mockRestore();
+    infoSpy?.mockRestore();
+    try {
+      await unlink(TEST_EXECUTION_FILE);
+    } catch {
+      // Ignore if file doesn't exist
+    }
+  });
+
+  test("should set structured_output with valid data", async () => {
+    await createMockExecutionFile({
+      is_flaky: true,
+      confidence: 0.85,
+      summary: "Test looks flaky",
+    });
+
+    await parseAndSetStructuredOutputs(TEST_EXECUTION_FILE);
+
+    expect(setOutputSpy).toHaveBeenCalledWith(
+      "structured_output",
+      '{"is_flaky":true,"confidence":0.85,"summary":"Test looks flaky"}',
+    );
+    expect(infoSpy).toHaveBeenCalledWith(
+      "Set structured_output with 3 field(s)",
+    );
+  });
+
+  test("should handle arrays and nested objects", async () => {
+    await createMockExecutionFile({
+      items: ["a", "b", "c"],
+      config: { key: "value", nested: { deep: true } },
+    });
+
+    await parseAndSetStructuredOutputs(TEST_EXECUTION_FILE);
+
+    const callArgs = setOutputSpy.mock.calls[0];
+    expect(callArgs[0]).toBe("structured_output");
+    const parsed = JSON.parse(callArgs[1]);
+    expect(parsed).toEqual({
+      items: ["a", "b", "c"],
+      config: { key: "value", nested: { deep: true } },
+    });
+  });
+
+  test("should handle special characters in field names", async () => {
+    await createMockExecutionFile({
+      "test-result": "passed",
+      "item.count": 10,
+      "user@email": "test",
+    });
+
+    await parseAndSetStructuredOutputs(TEST_EXECUTION_FILE);
+
+    const callArgs = setOutputSpy.mock.calls[0];
+    const parsed = JSON.parse(callArgs[1]);
+    expect(parsed["test-result"]).toBe("passed");
+    expect(parsed["item.count"]).toBe(10);
+    expect(parsed["user@email"]).toBe("test");
+  });
+
+  test("should throw error when result exists but structured_output is undefined", async () => {
+    const messages = [
+      { type: "system", subtype: "init" },
+      { type: "result", cost_usd: 0.01, duration_ms: 1000 },
+    ];
+    await writeFile(TEST_EXECUTION_FILE, JSON.stringify(messages));
+
+    await expect(
+      parseAndSetStructuredOutputs(TEST_EXECUTION_FILE),
+    ).rejects.toThrow(
+      "--json-schema was provided but Claude did not return structured_output",
+    );
+  });
+
+  test("should throw error when no result message exists", async () => {
+    const messages = [
+      { type: "system", subtype: "init" },
+      { type: "turn", content: "test" },
+    ];
+    await writeFile(TEST_EXECUTION_FILE, JSON.stringify(messages));
+
+    await expect(
+      parseAndSetStructuredOutputs(TEST_EXECUTION_FILE),
+    ).rejects.toThrow(
+      "--json-schema was provided but Claude did not return structured_output",
+    );
+  });
+
+  test("should throw error with malformed JSON", async () => {
+    await writeFile(TEST_EXECUTION_FILE, "{ invalid json");
+
+    await expect(
+      parseAndSetStructuredOutputs(TEST_EXECUTION_FILE),
+    ).rejects.toThrow();
+  });
+
+  test("should throw error when file does not exist", async () => {
+    await expect(
+      parseAndSetStructuredOutputs("/nonexistent/file.json"),
+    ).rejects.toThrow();
+  });
+
+  test("should handle empty structured_output object", async () => {
+    await createMockExecutionFile({});
+
+    await parseAndSetStructuredOutputs(TEST_EXECUTION_FILE);
+
+    expect(setOutputSpy).toHaveBeenCalledWith("structured_output", "{}");
+    expect(infoSpy).toHaveBeenCalledWith(
+      "Set structured_output with 0 field(s)",
+    );
+  });
+});