@mandujs/mcp 0.36.2 → 0.37.0

package/src/tools/run-tests.ts

@@ -1,424 +1,424 @@
-/**
- * MCP tool — `mandu.run.tests`
- *
- * Invokes `mandu test` as a child process via `Bun.spawn` and parses the
- * resulting Bun test output into a structured summary:
- *
- *   { passed, failed, skipped, failing_tests: [{ name, file, error }] }
- *
- * Design notes:
- *   • Input is validated against a minimal runtime schema (see `validateInput`).
- *     Bad input produces a structured `{ error, field, hint }` object — the
- *     error-handler's `isSoftErrorResult` detector will surface this as
- *     `isError: true` to MCP clients.
- *   • If no test files are discovered we return `{ passed: 0, failed: 0,
- *     skipped: 0, note: "no test files" }` without failing the caller.
- *   • The child process is spawned with a 10-minute ceiling via Promise.race —
- *     well above normal test suites but short enough that a stuck process
- *     never hangs the MCP server.
- */
-
-import type { Tool } from "@modelcontextprotocol/sdk/types.js";
-import { spawn } from "bun";
-import path from "path";
-
-// ─────────────────────────────────────────────────────────────────────────
-// Types
-// ─────────────────────────────────────────────────────────────────────────
-
-type RunTarget = "unit" | "integration" | "e2e" | "all";
-
-interface RunTestsInput {
-  target?: RunTarget;
-  filter?: string;
-  coverage?: boolean;
-}
-
-interface FailingTest {
-  name: string;
-  file?: string;
-  error?: string;
-}
-
-interface RunTestsResult {
-  target: RunTarget;
-  passed: number;
-  failed: number;
-  skipped: number;
-  duration_ms?: number;
-  failing_tests: FailingTest[];
-  exit_code: number;
-  note?: string;
-  /** Trailing 2000 chars of stdout for diagnostic context. */
-  stdout_tail?: string;
-  /** Trailing 2000 chars of stderr for diagnostic context. */
-  stderr_tail?: string;
-}
-
-// ─────────────────────────────────────────────────────────────────────────
-// Validation
-// ─────────────────────────────────────────────────────────────────────────
-
-const VALID_TARGETS = new Set<RunTarget>(["unit", "integration", "e2e", "all"]);
-const COMMAND_TIMEOUT_MS = 10 * 60_000;
-
-function validateInput(raw: Record<string, unknown>): {
-  ok: true;
-  value: Required<Pick<RunTestsInput, "target" | "coverage">> &
-    Pick<RunTestsInput, "filter">;
-} | { ok: false; error: string; field: string; hint: string } {
-  const target = raw.target ?? "all";
-  if (typeof target !== "string" || !VALID_TARGETS.has(target as RunTarget)) {
-    return {
-      ok: false,
-      error: "Invalid 'target' — expected 'unit', 'integration', 'e2e', or 'all'",
-      field: "target",
-      hint: "Omit to default to 'all'",
-    };
-  }
-
-  const filter = raw.filter;
-  if (filter !== undefined && typeof filter !== "string") {
-    return {
-      ok: false,
-      error: "'filter' must be a string",
-      field: "filter",
-      hint: "Pass a bun-test filter pattern, e.g. 'my-describe > my-case'",
-    };
-  }
-
-  const coverage = raw.coverage;
-  if (coverage !== undefined && typeof coverage !== "boolean") {
-    return {
-      ok: false,
-      error: "'coverage' must be a boolean",
-      field: "coverage",
-      hint: "Pass true to emit a coverage report",
-    };
-  }
-
-  return {
-    ok: true,
-    value: {
-      target: target as RunTarget,
-      coverage: coverage === true,
-      ...(typeof filter === "string" && filter.length > 0 ? { filter } : {}),
-    },
-  };
-}
-
-// ─────────────────────────────────────────────────────────────────────────
-// Parser — Bun test output → RunTestsResult
-// ─────────────────────────────────────────────────────────────────────────
-
-/**
- * Parse bun-test style output. Bun emits:
- *   `(pass) describe > test`
- *   `(fail) describe > test`
- *   `(skip) describe > test`
- *
- * And a trailing summary block:
- *   `N pass`
- *   `M fail`
- *   `K skipped`
- *   `Ran ... tests across ... files. [x.xxs]`
- *
- * This parser is intentionally forgiving: counts are taken from the
- * explicit summary lines when present, else derived from `(pass|fail|skip)`
- * markers.
- */
-export function parseBunTestOutput(raw: string): {
-  passed: number;
-  failed: number;
-  skipped: number;
-  duration_ms?: number;
-  failing_tests: FailingTest[];
-} {
-  const lines = raw.split(/\r?\n/);
-  let passed = 0;
-  let failed = 0;
-  let skipped = 0;
-  let duration_ms: number | undefined;
-  const failing_tests: FailingTest[] = [];
-
-  let currentFile: string | undefined;
-  let pendingFailure: FailingTest | null = null;
-
-  for (const line of lines) {
-    const trimmed = line.trim();
-
-    // Track the current file heading (e.g. "src/foo.test.ts:"):
-    const fileMatch = /^([^\s()]+\.(?:test|spec)\.(?:ts|tsx|js|jsx|mjs|cjs)):$/.exec(trimmed);
-    if (fileMatch) {
-      currentFile = fileMatch[1];
-      continue;
-    }
-
-    // `(fail) ...` → start a failing test record.
-    const failMatch = /^\(fail\)\s+(.+)$/.exec(trimmed);
-    if (failMatch) {
-      if (pendingFailure) {
-        failing_tests.push(pendingFailure);
-      }
-      pendingFailure = {
-        name: failMatch[1].trim(),
-        file: currentFile,
-      };
-      continue;
-    }
-
-    // `(skip) ...` counts as skipped but doesn't emit a record.
-    if (/^\(skip\)\s+/.test(trimmed)) {
-      continue;
-    }
-
-    // If we're inside a failure block, capture the first few non-empty
-    // lines that follow as error context.
-    if (pendingFailure && trimmed.length > 0) {
-      const isNextTestMarker = /^\(pass|fail|skip\)/.test(trimmed);
-      if (!isNextTestMarker) {
-        pendingFailure.error = pendingFailure.error
-          ? `${pendingFailure.error}\n${trimmed}`
-          : trimmed;
-        // Cap the captured error to keep payloads tight.
-        if (pendingFailure.error.length > 800) {
-          pendingFailure.error = pendingFailure.error.slice(0, 800);
-        }
-        continue;
-      }
-    }
-
-    // End-of-block marker flushes the current failure record.
-    if (pendingFailure && (trimmed.length === 0 || /^\d+\s+(pass|fail|skipped)\b/.test(trimmed))) {
-      failing_tests.push(pendingFailure);
-      pendingFailure = null;
-    }
-
-    // Totals block — authoritative if present.
-    const passMatch = /^(\d+)\s+pass\b/.exec(trimmed);
-    if (passMatch) {
-      passed = Number(passMatch[1]);
-      continue;
-    }
-    const failSumMatch = /^(\d+)\s+fail\b/.exec(trimmed);
-    if (failSumMatch) {
-      failed = Number(failSumMatch[1]);
-      continue;
-    }
-    const skipMatch = /^(\d+)\s+skipped\b/.exec(trimmed);
-    if (skipMatch) {
-      skipped = Number(skipMatch[1]);
-      continue;
-    }
-
-    // Duration: `Ran 123 tests across 10 files. [1.23s]`
-    const dur = /\[([\d.]+)s\]/.exec(trimmed);
-    if (dur && /Ran\s+\d+\s+tests/.test(trimmed)) {
-      duration_ms = Math.round(Number(dur[1]) * 1000);
-    }
-  }
-
-  if (pendingFailure) {
-    failing_tests.push(pendingFailure);
-  }
-
-  return { passed, failed, skipped, duration_ms, failing_tests };
-}
-
-// ─────────────────────────────────────────────────────────────────────────
-// Child process invocation
-// ─────────────────────────────────────────────────────────────────────────
-
-function tailString(s: string, max = 2000): string {
-  if (s.length <= max) return s;
-  return s.slice(-max);
-}
-
-function isNoTestFilesSignal(stdout: string, stderr: string): boolean {
-  // Bun reports "0 tests" or exits with a "No tests found" banner depending
-  // on version. We match on both variants.
-  const combined = `${stdout}\n${stderr}`;
-  if (/Ran\s+0\s+tests/i.test(combined)) return true;
-  if (/No tests found/i.test(combined)) return true;
-  if (/no test files/i.test(combined)) return true;
-  return false;
-}
-
-/**
- * Resolve the `mandu` CLI entry. We prefer the workspace binary
- * (`packages/cli/src/main.ts`) when running inside the monorepo, else
- * fall back to `mandu` on PATH.
- *
- * The CLI entry is invoked directly via `bun run <path>` so users get
- * the version bundled with their project without relying on global installs.
- */
-async function resolveManduCommand(projectRoot: string): Promise<string[]> {
-  // Prefer a local `.bin/mandu` if the project installed `@mandujs/cli`.
-  const localBin = path.join(projectRoot, "node_modules", ".bin", "mandu");
-  try {
-    const f = Bun.file(localBin);
-    if (await f.exists()) {
-      return ["bun", "run", localBin];
-    }
-  } catch {}
-
-  // Monorepo: packages/cli/src/main.ts is directly executable via bun.
-  const monorepoCli = path.resolve(projectRoot, "packages", "cli", "src", "main.ts");
-  try {
-    const f = Bun.file(monorepoCli);
-    if (await f.exists()) {
-      return ["bun", "run", monorepoCli];
-    }
-  } catch {}
-
-  // Fallback: rely on PATH.
-  return ["mandu"];
-}
-
-async function runProcess(
-  cmd: string[],
-  cwd: string,
-  timeoutMs: number,
-): Promise<{ stdout: string; stderr: string; exitCode: number; timedOut: boolean }> {
-  const proc = spawn(cmd, {
-    cwd,
-    stdout: "pipe",
-    stderr: "pipe",
-  });
-
-  let timedOut = false;
-  const timeoutHandle: ReturnType<typeof setTimeout> = setTimeout(() => {
-    timedOut = true;
-    try {
-      proc.kill();
-    } catch {}
-  }, timeoutMs);
-
-  try {
-    const [stdout, stderr, exitCode] = await Promise.all([
-      new Response(proc.stdout).text(),
-      new Response(proc.stderr).text(),
-      proc.exited,
-    ]);
-    return { stdout, stderr, exitCode: exitCode ?? 1, timedOut };
-  } finally {
-    clearTimeout(timeoutHandle);
-  }
-}
-
-// ─────────────────────────────────────────────────────────────────────────
-// Public handler
-// ─────────────────────────────────────────────────────────────────────────
-
-async function runManduTests(
-  projectRoot: string,
-  input: RunTestsInput,
-): Promise<RunTestsResult | { error: string; field?: string; hint?: string }> {
-  const validated = validateInput(input as Record<string, unknown>);
-  if (!validated.ok) {
-    return {
-      error: validated.error,
-      field: validated.field,
-      hint: validated.hint,
-    };
-  }
-
-  const { target, filter, coverage } = validated.value;
-
-  const base = await resolveManduCommand(projectRoot);
-  const args = [...base, "test"];
-  if (target !== "all") args.push(target);
-  if (filter) args.push("--filter", filter);
-  if (coverage) args.push("--coverage");
-
-  let proc: { stdout: string; stderr: string; exitCode: number; timedOut: boolean };
-  try {
-    proc = await runProcess(args, projectRoot, COMMAND_TIMEOUT_MS);
-  } catch (err) {
-    return {
-      error: `Failed to spawn test runner: ${err instanceof Error ? err.message : String(err)}`,
-      hint: "Verify that @mandujs/cli is installed and accessible",
-    };
-  }
-
-  // No-tests case: the caller gets a benign zeroed summary.
-  if (isNoTestFilesSignal(proc.stdout, proc.stderr) && proc.exitCode !== 0) {
-    return {
-      target,
-      passed: 0,
-      failed: 0,
-      skipped: 0,
-      failing_tests: [],
-      exit_code: proc.exitCode,
-      note: "no test files",
-      stdout_tail: tailString(proc.stdout),
-      stderr_tail: tailString(proc.stderr),
-    };
-  }
-
-  const parsed = parseBunTestOutput(`${proc.stdout}\n${proc.stderr}`);
-
-  const result: RunTestsResult = {
-    target,
-    passed: parsed.passed,
-    failed: parsed.failed,
-    skipped: parsed.skipped,
-    failing_tests: parsed.failing_tests,
-    exit_code: proc.exitCode,
-    stdout_tail: tailString(proc.stdout),
-    stderr_tail: tailString(proc.stderr),
-  };
-  if (parsed.duration_ms !== undefined) result.duration_ms = parsed.duration_ms;
-  if (proc.timedOut) result.note = "timed out";
-  if (parsed.passed === 0 && parsed.failed === 0 && parsed.skipped === 0) {
-    result.note = result.note ?? "no test files";
-  }
-
-  return result;
-}
-
-// ─────────────────────────────────────────────────────────────────────────
-// MCP tool definition + handler map
-// ─────────────────────────────────────────────────────────────────────────
-
-export const runTestsToolDefinitions: Tool[] = [
-  {
-    name: "mandu.run.tests",
-    description:
-      "Run the project's tests via `mandu test` and return a structured summary: passed / failed / skipped counts plus a list of failing tests with file and error context. Safe to call repeatedly — no writes, just spawns the child process and parses its output.",
-    annotations: {
-      readOnlyHint: true,
-    },
-    inputSchema: {
-      type: "object",
-      properties: {
-        target: {
-          type: "string",
-          enum: ["unit", "integration", "e2e", "all"],
-          description:
-            "Which test target to run (default: 'all'). Maps directly to `mandu test <target>`.",
-        },
-        filter: {
-          type: "string",
-          description:
-            "Forward `--filter <pattern>` to `bun test` — restricts to matching describe/it names.",
-        },
-        coverage: {
-          type: "boolean",
-          description: "Pass `--coverage` to emit a coverage report (default: false).",
-        },
-      },
-      required: [],
-    },
-  },
-];
-
-export function runTestsTools(projectRoot: string) {
-  const handlers: Record<string, (args: Record<string, unknown>) => Promise<unknown>> = {
-    "mandu.run.tests": async (args) => runManduTests(projectRoot, args as RunTestsInput),
-  };
-  return handlers;
-}
-
-// Re-export with canonical snake-case alias for parsimony (used by tests).
-export { parseBunTestOutput as parseTestOutput };
+/**
+ * MCP tool — `mandu.run.tests`
+ *
+ * Invokes `mandu test` as a child process via `Bun.spawn` and parses the
+ * resulting Bun test output into a structured summary:
+ *
+ *   { passed, failed, skipped, failing_tests: [{ name, file, error }] }
+ *
+ * Design notes:
+ *   • Input is validated against a minimal runtime schema (see `validateInput`).
+ *     Bad input produces a structured `{ error, field, hint }` object — the
+ *     error-handler's `isSoftErrorResult` detector will surface this as
+ *     `isError: true` to MCP clients.
+ *   • If no test files are discovered we return `{ passed: 0, failed: 0,
+ *     skipped: 0, note: "no test files" }` without failing the caller.
+ *   • The child process is spawned with a 10-minute ceiling via Promise.race —
+ *     well above normal test suites but short enough that a stuck process
+ *     never hangs the MCP server.
+ */
+
+import type { Tool } from "@modelcontextprotocol/sdk/types.js";
+import { spawn } from "bun";
+import path from "path";
+
+// ─────────────────────────────────────────────────────────────────────────
+// Types
+// ─────────────────────────────────────────────────────────────────────────
+
+type RunTarget = "unit" | "integration" | "e2e" | "all";
+
+interface RunTestsInput {
+  target?: RunTarget;
+  filter?: string;
+  coverage?: boolean;
+}
+
+interface FailingTest {
+  name: string;
+  file?: string;
+  error?: string;
+}
+
+interface RunTestsResult {
+  target: RunTarget;
+  passed: number;
+  failed: number;
+  skipped: number;
+  duration_ms?: number;
+  failing_tests: FailingTest[];
+  exit_code: number;
+  note?: string;
+  /** Trailing 2000 chars of stdout for diagnostic context. */
+  stdout_tail?: string;
+  /** Trailing 2000 chars of stderr for diagnostic context. */
+  stderr_tail?: string;
+}
+
+// ─────────────────────────────────────────────────────────────────────────
+// Validation
+// ─────────────────────────────────────────────────────────────────────────
+
+const VALID_TARGETS = new Set<RunTarget>(["unit", "integration", "e2e", "all"]);
+const COMMAND_TIMEOUT_MS = 10 * 60_000;
+
+function validateInput(raw: Record<string, unknown>): {
+  ok: true;
+  value: Required<Pick<RunTestsInput, "target" | "coverage">> &
+    Pick<RunTestsInput, "filter">;
+} | { ok: false; error: string; field: string; hint: string } {
+  const target = raw.target ?? "all";
+  if (typeof target !== "string" || !VALID_TARGETS.has(target as RunTarget)) {
+    return {
+      ok: false,
+      error: "Invalid 'target' — expected 'unit', 'integration', 'e2e', or 'all'",
+      field: "target",
+      hint: "Omit to default to 'all'",
+    };
+  }
+
+  const filter = raw.filter;
+  if (filter !== undefined && typeof filter !== "string") {
+    return {
+      ok: false,
+      error: "'filter' must be a string",
+      field: "filter",
+      hint: "Pass a bun-test filter pattern, e.g. 'my-describe > my-case'",
+    };
+  }
+
+  const coverage = raw.coverage;
+  if (coverage !== undefined && typeof coverage !== "boolean") {
+    return {
+      ok: false,
+      error: "'coverage' must be a boolean",
+      field: "coverage",
+      hint: "Pass true to emit a coverage report",
+    };
+  }
+
+  return {
+    ok: true,
+    value: {
+      target: target as RunTarget,
+      coverage: coverage === true,
+      ...(typeof filter === "string" && filter.length > 0 ? { filter } : {}),
+    },
+  };
+}
+
+// ─────────────────────────────────────────────────────────────────────────
+// Parser — Bun test output → RunTestsResult
+// ─────────────────────────────────────────────────────────────────────────
+
+/**
+ * Parse bun-test style output. Bun emits:
+ *   `(pass) describe > test`
+ *   `(fail) describe > test`
+ *   `(skip) describe > test`
+ *
+ * And a trailing summary block:
+ *   `N pass`
+ *   `M fail`
+ *   `K skipped`
+ *   `Ran ... tests across ... files. [x.xxs]`
+ *
+ * This parser is intentionally forgiving: counts are taken from the
+ * explicit summary lines when present, else derived from `(pass|fail|skip)`
+ * markers.
+ */
+export function parseBunTestOutput(raw: string): {
+  passed: number;
+  failed: number;
+  skipped: number;
+  duration_ms?: number;
+  failing_tests: FailingTest[];
+} {
+  const lines = raw.split(/\r?\n/);
+  let passed = 0;
+  let failed = 0;
+  let skipped = 0;
+  let duration_ms: number | undefined;
+  const failing_tests: FailingTest[] = [];
+
+  let currentFile: string | undefined;
+  let pendingFailure: FailingTest | null = null;
+
+  for (const line of lines) {
+    const trimmed = line.trim();
+
+    // Track the current file heading (e.g. "src/foo.test.ts:"):
+    const fileMatch = /^([^\s()]+\.(?:test|spec)\.(?:ts|tsx|js|jsx|mjs|cjs)):$/.exec(trimmed);
+    if (fileMatch) {
+      currentFile = fileMatch[1];
+      continue;
+    }
+
+    // `(fail) ...` → start a failing test record.
+    const failMatch = /^\(fail\)\s+(.+)$/.exec(trimmed);
+    if (failMatch) {
+      if (pendingFailure) {
+        failing_tests.push(pendingFailure);
+      }
+      pendingFailure = {
+        name: failMatch[1].trim(),
+        file: currentFile,
+      };
+      continue;
+    }
+
+    // `(skip) ...` counts as skipped but doesn't emit a record.
+    if (/^\(skip\)\s+/.test(trimmed)) {
+      continue;
+    }
+
+    // If we're inside a failure block, capture the first few non-empty
+    // lines that follow as error context.
+    if (pendingFailure && trimmed.length > 0) {
+      const isNextTestMarker = /^\(pass|fail|skip\)/.test(trimmed);
+      if (!isNextTestMarker) {
+        pendingFailure.error = pendingFailure.error
+          ? `${pendingFailure.error}\n${trimmed}`
+          : trimmed;
+        // Cap the captured error to keep payloads tight.
+        if (pendingFailure.error.length > 800) {
+          pendingFailure.error = pendingFailure.error.slice(0, 800);
+        }
+        continue;
+      }
+    }
+
+    // End-of-block marker flushes the current failure record.
+    if (pendingFailure && (trimmed.length === 0 || /^\d+\s+(pass|fail|skipped)\b/.test(trimmed))) {
+      failing_tests.push(pendingFailure);
+      pendingFailure = null;
+    }
+
+    // Totals block — authoritative if present.
+    const passMatch = /^(\d+)\s+pass\b/.exec(trimmed);
+    if (passMatch) {
+      passed = Number(passMatch[1]);
+      continue;
+    }
+    const failSumMatch = /^(\d+)\s+fail\b/.exec(trimmed);
+    if (failSumMatch) {
+      failed = Number(failSumMatch[1]);
+      continue;
+    }
+    const skipMatch = /^(\d+)\s+skipped\b/.exec(trimmed);
+    if (skipMatch) {
+      skipped = Number(skipMatch[1]);
+      continue;
+    }
+
+    // Duration: `Ran 123 tests across 10 files. [1.23s]`
+    const dur = /\[([\d.]+)s\]/.exec(trimmed);
+    if (dur && /Ran\s+\d+\s+tests/.test(trimmed)) {
+      duration_ms = Math.round(Number(dur[1]) * 1000);
+    }
+  }
+
+  if (pendingFailure) {
+    failing_tests.push(pendingFailure);
+  }
+
+  return { passed, failed, skipped, duration_ms, failing_tests };
+}
+
+// ─────────────────────────────────────────────────────────────────────────
+// Child process invocation
+// ─────────────────────────────────────────────────────────────────────────
+
+function tailString(s: string, max = 2000): string {
+  if (s.length <= max) return s;
+  return s.slice(-max);
+}
+
+function isNoTestFilesSignal(stdout: string, stderr: string): boolean {
+  // Bun reports "0 tests" or exits with a "No tests found" banner depending
+  // on version. We match on both variants.
+  const combined = `${stdout}\n${stderr}`;
+  if (/Ran\s+0\s+tests/i.test(combined)) return true;
+  if (/No tests found/i.test(combined)) return true;
+  if (/no test files/i.test(combined)) return true;
+  return false;
+}
+
+/**
+ * Resolve the `mandu` CLI entry. We prefer the workspace binary
+ * (`packages/cli/src/main.ts`) when running inside the monorepo, else
+ * fall back to `mandu` on PATH.
+ *
+ * The CLI entry is invoked directly via `bun run <path>` so users get
+ * the version bundled with their project without relying on global installs.
+ */
+async function resolveManduCommand(projectRoot: string): Promise<string[]> {
+  // Prefer a local `.bin/mandu` if the project installed `@mandujs/cli`.
+  const localBin = path.join(projectRoot, "node_modules", ".bin", "mandu");
+  try {
+    const f = Bun.file(localBin);
+    if (await f.exists()) {
+      return ["bun", "run", localBin];
+    }
+  } catch {}
+
+  // Monorepo: packages/cli/src/main.ts is directly executable via bun.
+  const monorepoCli = path.resolve(projectRoot, "packages", "cli", "src", "main.ts");
+  try {
+    const f = Bun.file(monorepoCli);
+    if (await f.exists()) {
+      return ["bun", "run", monorepoCli];
+    }
+  } catch {}
+
+  // Fallback: rely on PATH.
+  return ["mandu"];
+}
+
+async function runProcess(
+  cmd: string[],
+  cwd: string,
+  timeoutMs: number,
+): Promise<{ stdout: string; stderr: string; exitCode: number; timedOut: boolean }> {
+  const proc = spawn(cmd, {
+    cwd,
+    stdout: "pipe",
+    stderr: "pipe",
+  });
+
+  let timedOut = false;
+  const timeoutHandle: ReturnType<typeof setTimeout> = setTimeout(() => {
+    timedOut = true;
+    try {
+      proc.kill();
+    } catch {}
+  }, timeoutMs);
+
+  try {
+    const [stdout, stderr, exitCode] = await Promise.all([
+      new Response(proc.stdout).text(),
+      new Response(proc.stderr).text(),
+      proc.exited,
+    ]);
+    return { stdout, stderr, exitCode: exitCode ?? 1, timedOut };
+  } finally {
+    clearTimeout(timeoutHandle);
+  }
+}
+
+// ─────────────────────────────────────────────────────────────────────────
+// Public handler
+// ─────────────────────────────────────────────────────────────────────────
+
+async function runManduTests(
+  projectRoot: string,
+  input: RunTestsInput,
+): Promise<RunTestsResult | { error: string; field?: string; hint?: string }> {
+  const validated = validateInput(input as Record<string, unknown>);
+  if (!validated.ok) {
+    return {
+      error: validated.error,
+      field: validated.field,
+      hint: validated.hint,
+    };
+  }
+
+  const { target, filter, coverage } = validated.value;
+
+  const base = await resolveManduCommand(projectRoot);
+  const args = [...base, "test"];
+  if (target !== "all") args.push(target);
+  if (filter) args.push("--filter", filter);
+  if (coverage) args.push("--coverage");
+
+  let proc: { stdout: string; stderr: string; exitCode: number; timedOut: boolean };
+  try {
+    proc = await runProcess(args, projectRoot, COMMAND_TIMEOUT_MS);
+  } catch (err) {
+    return {
+      error: `Failed to spawn test runner: ${err instanceof Error ? err.message : String(err)}`,
+      hint: "Verify that @mandujs/cli is installed and accessible",
+    };
+  }
+
+  // No-tests case: the caller gets a benign zeroed summary.
+  if (isNoTestFilesSignal(proc.stdout, proc.stderr) && proc.exitCode !== 0) {
+    return {
+      target,
+      passed: 0,
+      failed: 0,
+      skipped: 0,
+      failing_tests: [],
+      exit_code: proc.exitCode,
+      note: "no test files",
+      stdout_tail: tailString(proc.stdout),
+      stderr_tail: tailString(proc.stderr),
+    };
+  }
+
+  const parsed = parseBunTestOutput(`${proc.stdout}\n${proc.stderr}`);
+
+  const result: RunTestsResult = {
+    target,
+    passed: parsed.passed,
+    failed: parsed.failed,
+    skipped: parsed.skipped,
+    failing_tests: parsed.failing_tests,
+    exit_code: proc.exitCode,
+    stdout_tail: tailString(proc.stdout),
+    stderr_tail: tailString(proc.stderr),
+  };
+  if (parsed.duration_ms !== undefined) result.duration_ms = parsed.duration_ms;
+  if (proc.timedOut) result.note = "timed out";
+  if (parsed.passed === 0 && parsed.failed === 0 && parsed.skipped === 0) {
+    result.note = result.note ?? "no test files";
+  }
+
+  return result;
+}
+
+// ─────────────────────────────────────────────────────────────────────────
+// MCP tool definition + handler map
+// ─────────────────────────────────────────────────────────────────────────
+
+export const runTestsToolDefinitions: Tool[] = [
+  {
+    name: "mandu.run.tests",
+    description:
+      "Run the project's tests via `mandu test` and return a structured summary: passed / failed / skipped counts plus a list of failing tests with file and error context. Safe to call repeatedly — no writes, just spawns the child process and parses its output.",
+    annotations: {
+      readOnlyHint: true,
+    },
+    inputSchema: {
+      type: "object",
+      properties: {
+        target: {
+          type: "string",
+          enum: ["unit", "integration", "e2e", "all"],
+          description:
+            "Which test target to run (default: 'all'). Maps directly to `mandu test <target>`.",
+        },
+        filter: {
+          type: "string",
+          description:
+            "Forward `--filter <pattern>` to `bun test` — restricts to matching describe/it names.",
+        },
+        coverage: {
+          type: "boolean",
+          description: "Pass `--coverage` to emit a coverage report (default: false).",
+        },
+      },
+      required: [],
+    },
+  },
+];
+
+export function runTestsTools(projectRoot: string) {
+  const handlers: Record<string, (args: Record<string, unknown>) => Promise<unknown>> = {
+    "mandu.run.tests": async (args) => runManduTests(projectRoot, args as RunTestsInput),
+  };
+  return handlers;
+}
+
+// Re-export with canonical snake-case alias for parsimony (used by tests).
+export { parseBunTestOutput as parseTestOutput };