metaharness 0.4.0 → 0.4.2

package/templates/vertical_ai/bin/cli.js.tmpl

@@ -1,100 +1,100 @@
-#!/usr/bin/env node
-// SPDX-License-Identifier: MIT
-// Generated by metaharness — the `{{name}}` CLI entry point.
-//
-// This is plain ESM JavaScript on purpose: it runs as-is via `npx {{name}}`
-// with NO build step. `npm run build` (tsc) is only needed if you extend the
-// TypeScript in src/. The published package ships this file directly (see the
-// "bin" + "files" fields in package.json), so `npx {{name}}` works the moment
-// `npm install` has resolved @metaharness/kernel + @metaharness/host-{{host}}.
-
-import { loadKernel } from '@metaharness/kernel';
-import adapter from '@metaharness/host-{{host}}';
-
-const HARNESS_NAME = '{{name}}';
-
-/** `{{name}} init` — boot the kernel + host adapter and report status. */
-async function init() {
-  const kernel = await loadKernel();
-  const info = kernel.kernelInfo();
-  console.log(`${HARNESS_NAME} — kernel ${info.version} (${kernel.backend})`);
-  console.log(`Host adapter: ${adapter.name}`);
-  console.log(`Run \`${HARNESS_NAME} doctor\` to verify the install.`);
-  return 0;
-}
-
-/** `{{name}} doctor` — verify the install end-to-end (kernel + host resolve). */
-async function doctor() {
-  const kernel = await loadKernel();
-  const info = kernel.kernelInfo();
-  const checks = [
-    ['kernel loads', !!kernel],
-    ['kernel reports a version', typeof info.version === 'string' && info.version.length > 0],
-    ['kernel backend is native|wasm|js', ['native', 'wasm', 'js'].includes(kernel.backend)],
-    ['host adapter has a name', typeof adapter?.name === 'string' && adapter.name.length > 0],
-  ];
-  let ok = true;
-  for (const [label, pass] of checks) {
-    console.log(`${pass ? 'PASS' : 'FAIL'} ${label}`);
-    if (!pass) ok = false;
-  }
-  console.log(
-    ok
-      ? `\n${HARNESS_NAME}: all checks passed (kernel ${info.version}, ${kernel.backend} backend, host ${adapter.name})`
-      : `\n${HARNESS_NAME}: doctor found problems`,
-  );
-  return ok ? 0 : 1;
-}
-
-/**
- * Dispatch one CLI invocation. Exported (not just run on import) so a test can
- * drive it without spawning a subprocess. Returns the intended exit code.
- */
-export async function run(argv) {
-  const cmd = argv[0] ?? 'init';
-  switch (cmd) {
-    case 'init':
-      return init();
-    case 'doctor':
-      return doctor();
-    case '--version':
-    case '-v': {
-      const kernel = await loadKernel();
-      console.log(kernel.version());
-      return 0;
-    }
-    case '--help':
-    case '-h':
-      console.log(`Usage: ${HARNESS_NAME} <command>\n\n  init     boot the kernel + host adapter (default)\n  doctor   verify the install end-to-end\n  --version  print the kernel version`);
-      return 0;
-    default:
-      console.error(`Unknown command: ${cmd}. Try \`${HARNESS_NAME} --help\`.`);
-      return 2;
-  }
-}
-
-// CLI guard: execute only when invoked directly (not when imported by a test).
-// npm's bin shims pass a NON-normalized argv[1] (e.g. ".../.bin/../<pkg>/bin/cli.js"
-// on Windows) and may differ in case, so realpath BOTH sides before comparing —
-// a naive string === misses the npx/shim path and the CLI silently no-ops.
-import { fileURLToPath } from 'node:url';
-import { realpathSync } from 'node:fs';
-import { argv } from 'node:process';
-const invokedDirectly = (() => {
-  if (!argv[1]) return false;
-  try {
-    const a = realpathSync(argv[1]);
-    const b = realpathSync(fileURLToPath(import.meta.url));
-    return process.platform === 'win32' ? a.toLowerCase() === b.toLowerCase() : a === b;
-  } catch {
-    return false;
-  }
-})();
-if (invokedDirectly) {
-  run(argv.slice(2))
-    .then((code) => process.exit(code))
-    .catch((err) => {
-      console.error(err);
-      process.exit(1);
-    });
-}
+#!/usr/bin/env node
+// SPDX-License-Identifier: MIT
+// Generated by metaharness — the `{{name}}` CLI entry point.
+//
+// This is plain ESM JavaScript on purpose: it runs as-is via `npx {{name}}`
+// with NO build step. `npm run build` (tsc) is only needed if you extend the
+// TypeScript in src/. The published package ships this file directly (see the
+// "bin" + "files" fields in package.json), so `npx {{name}}` works the moment
+// `npm install` has resolved @metaharness/kernel + @metaharness/host-{{host}}.
+
+import { loadKernel } from '@metaharness/kernel';
+import adapter from '@metaharness/host-{{host}}';
+
+const HARNESS_NAME = '{{name}}';
+
+/** `{{name}} init` — boot the kernel + host adapter and report status. */
+async function init() {
+  const kernel = await loadKernel();
+  const info = kernel.kernelInfo();
+  console.log(`${HARNESS_NAME} — kernel ${info.version} (${kernel.backend})`);
+  console.log(`Host adapter: ${adapter.name}`);
+  console.log(`Run \`${HARNESS_NAME} doctor\` to verify the install.`);
+  return 0;
+}
+
+/** `{{name}} doctor` — verify the install end-to-end (kernel + host resolve). */
+async function doctor() {
+  const kernel = await loadKernel();
+  const info = kernel.kernelInfo();
+  const checks = [
+    ['kernel loads', !!kernel],
+    ['kernel reports a version', typeof info.version === 'string' && info.version.length > 0],
+    ['kernel backend is native|wasm|js', ['native', 'wasm', 'js'].includes(kernel.backend)],
+    ['host adapter has a name', typeof adapter?.name === 'string' && adapter.name.length > 0],
+  ];
+  let ok = true;
+  for (const [label, pass] of checks) {
+    console.log(`${pass ? 'PASS' : 'FAIL'} ${label}`);
+    if (!pass) ok = false;
+  }
+  console.log(
+    ok
+      ? `\n${HARNESS_NAME}: all checks passed (kernel ${info.version}, ${kernel.backend} backend, host ${adapter.name})`
+      : `\n${HARNESS_NAME}: doctor found problems`,
+  );
+  return ok ? 0 : 1;
+}
+
+/**
+ * Dispatch one CLI invocation. Exported (not just run on import) so a test can
+ * drive it without spawning a subprocess. Returns the intended exit code.
+ */
+export async function run(argv) {
+  const cmd = argv[0] ?? 'init';
+  switch (cmd) {
+    case 'init':
+      return init();
+    case 'doctor':
+      return doctor();
+    case '--version':
+    case '-v': {
+      const kernel = await loadKernel();
+      console.log(kernel.version());
+      return 0;
+    }
+    case '--help':
+    case '-h':
+      console.log(`Usage: ${HARNESS_NAME} <command>\n\n  init     boot the kernel + host adapter (default)\n  doctor   verify the install end-to-end\n  --version  print the kernel version`);
+      return 0;
+    default:
+      console.error(`Unknown command: ${cmd}. Try \`${HARNESS_NAME} --help\`.`);
+      return 2;
+  }
+}
+
+// CLI guard: execute only when invoked directly (not when imported by a test).
+// npm's bin shims pass a NON-normalized argv[1] (e.g. ".../.bin/../<pkg>/bin/cli.js"
+// on Windows) and may differ in case, so realpath BOTH sides before comparing —
+// a naive string === misses the npx/shim path and the CLI silently no-ops.
+import { fileURLToPath } from 'node:url';
+import { realpathSync } from 'node:fs';
+import { argv } from 'node:process';
+const invokedDirectly = (() => {
+  if (!argv[1]) return false;
+  try {
+    const a = realpathSync(argv[1]);
+    const b = realpathSync(fileURLToPath(import.meta.url));
+    return process.platform === 'win32' ? a.toLowerCase() === b.toLowerCase() : a === b;
+  } catch {
+    return false;
+  }
+})();
+if (invokedDirectly) {
+  run(argv.slice(2))
+    .then((code) => process.exit(code))
+    .catch((err) => {
+      console.error(err);
+      process.exit(1);
+    });
+}

package/templates/vertical_ai/manifest.json

@@ -1,106 +1,106 @@
-{
-  "id": "vertical:ai",
-  "description": "An ML pod — data-curator, trainer, evaluator, and deployer over the model lifecycle.",
-  "domain": "ai/ml-lifecycle",
-  "category": "Engineering",
-  "files": [
-    {
-      "src": "package.json.tmpl",
-      "dst": "package.json",
-      "render": true
-    },
-    {
-      "src": "CLAUDE.md.tmpl",
-      "dst": "CLAUDE.md",
-      "render": true
-    },
-    {
-      "src": "README.md.tmpl",
-      "dst": "README.md",
-      "render": true
-    },
-    {
-      "src": ".claude/settings.json.tmpl",
-      "dst": ".claude/settings.json",
-      "render": true
-    },
-    {
-      "src": ".claude-plugin/plugin.json.tmpl",
-      "dst": ".claude-plugin/plugin.json",
-      "render": true
-    },
-    {
-      "src": "src/init.ts.tmpl",
-      "dst": "src/init.ts",
-      "render": true
-    },
-    {
-      "src": "src/agents/data-curator.ts.tmpl",
-      "dst": "src/agents/data-curator.ts",
-      "render": true
-    },
-    {
-      "src": "src/agents/trainer.ts.tmpl",
-      "dst": "src/agents/trainer.ts",
-      "render": true
-    },
-    {
-      "src": "src/agents/evaluator.ts.tmpl",
-      "dst": "src/agents/evaluator.ts",
-      "render": true
-    },
-    {
-      "src": "src/agents/deployer.ts.tmpl",
-      "dst": "src/agents/deployer.ts",
-      "render": true
-    },
-    {
-      "src": ".claude/skills/eval-report/SKILL.md.tmpl",
-      "dst": ".claude/skills/eval-report/SKILL.md",
-      "render": true
-    },
-    {
-      "src": ".claude/commands/doctor.md.tmpl",
-      "dst": ".claude/commands/doctor.md",
-      "render": true
-    },
-    {
-      "src": "tsconfig.json.tmpl",
-      "dst": "tsconfig.json",
-      "render": true
-    },
-    {
-      "src": "bin/cli.js.tmpl",
-      "dst": "bin/cli.js",
-      "render": true
-    },
-    {
-      "src": "__tests__/smoke.test.ts.tmpl",
-      "dst": "__tests__/smoke.test.ts",
-      "render": true
-    }
-  ],
-  "vars": [
-    {
-      "name": "name",
-      "prompt": "Harness name (kebab-case)",
-      "validate": "^[a-z0-9-]+$"
-    },
-    {
-      "name": "description",
-      "prompt": "One-line description",
-      "default": "Curate data, train, evaluate, and deploy models"
-    },
-    {
-      "name": "host",
-      "prompt": "Host adapter",
-      "default": "claude-code",
-      "choices": [
-        "claude-code",
-        "codex",
-        "pi-dev",
-        "hermes"
-      ]
-    }
-  ]
-}
+{
+  "id": "vertical:ai",
+  "description": "An ML pod — data-curator, trainer, evaluator, and deployer over the model lifecycle.",
+  "domain": "ai/ml-lifecycle",
+  "category": "Engineering",
+  "files": [
+    {
+      "src": "package.json.tmpl",
+      "dst": "package.json",
+      "render": true
+    },
+    {
+      "src": "CLAUDE.md.tmpl",
+      "dst": "CLAUDE.md",
+      "render": true
+    },
+    {
+      "src": "README.md.tmpl",
+      "dst": "README.md",
+      "render": true
+    },
+    {
+      "src": ".claude/settings.json.tmpl",
+      "dst": ".claude/settings.json",
+      "render": true
+    },
+    {
+      "src": ".claude-plugin/plugin.json.tmpl",
+      "dst": ".claude-plugin/plugin.json",
+      "render": true
+    },
+    {
+      "src": "src/init.ts.tmpl",
+      "dst": "src/init.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/data-curator.ts.tmpl",
+      "dst": "src/agents/data-curator.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/trainer.ts.tmpl",
+      "dst": "src/agents/trainer.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/evaluator.ts.tmpl",
+      "dst": "src/agents/evaluator.ts",
+      "render": true
+    },
+    {
+      "src": "src/agents/deployer.ts.tmpl",
+      "dst": "src/agents/deployer.ts",
+      "render": true
+    },
+    {
+      "src": ".claude/skills/eval-report/SKILL.md.tmpl",
+      "dst": ".claude/skills/eval-report/SKILL.md",
+      "render": true
+    },
+    {
+      "src": ".claude/commands/doctor.md.tmpl",
+      "dst": ".claude/commands/doctor.md",
+      "render": true
+    },
+    {
+      "src": "tsconfig.json.tmpl",
+      "dst": "tsconfig.json",
+      "render": true
+    },
+    {
+      "src": "bin/cli.js.tmpl",
+      "dst": "bin/cli.js",
+      "render": true
+    },
+    {
+      "src": "__tests__/smoke.test.ts.tmpl",
+      "dst": "__tests__/smoke.test.ts",
+      "render": true
+    }
+  ],
+  "vars": [
+    {
+      "name": "name",
+      "prompt": "Harness name (kebab-case)",
+      "validate": "^[a-z0-9-]+$"
+    },
+    {
+      "name": "description",
+      "prompt": "One-line description",
+      "default": "Curate data, train, evaluate, and deploy models"
+    },
+    {
+      "name": "host",
+      "prompt": "Host adapter",
+      "default": "claude-code",
+      "choices": [
+        "claude-code",
+        "codex",
+        "pi-dev",
+        "hermes"
+      ]
+    }
+  ]
+}

package/templates/vertical_ai/package.json.tmpl

@@ -1,30 +1,30 @@
-{
-  "name": "{{name}}",
-  "version": "0.1.0",
-  "description": "{{description}}",
-  "type": "module",
-  "bin": {
-    "{{name}}": "bin/cli.js"
-  },
-  "files": ["bin/**", "dist/**", "src/**", "tsconfig.json", ".claude/**", "CLAUDE.md", "README.md", "LICENSE"],
-  "scripts": {
-    "build": "tsc",
-    "test": "vitest run",
-    "init": "node ./bin/cli.js init", "doctor": "node ./bin/cli.js doctor"
-  },
-  "dependencies": {
-    "@metaharness/kernel": "^0.1.0",
-    "@metaharness/host-{{host}}": "^0.1.0"
-  },
-  "devDependencies": {
-    "@types/node": "^20.0.0",
-    "typescript": "^5.4.0",
-    "vitest": "^3.0.0"
-  },
-  "engines": {
-    "node": ">=20.0.0"
-  },
-  "publishConfig": {
-    "access": "public"
-  }
-}
+{
+  "name": "{{name}}",
+  "version": "0.1.0",
+  "description": "{{description}}",
+  "type": "module",
+  "bin": {
+    "{{name}}": "bin/cli.js"
+  },
+  "files": ["bin/**", "dist/**", "src/**", "tsconfig.json", ".claude/**", "CLAUDE.md", "README.md", "LICENSE"],
+  "scripts": {
+    "build": "tsc",
+    "test": "vitest run",
+    "init": "node ./bin/cli.js init", "doctor": "node ./bin/cli.js doctor"
+  },
+  "dependencies": {
+    "@metaharness/kernel": "^0.1.0",
+    "@metaharness/host-{{host}}": "^0.1.0"
+  },
+  "devDependencies": {
+    "@types/node": "^20.0.0",
+    "typescript": "^5.4.0",
+    "vitest": "^3.0.0"
+  },
+  "engines": {
+    "node": ">=20.0.0"
+  },
+  "publishConfig": {
+    "access": "public"
+  }
+}

package/templates/vertical_ai/src/agents/data-curator.ts.tmpl

@@ -1,7 +1,7 @@
-// SPDX-License-Identifier: MIT
-// Data Curator agent — Builds and documents the dataset.
-
-export const SYSTEM_PROMPT = `You curate the dataset: source it, clean it, split it without leakage, and document its provenance and biases in a datasheet. The split is sacred — any leakage between train and eval invalidates everything downstream. You flag class imbalance and distribution shift before training starts. You operate inside the {{name}} harness; defer destructive actions to the user.`;
-
-export const NAME = 'data-curator';
-export const TIER = 'sonnet' as const;
+// SPDX-License-Identifier: MIT
+// Data Curator agent — Builds and documents the dataset.
+
+export const SYSTEM_PROMPT = `You curate the dataset: source it, clean it, split it without leakage, and document its provenance and biases in a datasheet. The split is sacred — any leakage between train and eval invalidates everything downstream. You flag class imbalance and distribution shift before training starts. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'data-curator';
+export const TIER = 'sonnet' as const;

package/templates/vertical_ai/src/agents/deployer.ts.tmpl

@@ -1,7 +1,7 @@
-// SPDX-License-Identifier: MIT
-// Deployer agent — Ships behind a guardrail.
-
-export const SYSTEM_PROMPT = `You deploy only models that passed the evaluator. Ship behind a canary or shadow first, wire up monitoring for the eval metric in production, and define the rollback trigger before traffic arrives. A model with no monitoring is not deployed — it is abandoned. You operate inside the {{name}} harness; defer destructive actions to the user.`;
-
-export const NAME = 'deployer';
-export const TIER = 'sonnet' as const;
+// SPDX-License-Identifier: MIT
+// Deployer agent — Ships behind a guardrail.
+
+export const SYSTEM_PROMPT = `You deploy only models that passed the evaluator. Ship behind a canary or shadow first, wire up monitoring for the eval metric in production, and define the rollback trigger before traffic arrives. A model with no monitoring is not deployed — it is abandoned. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'deployer';
+export const TIER = 'sonnet' as const;

package/templates/vertical_ai/src/agents/evaluator.ts.tmpl

@@ -1,7 +1,7 @@
-// SPDX-License-Identifier: MIT
-// Evaluator agent — The honest eval gate.
-
-export const SYSTEM_PROMPT = `You are the eval gate. Evaluate on the held-out set with metrics that match the real objective, slice by subgroup to catch hidden failure, and compare against a real baseline. You report the number that matters, including where the model is worse. No model ships on a cherry-picked metric. You operate inside the {{name}} harness; defer destructive actions to the user.`;
-
-export const NAME = 'evaluator';
-export const TIER = 'opus' as const;
+// SPDX-License-Identifier: MIT
+// Evaluator agent — The honest eval gate.
+
+export const SYSTEM_PROMPT = `You are the eval gate. Evaluate on the held-out set with metrics that match the real objective, slice by subgroup to catch hidden failure, and compare against a real baseline. You report the number that matters, including where the model is worse. No model ships on a cherry-picked metric. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'evaluator';
+export const TIER = 'opus' as const;

package/templates/vertical_ai/src/agents/trainer.ts.tmpl

@@ -1,7 +1,7 @@
-// SPDX-License-Identifier: MIT
-// Trainer agent — Runs reproducible training jobs.
-
-export const SYSTEM_PROMPT = `You run training jobs reproducibly: fixed seeds, logged hyperparameters, and every run tracked in the experiments MCP. You change one variable at a time so results are attributable. You report training/val curves and stop early on overfitting. You operate inside the {{name}} harness; defer destructive actions to the user.`;
-
-export const NAME = 'trainer';
-export const TIER = 'sonnet' as const;
+// SPDX-License-Identifier: MIT
+// Trainer agent — Runs reproducible training jobs.
+
+export const SYSTEM_PROMPT = `You run training jobs reproducibly: fixed seeds, logged hyperparameters, and every run tracked in the experiments MCP. You change one variable at a time so results are attributable. You report training/val curves and stop early on overfitting. You operate inside the {{name}} harness; defer destructive actions to the user.`;
+
+export const NAME = 'trainer';
+export const TIER = 'sonnet' as const;

package/templates/vertical_ai/src/init.ts.tmpl

@@ -1,21 +1,21 @@
-// SPDX-License-Identifier: MIT
-// Generated by create-agent-harness — your harness's `{{name}} init` entry.
-
-import { loadKernel } from '@metaharness/kernel';
-import adapter from '@metaharness/host-{{host}}';
-
-const HARNESS_NAME = '{{name}}';
-
-async function main(): Promise<number> {
-  const kernel = await loadKernel();
-  const info = kernel.kernelInfo();
-  console.log(`${HARNESS_NAME} — kernel ${info.version} (${kernel.backend})`);
-  console.log(`Host adapter: ${adapter.name}`);
-  console.log(`Run \`${HARNESS_NAME} doctor\` to verify the install.`);
-  return 0;
-}
-
-main().then(c => process.exit(c)).catch(err => {
-  console.error(err);
-  process.exit(1);
-});
+// SPDX-License-Identifier: MIT
+// Generated by create-agent-harness — your harness's `{{name}} init` entry.
+
+import { loadKernel } from '@metaharness/kernel';
+import adapter from '@metaharness/host-{{host}}';
+
+const HARNESS_NAME = '{{name}}';
+
+async function main(): Promise<number> {
+  const kernel = await loadKernel();
+  const info = kernel.kernelInfo();
+  console.log(`${HARNESS_NAME} — kernel ${info.version} (${kernel.backend})`);
+  console.log(`Host adapter: ${adapter.name}`);
+  console.log(`Run \`${HARNESS_NAME} doctor\` to verify the install.`);
+  return 0;
+}
+
+main().then(c => process.exit(c)).catch(err => {
+  console.error(err);
+  process.exit(1);
+});

package/templates/vertical_ai/tsconfig.json.tmpl

@@ -1,19 +1,19 @@
-{
-  "compilerOptions": {
-    "target": "ES2022",
-    "module": "NodeNext",
-    "moduleResolution": "NodeNext",
-    "lib": ["ES2022"],
-    "outDir": "./dist",
-    "rootDir": "./src",
-    "declaration": true,
-    "declarationMap": true,
-    "sourceMap": true,
-    "strict": true,
-    "esModuleInterop": true,
-    "skipLibCheck": true,
-    "forceConsistentCasingInFileNames": true
-  },
-  "include": ["src/**/*.ts"],
-  "exclude": ["node_modules", "dist", "__tests__"]
-}
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "NodeNext",
+    "moduleResolution": "NodeNext",
+    "lib": ["ES2022"],
+    "outDir": "./dist",
+    "rootDir": "./src",
+    "declaration": true,
+    "declarationMap": true,
+    "sourceMap": true,
+    "strict": true,
+    "esModuleInterop": true,
+    "skipLibCheck": true,
+    "forceConsistentCasingInFileNames": true
+  },
+  "include": ["src/**/*.ts"],
+  "exclude": ["node_modules", "dist", "__tests__"]
+}