feat: implement meta-harness functionality with harness loader, presets, and validation

- Add harnessLoader.ts for loading and validating harness candidates.
- Introduce metaHarness.ts for managing meta-harness execution and context building.
- Create presets.ts to define built-in harness candidates with policies.
- Define types in types.ts for harness candidates and validation results.
- Implement tests for harness loader, harness applier, and meta-harness functionality.
- Ensure proper handling of file operations and context management in tests.

Author: lhy0718

src/cli/args.ts

@@ -5,6 +5,8 @@ export type CliAction =
   | { kind: "web"; host?: string; port?: number }
   | { kind: "compare-analysis"; runId: string; limit: number; judge: boolean }
   | { kind: "eval-harness"; runIds: string[]; limit: number; outputPath?: string; noHistory?: boolean }
+  | { kind: "evolve"; maxCycles: number; target: "skills" | "prompts" | "all"; dryRun: boolean }
+  | { kind: "meta-harness"; runs: number; nodes: ("analyze_results" | "review")[]; noApply: boolean; dryRun: boolean }
   | { kind: "help" }
   | { kind: "version" }
   | { kind: "error"; message: string };
@@ -156,10 +158,113 @@ export function resolveCliAction(args: string[]): CliAction {
     return { kind: "eval-harness", runIds, limit, outputPath, noHistory };
   }
 
+  if (first === "evolve") {
+    let maxCycles = 3;
+    let target: "skills" | "prompts" | "all" = "all";
+    let dryRun = false;
+    for (let index = 1; index < args.length; index += 1) {
+      const token = args[index];
+      if (token === "--max-cycles") {
+        const value = args[index + 1];
+        if (!value) {
+          return { kind: "error", message: "Missing value for --max-cycles." };
+        }
+        const parsed = Number(value);
+        if (!Number.isFinite(parsed) || parsed <= 0) {
+          return { kind: "error", message: `Invalid max cycle count: ${value}` };
+        }
+        maxCycles = Math.floor(parsed);
+        index += 1;
+        continue;
+      }
+      if (token === "--target") {
+        const value = args[index + 1];
+        if (!value) {
+          return { kind: "error", message: "Missing value for --target." };
+        }
+        if (value !== "skills" && value !== "prompts" && value !== "all") {
+          return {
+            kind: "error",
+            message: `Unsupported evolve target: ${value}. Expected one of skills, prompts, all.`
+          };
+        }
+        target = value;
+        index += 1;
+        continue;
+      }
+      if (token === "--dry-run") {
+        dryRun = true;
+        continue;
+      }
+      return {
+        kind: "error",
+        message: `Unsupported evolve argument: ${token}`
+      };
+    }
+    return { kind: "evolve", maxCycles, target, dryRun };
+  }
+
+  if (first === "meta-harness") {
+    let runs = 5;
+    const nodes: ("analyze_results" | "review")[] = [];
+    let noApply = false;
+    let dryRun = false;
+    for (let index = 1; index < args.length; index += 1) {
+      const token = args[index];
+      if (token === "--runs") {
+        const value = args[index + 1];
+        if (!value) {
+          return { kind: "error", message: "Missing value for --runs." };
+        }
+        const parsed = Number(value);
+        if (!Number.isFinite(parsed) || parsed <= 0) {
+          return { kind: "error", message: `Invalid run count: ${value}` };
+        }
+        runs = Math.floor(parsed);
+        index += 1;
+        continue;
+      }
+      if (token === "--node") {
+        const value = args[index + 1];
+        if (!value) {
+          return { kind: "error", message: "Missing value for --node." };
+        }
+        if (value !== "analyze_results" && value !== "review") {
+          return {
+            kind: "error",
+            message: `Unsupported meta-harness node: ${value}. Expected analyze_results or review.`
+          };
+        }
+        nodes.push(value);
+        index += 1;
+        continue;
+      }
+      if (token === "--no-apply") {
+        noApply = true;
+        continue;
+      }
+      if (token === "--dry-run") {
+        dryRun = true;
+        continue;
+      }
+      return {
+        kind: "error",
+        message: `Unsupported meta-harness argument: ${token}`
+      };
+    }
+    return {
+      kind: "meta-harness",
+      runs,
+      nodes: nodes.length > 0 ? nodes : ["analyze_results", "review"],
+      noApply,
+      dryRun
+    };
+  }
+
   return {
     kind: "error",
     message:
-      "Unsupported CLI arguments. Run `autolabos`, `autolabos web`, `autolabos compare-analysis`, `autolabos eval-harness`, or use slash commands inside the TUI."
+      "Unsupported CLI arguments. Run `autolabos`, `autolabos web`, `autolabos compare-analysis`, `autolabos eval-harness`, `autolabos evolve`, `autolabos meta-harness`, or use slash commands inside the TUI."
   };
 }
 

src/cli/evolveRun.ts

@@ -0,0 +1,16 @@
+import { runEvolveLoop } from "../core/evolution/evolveRun.js";
+
+export async function runEvolveCli(input: {
+  cwd: string;
+  maxCycles: number;
+  target: "skills" | "prompts" | "all";
+  dryRun?: boolean;
+}): Promise<void> {
+  const report = await runEvolveLoop({
+    cwd: input.cwd,
+    maxCycles: input.maxCycles,
+    target: input.target,
+    dryRun: input.dryRun
+  });
+  process.stdout.write(`${report.lines.join("\n")}\n`);
+}

src/cli/main.ts

@@ -4,6 +4,8 @@ import { resolveCliAction } from "./args.js";
 import { runAutoLabOSWebServer } from "../web/server.js";
 import { runCompareAnalysisCli } from "./compareAnalysis.js";
 import { runEvalHarnessCli } from "./evalHarness.js";
+import { runEvolveCli } from "./evolveRun.js";
+import { runMetaHarnessCli } from "./metaHarness.js";
 
 function printHelp(): void {
   process.stdout.write([
@@ -18,6 +20,8 @@ function printHelp(): void {
     "  autolabos web [--host 127.0.0.1] [--port 4317]",
     "  autolabos compare-analysis --run <run-id> [--limit 3] [--no-judge]",
     "  autolabos eval-harness [--run <run-id>] [--limit 10] [--output outputs/eval-harness/latest.json] [--no-history]",
+    "  autolabos evolve [--max-cycles 3] [--target skills|prompts|all] [--dry-run]",
+    "  autolabos meta-harness [--runs 5] [--node analyze_results|review] [--no-apply] [--dry-run]",
     "  autolabos --help",
     "  autolabos --version"
   ].join("\n") + "\n");
@@ -72,6 +76,27 @@ async function main(): Promise<void> {
     return;
   }
 
+  if (action.kind === "evolve") {
+    await runEvolveCli({
+      cwd: process.cwd(),
+      maxCycles: action.maxCycles,
+      target: action.target,
+      dryRun: action.dryRun
+    });
+    return;
+  }
+
+  if (action.kind === "meta-harness") {
+    await runMetaHarnessCli({
+      cwd: process.cwd(),
+      runs: action.runs,
+      nodes: action.nodes,
+      noApply: action.noApply,
+      dryRun: action.dryRun
+    });
+    return;
+  }
+
   await runAutoLabOSApp({
     packageName: action.kind === "run" ? action.packageName : undefined
   });

src/cli/metaHarness.ts

@@ -0,0 +1,18 @@
+import { runMetaHarness, type MetaHarnessNode } from "../core/metaHarness/metaHarness.js";
+
+export async function runMetaHarnessCli(input: {
+  cwd: string;
+  runs: number;
+  nodes: MetaHarnessNode[];
+  noApply?: boolean;
+  dryRun?: boolean;
+}): Promise<void> {
+  const result = await runMetaHarness({
+    cwd: input.cwd,
+    runs: input.runs,
+    nodes: input.nodes,
+    noApply: input.noApply,
+    dryRun: input.dryRun
+  });
+  process.stdout.write(`${result.lines.join("\n")}\n`);
+}