agentic-flow 2.0.14 → 2.1.0

package/dist/mcp/tools/harness-tools.js

@@ -0,0 +1,90 @@
+/**
+ * Harness MCP tools (ADR-075, Track A): expose Darwin harness evolution/repair
+ * and provenance over MCP, mirroring the orchestration-side `metaharness_*` tools
+ * already present in claude-flow.
+ *
+ * Each tool validates its own args with its zod schema (so `execute` can be
+ * unit-tested by calling it with a plain object), and returns a JSON string —
+ * the FastMCP content convention used elsewhere in this server.
+ *
+ * @see docs/adr/ADR-075-metaharness-harness-evolution-and-provenance.md
+ */
+import { z } from 'zod';
+import { repair } from '../../repair/darwin-repair.js';
+import { buildManifest, verifySignedManifest } from '../../harness/provenance.js';
+const repairParams = z.object({
+    repoRoot: z.string().describe('Path to the repo to evolve/repair'),
+    generations: z.number().int().positive().optional().describe('Generations to run (default 3)'),
+    children: z.number().int().positive().optional().describe('Children per parent per generation (default 4)'),
+    seed: z.number().int().optional().describe('Deterministic seed (default 0)'),
+    sandboxMode: z
+        .enum(['real', 'mock', 'agent'])
+        .optional()
+        .describe("Evaluation substrate: 'real' (test-driven, default), 'mock' (deterministic/Docker-free), 'agent'"),
+});
+export const harnessRepairTool = {
+    name: 'harness_repair',
+    description: 'Evolve/repair a repo with Darwin Mode (ADR-074): freeze the model, evolve the harness; keep only variants that measurably improve under a frozen scorer + safety gate. Default sandbox "real" gates on the repo tests; "mock" is deterministic and Docker-free.',
+    parameters: repairParams,
+    execute: async (raw) => {
+        const a = repairParams.parse(raw);
+        const res = await repair({
+            repoRoot: a.repoRoot,
+            generations: a.generations,
+            childrenPerGeneration: a.children,
+            seed: a.seed,
+            sandboxMode: a.sandboxMode,
+        });
+        return JSON.stringify({
+            improved: res.improved,
+            winnerId: res.winnerId,
+            winnerLineage: res.winnerLineage,
+            baselineScore: res.baselineScore,
+            winnerScore: res.winnerScore,
+            deltaOverBaseline: res.deltaOverBaseline,
+            generations: res.generations,
+            variantsEvaluated: res.variantsEvaluated,
+        }, null, 2);
+    },
+};
+const manifestParams = z.object({
+    files: z.array(z.string()).min(1).describe('Files to include in the provenance manifest'),
+    createdAt: z.string().optional().describe('Optional ISO timestamp to embed in the manifest'),
+});
+export const harnessManifestTool = {
+    name: 'harness_manifest',
+    description: 'Build a provenance manifest (sha256 per file) over harness/agent config files (ADR-075). Sign it locally with the agentic-flow/harness provenance API to produce an Ed25519 witness.',
+    parameters: manifestParams,
+    execute: async (raw) => {
+        const a = manifestParams.parse(raw);
+        return JSON.stringify(buildManifest(a.files, { createdAt: a.createdAt }), null, 2);
+    },
+};
+const signedManifestSchema = z.object({
+    manifest: z.object({
+        version: z.literal(1),
+        createdAt: z.string().optional(),
+        entries: z.array(z.object({ path: z.string(), sha256: z.string() })),
+    }),
+    signature: z.string(),
+    publicKey: z.string(),
+});
+const verifyParams = z.object({
+    signed: signedManifestSchema.describe('A signed manifest bundle { manifest, signature, publicKey }'),
+});
+export const harnessVerifyTool = {
+    name: 'harness_verify',
+    description: 'Verify a signed harness manifest (Ed25519) and report on-disk drift vs the signed digests (ADR-075). Returns { signatureValid, filesIntact, drift }.',
+    parameters: verifyParams,
+    execute: async (raw) => {
+        const a = verifyParams.parse(raw);
+        return JSON.stringify(verifySignedManifest(a.signed), null, 2);
+    },
+};
+export const HARNESS_TOOLS = [harnessRepairTool, harnessManifestTool, harnessVerifyTool];
+/** Register all harness tools on a FastMCP-compatible server. */
+export function registerHarnessTools(server) {
+    for (const tool of HARNESS_TOOLS)
+        server.addTool(tool);
+}
+//# sourceMappingURL=harness-tools.js.map

package/dist/mcp/tools/harness-tools.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"harness-tools.js","sourceRoot":"","sources":["../../../src/mcp/tools/harness-tools.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,MAAM,EAAE,MAAM,+BAA+B,CAAC;AACvD,OAAO,EAAE,aAAa,EAAE,oBAAoB,EAAuB,MAAM,6BAA6B,CAAC;AAUvG,MAAM,YAAY,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5B,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,mCAAmC,CAAC;IAClE,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gCAAgC,CAAC;IAC9F,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gDAAgD,CAAC;IAC3G,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,gCAAgC,CAAC;IAC5E,WAAW,EAAE,CAAC;SACX,IAAI,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;SAC/B,QAAQ,EAAE;SACV,QAAQ,CAAC,kGAAkG,CAAC;CAChH,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,iBAAiB,GAAgB;IAC5C,IAAI,EAAE,gBAAgB;IACtB,WAAW,EACT,iQAAiQ;IACnQ,UAAU,EAAE,YAAY;IACxB,OAAO,EAAE,KAAK,EAAE,GAAY,EAAmB,EAAE;QAC/C,MAAM,CAAC,GAAG,YAAY,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAClC,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC;YACvB,QAAQ,EAAE,CAAC,CAAC,QAAQ;YACpB,WAAW,EAAE,CAAC,CAAC,WAAW;YAC1B,qBAAqB,EAAE,CAAC,CAAC,QAAQ;YACjC,IAAI,EAAE,CAAC,CAAC,IAAI;YACZ,WAAW,EAAE,CAAC,CAAC,WAAW;SAC3B,CAAC,CAAC;QACH,OAAO,IAAI,CAAC,SAAS,CACnB;YACE,QAAQ,EAAE,GAAG,CAAC,QAAQ;YACtB,QAAQ,EAAE,GAAG,CAAC,QAAQ;YACtB,aAAa,EAAE,GAAG,CAAC,aAAa;YAChC,aAAa,EAAE,GAAG,CAAC,aAAa;YAChC,WAAW,EAAE,GAAG,CAAC,WAAW;YAC5B,iBAAiB,EAAE,GAAG,CAAC,iBAAiB;YACxC,WAAW,EAAE,GAAG,CAAC,WAAW;YAC5B,iBAAiB,EAAE,GAAG,CAAC,iBAAiB;SACzC,EACD,IAAI,EACJ,CAAC,CACF,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,cAAc,GAAG,CAAC,CAAC,MAAM,CAAC;IAC9B,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,6CAA6C,CAAC;IACzF,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,iDAAiD,CAAC;CAC7F,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,mBAAmB,GAAgB;IAC9C,IAAI,EAAE,kBAAkB;IACxB,WAAW,EACT,sLAAsL;IACxL,UAAU,EAAE,cAAc;IAC1B,OAAO,EAAE,KAAK,EAAE,GAAY,EAAmB,EAAE;QAC/C,MAAM,CAAC,GAAG,cAAc,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QACpC,OAAO,IAAI,CAAC,SAAS,CAAC,aAAa,CAAC,CAAC,CAAC,KAAK,EAAE,EAAE,SAAS,EAAE,CAAC,CAAC,SAAS,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;IACrF,CAAC;CACF,CAAC;AAEF,MAAM,oBAAoB,GAAG,CAAC,CAAC,MAAM,CAAC;IACpC,QAAQ,EAAE,CAAC,CAAC,MAAM,CAAC;QACjB,OAAO,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;QACrB,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;QAChC,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;KACrE,CAAC;IACF,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE;IACrB,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE;CACtB,CAAC,CAAC;AAEH,MAAM,YAAY,GAAG,CAAC,CAAC,MAAM,CAAC;IAC5B,MAAM,EAAE,oBAAoB,CAAC,QAAQ,CAAC,6DAA6D,CAAC;CACrG,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,iBAAiB,GAAgB;IAC5C,IAAI,EAAE,gBAAgB;IACtB,WAAW,EACT,sJAAsJ;IACxJ,UAAU,EAAE,YAAY;IACxB,OAAO,EAAE,KAAK,EAAE,GAAY,EAAmB,EAAE;QAC/C,MAAM,CAAC,GAAG,YAAY,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAClC,OAAO,IAAI,CAAC,SAAS,CAAC,oBAAoB,CAAC,CAAC,CAAC,MAAwB,CAAC,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;IACnF,CAAC;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,aAAa,GAA2B,CAAC,iBAAiB,EAAE,mBAAmB,EAAE,iBAAiB,CAAC,CAAC;AAEjH,iEAAiE;AACjE,MAAM,UAAU,oBAAoB,CAAC,MAAgD;IACnF,KAAK,MAAM,IAAI,IAAI,aAAa;QAAE,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;AACzD,CAAC","sourcesContent":["/**\n * Harness MCP tools (ADR-075, Track A): expose Darwin harness evolution/repair\n * and provenance over MCP, mirroring the orchestration-side `metaharness_*` tools\n * already present in claude-flow.\n *\n * Each tool validates its own args with its zod schema (so `execute` can be\n * unit-tested by calling it with a plain object), and returns a JSON string —\n * the FastMCP content convention used elsewhere in this server.\n *\n * @see docs/adr/ADR-075-metaharness-harness-evolution-and-provenance.md\n */\n\nimport { z } from 'zod';\nimport { repair } from '../../repair/darwin-repair.js';\nimport { buildManifest, verifySignedManifest, type SignedManifest } from '../../harness/provenance.js';\n\n/** Minimal FastMCP-compatible tool descriptor (subset we use + test). */\nexport interface HarnessTool {\n  name: string;\n  description: string;\n  parameters: z.ZodTypeAny;\n  execute: (args: unknown) => Promise<string>;\n}\n\nconst repairParams = z.object({\n  repoRoot: z.string().describe('Path to the repo to evolve/repair'),\n  generations: z.number().int().positive().optional().describe('Generations to run (default 3)'),\n  children: z.number().int().positive().optional().describe('Children per parent per generation (default 4)'),\n  seed: z.number().int().optional().describe('Deterministic seed (default 0)'),\n  sandboxMode: z\n    .enum(['real', 'mock', 'agent'])\n    .optional()\n    .describe(\"Evaluation substrate: 'real' (test-driven, default), 'mock' (deterministic/Docker-free), 'agent'\"),\n});\n\nexport const harnessRepairTool: HarnessTool = {\n  name: 'harness_repair',\n  description:\n    'Evolve/repair a repo with Darwin Mode (ADR-074): freeze the model, evolve the harness; keep only variants that measurably improve under a frozen scorer + safety gate. Default sandbox \"real\" gates on the repo tests; \"mock\" is deterministic and Docker-free.',\n  parameters: repairParams,\n  execute: async (raw: unknown): Promise<string> => {\n    const a = repairParams.parse(raw);\n    const res = await repair({\n      repoRoot: a.repoRoot,\n      generations: a.generations,\n      childrenPerGeneration: a.children,\n      seed: a.seed,\n      sandboxMode: a.sandboxMode,\n    });\n    return JSON.stringify(\n      {\n        improved: res.improved,\n        winnerId: res.winnerId,\n        winnerLineage: res.winnerLineage,\n        baselineScore: res.baselineScore,\n        winnerScore: res.winnerScore,\n        deltaOverBaseline: res.deltaOverBaseline,\n        generations: res.generations,\n        variantsEvaluated: res.variantsEvaluated,\n      },\n      null,\n      2,\n    );\n  },\n};\n\nconst manifestParams = z.object({\n  files: z.array(z.string()).min(1).describe('Files to include in the provenance manifest'),\n  createdAt: z.string().optional().describe('Optional ISO timestamp to embed in the manifest'),\n});\n\nexport const harnessManifestTool: HarnessTool = {\n  name: 'harness_manifest',\n  description:\n    'Build a provenance manifest (sha256 per file) over harness/agent config files (ADR-075). Sign it locally with the agentic-flow/harness provenance API to produce an Ed25519 witness.',\n  parameters: manifestParams,\n  execute: async (raw: unknown): Promise<string> => {\n    const a = manifestParams.parse(raw);\n    return JSON.stringify(buildManifest(a.files, { createdAt: a.createdAt }), null, 2);\n  },\n};\n\nconst signedManifestSchema = z.object({\n  manifest: z.object({\n    version: z.literal(1),\n    createdAt: z.string().optional(),\n    entries: z.array(z.object({ path: z.string(), sha256: z.string() })),\n  }),\n  signature: z.string(),\n  publicKey: z.string(),\n});\n\nconst verifyParams = z.object({\n  signed: signedManifestSchema.describe('A signed manifest bundle { manifest, signature, publicKey }'),\n});\n\nexport const harnessVerifyTool: HarnessTool = {\n  name: 'harness_verify',\n  description:\n    'Verify a signed harness manifest (Ed25519) and report on-disk drift vs the signed digests (ADR-075). Returns { signatureValid, filesIntact, drift }.',\n  parameters: verifyParams,\n  execute: async (raw: unknown): Promise<string> => {\n    const a = verifyParams.parse(raw);\n    return JSON.stringify(verifySignedManifest(a.signed as SignedManifest), null, 2);\n  },\n};\n\nexport const HARNESS_TOOLS: readonly HarnessTool[] = [harnessRepairTool, harnessManifestTool, harnessVerifyTool];\n\n/** Register all harness tools on a FastMCP-compatible server. */\nexport function registerHarnessTools(server: { addTool: (tool: HarnessTool) => void }): void {\n  for (const tool of HARNESS_TOOLS) server.addTool(tool);\n}\n"]}

package/dist/optimizations/configuration-tuning.d.ts

@@ -122,7 +122,7 @@ export declare class ConfigurationTuning {
             entries: number;
         };
         topology: {
-            mode: "mesh" | "hierarchical" | "ring" | "star" | "auto";
+            mode: "auto" | "mesh" | "hierarchical" | "ring" | "star";
             changes: number;
             improvement: string;
         };

package/dist/repair/cli.d.ts

@@ -0,0 +1,24 @@
+#!/usr/bin/env node
+/**
+ * `agentic-flow-repair` — thin CLI over {@link repair} (ADR-074).
+ *
+ * Usage:
+ *   node dist/repair/cli.js <repoRoot> [--generations N] [--children N]
+ *                           [--seed N] [--mock | --agent]
+ *
+ * Default substrate is 'real' (Test-Driven Repair): the repo's own test command
+ * gates every promotion. `--mock` is the deterministic, Docker-free smoke path.
+ * The full SWE-bench-Lite TDR product (issue checkout + Docker grading) is run
+ * via Darwin's own `metaharness-darwin` CLI — see ADR-074.
+ */
+import { type SandboxMode } from './darwin-repair.js';
+interface CliArgs {
+    repoRoot: string;
+    generations: number;
+    children: number;
+    seed: number;
+    sandboxMode: SandboxMode;
+}
+export declare function parseArgs(argv: string[]): CliArgs;
+export {};
+//# sourceMappingURL=cli.d.ts.map

package/dist/repair/cli.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli.d.ts","sourceRoot":"","sources":["../../src/repair/cli.ts"],"names":[],"mappings":";AACA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAU,KAAK,WAAW,EAAE,MAAM,oBAAoB,CAAC;AAE9D,UAAU,OAAO;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,WAAW,CAAC;CAC1B;AAED,wBAAgB,SAAS,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,OAAO,CAoBjD"}

package/dist/repair/cli.js

@@ -0,0 +1,69 @@
+#!/usr/bin/env node
+/**
+ * `agentic-flow-repair` — thin CLI over {@link repair} (ADR-074).
+ *
+ * Usage:
+ *   node dist/repair/cli.js <repoRoot> [--generations N] [--children N]
+ *                           [--seed N] [--mock | --agent]
+ *
+ * Default substrate is 'real' (Test-Driven Repair): the repo's own test command
+ * gates every promotion. `--mock` is the deterministic, Docker-free smoke path.
+ * The full SWE-bench-Lite TDR product (issue checkout + Docker grading) is run
+ * via Darwin's own `metaharness-darwin` CLI — see ADR-074.
+ */
+import { repair } from './darwin-repair.js';
+export function parseArgs(argv) {
+    const args = {
+        repoRoot: '.',
+        generations: 3,
+        children: 4,
+        seed: 0,
+        sandboxMode: 'real',
+    };
+    const positional = [];
+    for (let i = 0; i < argv.length; i++) {
+        const a = argv[i];
+        if (a === '--generations')
+            args.generations = Number(argv[++i]);
+        else if (a === '--children')
+            args.children = Number(argv[++i]);
+        else if (a === '--seed')
+            args.seed = Number(argv[++i]);
+        else if (a === '--mock')
+            args.sandboxMode = 'mock';
+        else if (a === '--agent')
+            args.sandboxMode = 'agent';
+        else if (!a.startsWith('-'))
+            positional.push(a);
+    }
+    if (positional[0])
+        args.repoRoot = positional[0];
+    return args;
+}
+async function main() {
+    const a = parseArgs(process.argv.slice(2));
+    const res = await repair({
+        repoRoot: a.repoRoot,
+        generations: a.generations,
+        childrenPerGeneration: a.children,
+        seed: a.seed,
+        sandboxMode: a.sandboxMode,
+    });
+    console.log(`\nDarwin Repair — ${a.repoRoot} (${a.sandboxMode})`);
+    console.log(`  baseline finalScore : ${res.baselineScore.toFixed(3)}`);
+    if (res.winnerId) {
+        const sign = res.deltaOverBaseline >= 0 ? '+' : '';
+        console.log(`  winner              : ${res.winnerId}  (Δ ${sign}${res.deltaOverBaseline.toFixed(3)})`);
+    }
+    console.log(`  lineage             : ${res.winnerLineage.join(' → ') || '(baseline only)'}`);
+    console.log(`  variants evaluated  : ${res.variantsEvaluated} over ${res.generations} generation(s)`);
+    console.log(res.improved ? '  ✅ improved over baseline\n' : '  — no promoted improvement over baseline\n');
+}
+// Run only when invoked directly (ESM-safe; no CommonJS require.main).
+if (import.meta.url === `file://${process.argv[1]}`) {
+    main().catch((err) => {
+        console.error('repair failed:', err);
+        process.exit(1);
+    });
+}
+//# sourceMappingURL=cli.js.map

package/dist/repair/cli.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli.js","sourceRoot":"","sources":["../../src/repair/cli.ts"],"names":[],"mappings":";AACA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,MAAM,EAAoB,MAAM,oBAAoB,CAAC;AAU9D,MAAM,UAAU,SAAS,CAAC,IAAc;IACtC,MAAM,IAAI,GAAY;QACpB,QAAQ,EAAE,GAAG;QACb,WAAW,EAAE,CAAC;QACd,QAAQ,EAAE,CAAC;QACX,IAAI,EAAE,CAAC;QACP,WAAW,EAAE,MAAM;KACpB,CAAC;IACF,MAAM,UAAU,GAAa,EAAE,CAAC;IAChC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACrC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,IAAI,CAAC,KAAK,eAAe;YAAE,IAAI,CAAC,WAAW,GAAG,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;aAC3D,IAAI,CAAC,KAAK,YAAY;YAAE,IAAI,CAAC,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;aAC1D,IAAI,CAAC,KAAK,QAAQ;YAAE,IAAI,CAAC,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;aAClD,IAAI,CAAC,KAAK,QAAQ;YAAE,IAAI,CAAC,WAAW,GAAG,MAAM,CAAC;aAC9C,IAAI,CAAC,KAAK,SAAS;YAAE,IAAI,CAAC,WAAW,GAAG,OAAO,CAAC;aAChD,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC;YAAE,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClD,CAAC;IACD,IAAI,UAAU,CAAC,CAAC,CAAC;QAAE,IAAI,CAAC,QAAQ,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC;IACjD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,KAAK,UAAU,IAAI;IACjB,MAAM,CAAC,GAAG,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3C,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC;QACvB,QAAQ,EAAE,CAAC,CAAC,QAAQ;QACpB,WAAW,EAAE,CAAC,CAAC,WAAW;QAC1B,qBAAqB,EAAE,CAAC,CAAC,QAAQ;QACjC,IAAI,EAAE,CAAC,CAAC,IAAI;QACZ,WAAW,EAAE,CAAC,CAAC,WAAW;KAC3B,CAAC,CAAC;IAEH,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC,CAAC,QAAQ,KAAK,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC;IAClE,OAAO,CAAC,GAAG,CAAC,2BAA2B,GAAG,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IACvE,IAAI,GAAG,CAAC,QAAQ,EAAE,CAAC;QACjB,MAAM,IAAI,GAAG,GAAG,CAAC,iBAAiB,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QACnD,OAAO,CAAC,GAAG,CAAC,2BAA2B,GAAG,CAAC,QAAQ,QAAQ,IAAI,GAAG,GAAG,CAAC,iBAAiB,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IACzG,CAAC;IACD,OAAO,CAAC,GAAG,CAAC,2BAA2B,GAAG,CAAC,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,iBAAiB,EAAE,CAAC,CAAC;IAC7F,OAAO,CAAC,GAAG,CAAC,2BAA2B,GAAG,CAAC,iBAAiB,SAAS,GAAG,CAAC,WAAW,gBAAgB,CAAC,CAAC;IACtG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,8BAA8B,CAAC,CAAC,CAAC,6CAA6C,CAAC,CAAC;AAC7G,CAAC;AAED,uEAAuE;AACvE,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,KAAK,UAAU,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;IACpD,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;QACnB,OAAO,CAAC,KAAK,CAAC,gBAAgB,EAAE,GAAG,CAAC,CAAC;QACrC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC,CAAC,CAAC;AACL,CAAC","sourcesContent":["#!/usr/bin/env node\n/**\n * `agentic-flow-repair` — thin CLI over {@link repair} (ADR-074).\n *\n * Usage:\n *   node dist/repair/cli.js <repoRoot> [--generations N] [--children N]\n *                           [--seed N] [--mock | --agent]\n *\n * Default substrate is 'real' (Test-Driven Repair): the repo's own test command\n * gates every promotion. `--mock` is the deterministic, Docker-free smoke path.\n * The full SWE-bench-Lite TDR product (issue checkout + Docker grading) is run\n * via Darwin's own `metaharness-darwin` CLI — see ADR-074.\n */\n\nimport { repair, type SandboxMode } from './darwin-repair.js';\n\ninterface CliArgs {\n  repoRoot: string;\n  generations: number;\n  children: number;\n  seed: number;\n  sandboxMode: SandboxMode;\n}\n\nexport function parseArgs(argv: string[]): CliArgs {\n  const args: CliArgs = {\n    repoRoot: '.',\n    generations: 3,\n    children: 4,\n    seed: 0,\n    sandboxMode: 'real',\n  };\n  const positional: string[] = [];\n  for (let i = 0; i < argv.length; i++) {\n    const a = argv[i];\n    if (a === '--generations') args.generations = Number(argv[++i]);\n    else if (a === '--children') args.children = Number(argv[++i]);\n    else if (a === '--seed') args.seed = Number(argv[++i]);\n    else if (a === '--mock') args.sandboxMode = 'mock';\n    else if (a === '--agent') args.sandboxMode = 'agent';\n    else if (!a.startsWith('-')) positional.push(a);\n  }\n  if (positional[0]) args.repoRoot = positional[0];\n  return args;\n}\n\nasync function main(): Promise<void> {\n  const a = parseArgs(process.argv.slice(2));\n  const res = await repair({\n    repoRoot: a.repoRoot,\n    generations: a.generations,\n    childrenPerGeneration: a.children,\n    seed: a.seed,\n    sandboxMode: a.sandboxMode,\n  });\n\n  console.log(`\\nDarwin Repair — ${a.repoRoot} (${a.sandboxMode})`);\n  console.log(`  baseline finalScore : ${res.baselineScore.toFixed(3)}`);\n  if (res.winnerId) {\n    const sign = res.deltaOverBaseline >= 0 ? '+' : '';\n    console.log(`  winner              : ${res.winnerId}  (Δ ${sign}${res.deltaOverBaseline.toFixed(3)})`);\n  }\n  console.log(`  lineage             : ${res.winnerLineage.join(' → ') || '(baseline only)'}`);\n  console.log(`  variants evaluated  : ${res.variantsEvaluated} over ${res.generations} generation(s)`);\n  console.log(res.improved ? '  ✅ improved over baseline\\n' : '  — no promoted improvement over baseline\\n');\n}\n\n// Run only when invoked directly (ESM-safe; no CommonJS require.main).\nif (import.meta.url === `file://${process.argv[1]}`) {\n  main().catch((err) => {\n    console.error('repair failed:', err);\n    process.exit(1);\n  });\n}\n"]}

package/dist/repair/darwin-repair.d.ts

@@ -0,0 +1,80 @@
+/**
+ * Darwin Repair — autonomous harness evolution / Test-Driven Repair (ADR-074).
+ *
+ * A typed wrapper over `@metaharness/darwin`'s `evolve()`: freeze the model and
+ * evolve the harness around it (planner / context / reviewer / retry / tool /
+ * memory / score policy), keeping only variants that *measurably* improve under
+ * a frozen, reproducible scorer + safety gate.
+ *
+ * Modes (via `sandboxMode`):
+ *   - 'real'  (default) — Test-Driven Repair: the repo's own test command is the
+ *               oracle, run in Darwin's shell-free, env-scrubbed sandbox.
+ *   - 'mock'  — deterministic, surface-driven loop. No repo test, no Docker,
+ *               no network — used for hermetic smoke tests of the pipeline.
+ *   - 'agent' — runs the variant's real surface code (Node ≥ 22).
+ *
+ * NOTE: the headline SWE-bench-Lite TDR *product* (≈68.3% with-test) additionally
+ * needs the official `swebench` Docker harness for issue checkout + grading — see
+ * ADR-074 for that deployment path. This wrapper exposes the runnable `evolve()`
+ * core that does not require Docker.
+ *
+ * @see docs/adr/ADR-074-metaharness-darwin-test-driven-repair.md
+ */
+import { type EvolutionResult } from '@metaharness/darwin';
+export type SandboxMode = 'real' | 'mock' | 'agent';
+export interface RepairOptions {
+    /** Path to the repo to repair/evolve. */
+    repoRoot: string;
+    /** Work tree for Darwin artifacts. Default `<repoRoot>/.metaharness`. */
+    workRoot?: string;
+    /** Generations to run. Default 3. */
+    generations?: number;
+    /** Children produced per parent per generation. Default 4. */
+    childrenPerGeneration?: number;
+    /** Max variants evaluated concurrently. Default 4. */
+    concurrency?: number;
+    /** Minimum finalScore margin a child must beat its parent by. Default 0.05. */
+    promotionDelta?: number;
+    /** Deterministic seed. Default 0. */
+    seed?: number;
+    /** Fixed scoring tasks (the variant cannot edit these). Defaults to REPAIR_TASKS. */
+    tasks?: string[];
+    /** Evaluation substrate. Default 'real' (Test-Driven Repair). */
+    sandboxMode?: SandboxMode;
+    /** Per-variant test-command wall-clock budget (ms). Default Darwin's 120000. */
+    taskTimeoutMs?: number;
+}
+export interface RepairResult {
+    /** True iff a child beat the baseline and was promoted. */
+    improved: boolean;
+    /** Winner variant id (lineage tail), or null when nothing beat the baseline. */
+    winnerId: string | null;
+    /** baseline → … → winner ids. */
+    winnerLineage: string[];
+    /** Baseline finalScore (0 if unevaluated). */
+    baselineScore: number;
+    /** Winner finalScore, or null. */
+    winnerScore: number | null;
+    /** winnerScore − baselineScore. */
+    deltaOverBaseline: number;
+    generations: number;
+    /** Total variants in the archive (baseline + descendants). */
+    variantsEvaluated: number;
+    /** Full Darwin result for callers that need the archive/traces. */
+    raw: EvolutionResult;
+}
+/** Default scoring tasks for a repair run. */
+export declare const REPAIR_TASKS: readonly string[];
+/**
+ * Run an autonomous repair/evolution pass over a repo and return a friendly
+ * summary. Defaults to Test-Driven Repair ('real' sandbox): the repo's own
+ * tests gate every promotion.
+ */
+export declare function repair(opts: RepairOptions): Promise<RepairResult>;
+/** Reusable repair runner with bound defaults (e.g. a fixed sandbox mode). */
+export declare class DarwinRepair {
+    private readonly defaults;
+    constructor(defaults?: Partial<RepairOptions>);
+    repair(opts: RepairOptions): Promise<RepairResult>;
+}
+//# sourceMappingURL=darwin-repair.d.ts.map

package/dist/repair/darwin-repair.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"darwin-repair.d.ts","sourceRoot":"","sources":["../../src/repair/darwin-repair.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAGH,OAAO,EAAgC,KAAK,eAAe,EAAsB,MAAM,qBAAqB,CAAC;AAE7G,MAAM,MAAM,WAAW,GAAG,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC;AAEpD,MAAM,WAAW,aAAa;IAC5B,yCAAyC;IACzC,QAAQ,EAAE,MAAM,CAAC;IACjB,yEAAyE;IACzE,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,qCAAqC;IACrC,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,8DAA8D;IAC9D,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,sDAAsD;IACtD,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,+EAA+E;IAC/E,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,qCAAqC;IACrC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,qFAAqF;IACrF,KAAK,CAAC,EAAE,MAAM,EAAE,CAAC;IACjB,iEAAiE;IACjE,WAAW,CAAC,EAAE,WAAW,CAAC;IAC1B,gFAAgF;IAChF,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,YAAY;IAC3B,2DAA2D;IAC3D,QAAQ,EAAE,OAAO,CAAC;IAClB,gFAAgF;IAChF,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,iCAAiC;IACjC,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,8CAA8C;IAC9C,aAAa,EAAE,MAAM,CAAC;IACtB,kCAAkC;IAClC,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,mCAAmC;IACnC,iBAAiB,EAAE,MAAM,CAAC;IAC1B,WAAW,EAAE,MAAM,CAAC;IACpB,8DAA8D;IAC9D,iBAAiB,EAAE,MAAM,CAAC;IAC1B,mEAAmE;IACnE,GAAG,EAAE,eAAe,CAAC;CACtB;AAED,8CAA8C;AAC9C,eAAO,MAAM,YAAY,EAAE,SAAS,MAAM,EAIzC,CAAC;AAMF;;;;GAIG;AACH,wBAAsB,MAAM,CAAC,IAAI,EAAE,aAAa,GAAG,OAAO,CAAC,YAAY,CAAC,CAkCvE;AAED,8EAA8E;AAC9E,qBAAa,YAAY;IACX,OAAO,CAAC,QAAQ,CAAC,QAAQ;gBAAR,QAAQ,GAAE,OAAO,CAAC,aAAa,CAAM;IAElE,MAAM,CAAC,IAAI,EAAE,aAAa,GAAG,OAAO,CAAC,YAAY,CAAC;CAGnD"}

package/dist/repair/darwin-repair.js

@@ -0,0 +1,81 @@
+/**
+ * Darwin Repair — autonomous harness evolution / Test-Driven Repair (ADR-074).
+ *
+ * A typed wrapper over `@metaharness/darwin`'s `evolve()`: freeze the model and
+ * evolve the harness around it (planner / context / reviewer / retry / tool /
+ * memory / score policy), keeping only variants that *measurably* improve under
+ * a frozen, reproducible scorer + safety gate.
+ *
+ * Modes (via `sandboxMode`):
+ *   - 'real'  (default) — Test-Driven Repair: the repo's own test command is the
+ *               oracle, run in Darwin's shell-free, env-scrubbed sandbox.
+ *   - 'mock'  — deterministic, surface-driven loop. No repo test, no Docker,
+ *               no network — used for hermetic smoke tests of the pipeline.
+ *   - 'agent' — runs the variant's real surface code (Node ≥ 22).
+ *
+ * NOTE: the headline SWE-bench-Lite TDR *product* (≈68.3% with-test) additionally
+ * needs the official `swebench` Docker harness for issue checkout + grading — see
+ * ADR-074 for that deployment path. This wrapper exposes the runnable `evolve()`
+ * core that does not require Docker.
+ *
+ * @see docs/adr/ADR-074-metaharness-darwin-test-driven-repair.md
+ */
+import { resolve } from 'node:path';
+import { evolve } from '@metaharness/darwin';
+/** Default scoring tasks for a repair run. */
+export const REPAIR_TASKS = [
+    'run repository test suite',
+    'verify generated harness safety',
+    'check trace quality',
+];
+function finalScoreOf(record) {
+    return record?.score?.finalScore ?? null;
+}
+/**
+ * Run an autonomous repair/evolution pass over a repo and return a friendly
+ * summary. Defaults to Test-Driven Repair ('real' sandbox): the repo's own
+ * tests gate every promotion.
+ */
+export async function repair(opts) {
+    const repoRoot = resolve(opts.repoRoot);
+    const config = {
+        repoRoot,
+        workRoot: opts.workRoot ? resolve(opts.workRoot) : resolve(repoRoot, '.metaharness'),
+        generations: opts.generations ?? 3,
+        childrenPerGeneration: opts.childrenPerGeneration ?? 4,
+        tasks: opts.tasks ? [...opts.tasks] : [...REPAIR_TASKS],
+        promotionDelta: opts.promotionDelta ?? 0.05,
+        concurrency: opts.concurrency ?? 4,
+        seed: opts.seed ?? 0,
+        sandboxMode: opts.sandboxMode ?? 'real',
+        ...(opts.taskTimeoutMs ? { taskTimeoutMs: opts.taskTimeoutMs } : {}),
+    };
+    const result = await evolve(config);
+    const baselineScore = finalScoreOf(result.baseline) ?? 0;
+    const winnerScore = finalScoreOf(result.winner);
+    const delta = (winnerScore ?? baselineScore) - baselineScore;
+    // winnerLineage is baseline → … → winner, so the tail is the winner id.
+    const winnerId = result.winner ? result.winnerLineage[result.winnerLineage.length - 1] ?? null : null;
+    return {
+        improved: result.winner != null && delta > 0,
+        winnerId,
+        winnerLineage: result.winnerLineage,
+        baselineScore,
+        winnerScore,
+        deltaOverBaseline: delta,
+        generations: result.generations,
+        variantsEvaluated: result.records.length,
+        raw: result,
+    };
+}
+/** Reusable repair runner with bound defaults (e.g. a fixed sandbox mode). */
+export class DarwinRepair {
+    defaults;
+    constructor(defaults = {}) {
+        this.defaults = defaults;
+    }
+    repair(opts) {
+        return repair({ ...this.defaults, ...opts });
+    }
+}
+//# sourceMappingURL=darwin-repair.js.map

package/dist/repair/darwin-repair.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"darwin-repair.js","sourceRoot":"","sources":["../../src/repair/darwin-repair.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,MAAM,EAAkE,MAAM,qBAAqB,CAAC;AA+C7G,8CAA8C;AAC9C,MAAM,CAAC,MAAM,YAAY,GAAsB;IAC7C,2BAA2B;IAC3B,iCAAiC;IACjC,qBAAqB;CACtB,CAAC;AAEF,SAAS,YAAY,CAAC,MAA4B;IAChD,OAAO,MAAM,EAAE,KAAK,EAAE,UAAU,IAAI,IAAI,CAAC;AAC3C,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,MAAM,CAAC,IAAmB;IAC9C,MAAM,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACxC,MAAM,MAAM,GAAoB;QAC9B,QAAQ;QACR,QAAQ,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,EAAE,cAAc,CAAC;QACpF,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,CAAC;QAClC,qBAAqB,EAAE,IAAI,CAAC,qBAAqB,IAAI,CAAC;QACtD,KAAK,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,YAAY,CAAC;QACvD,cAAc,EAAE,IAAI,CAAC,cAAc,IAAI,IAAI;QAC3C,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,CAAC;QAClC,IAAI,EAAE,IAAI,CAAC,IAAI,IAAI,CAAC;QACpB,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,MAAM;QACvC,GAAG,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,IAAI,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACrE,CAAC;IAEF,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,MAAM,CAAC,CAAC;IAEpC,MAAM,aAAa,GAAG,YAAY,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IACzD,MAAM,WAAW,GAAG,YAAY,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAChD,MAAM,KAAK,GAAG,CAAC,WAAW,IAAI,aAAa,CAAC,GAAG,aAAa,CAAC;IAC7D,wEAAwE;IACxE,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;IAEtG,OAAO;QACL,QAAQ,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI,IAAI,KAAK,GAAG,CAAC;QAC5C,QAAQ;QACR,aAAa,EAAE,MAAM,CAAC,aAAa;QACnC,aAAa;QACb,WAAW;QACX,iBAAiB,EAAE,KAAK;QACxB,WAAW,EAAE,MAAM,CAAC,WAAW;QAC/B,iBAAiB,EAAE,MAAM,CAAC,OAAO,CAAC,MAAM;QACxC,GAAG,EAAE,MAAM;KACZ,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,MAAM,OAAO,YAAY;IACM;IAA7B,YAA6B,WAAmC,EAAE;QAArC,aAAQ,GAAR,QAAQ,CAA6B;IAAG,CAAC;IAEtE,MAAM,CAAC,IAAmB;QACxB,OAAO,MAAM,CAAC,EAAE,GAAG,IAAI,CAAC,QAAQ,EAAE,GAAG,IAAI,EAAE,CAAC,CAAC;IAC/C,CAAC;CACF","sourcesContent":["/**\n * Darwin Repair — autonomous harness evolution / Test-Driven Repair (ADR-074).\n *\n * A typed wrapper over `@metaharness/darwin`'s `evolve()`: freeze the model and\n * evolve the harness around it (planner / context / reviewer / retry / tool /\n * memory / score policy), keeping only variants that *measurably* improve under\n * a frozen, reproducible scorer + safety gate.\n *\n * Modes (via `sandboxMode`):\n *   - 'real'  (default) — Test-Driven Repair: the repo's own test command is the\n *               oracle, run in Darwin's shell-free, env-scrubbed sandbox.\n *   - 'mock'  — deterministic, surface-driven loop. No repo test, no Docker,\n *               no network — used for hermetic smoke tests of the pipeline.\n *   - 'agent' — runs the variant's real surface code (Node ≥ 22).\n *\n * NOTE: the headline SWE-bench-Lite TDR *product* (≈68.3% with-test) additionally\n * needs the official `swebench` Docker harness for issue checkout + grading — see\n * ADR-074 for that deployment path. This wrapper exposes the runnable `evolve()`\n * core that does not require Docker.\n *\n * @see docs/adr/ADR-074-metaharness-darwin-test-driven-repair.md\n */\n\nimport { resolve } from 'node:path';\nimport { evolve, type EvolutionConfig, type EvolutionResult, type ArchiveRecord } from '@metaharness/darwin';\n\nexport type SandboxMode = 'real' | 'mock' | 'agent';\n\nexport interface RepairOptions {\n  /** Path to the repo to repair/evolve. */\n  repoRoot: string;\n  /** Work tree for Darwin artifacts. Default `<repoRoot>/.metaharness`. */\n  workRoot?: string;\n  /** Generations to run. Default 3. */\n  generations?: number;\n  /** Children produced per parent per generation. Default 4. */\n  childrenPerGeneration?: number;\n  /** Max variants evaluated concurrently. Default 4. */\n  concurrency?: number;\n  /** Minimum finalScore margin a child must beat its parent by. Default 0.05. */\n  promotionDelta?: number;\n  /** Deterministic seed. Default 0. */\n  seed?: number;\n  /** Fixed scoring tasks (the variant cannot edit these). Defaults to REPAIR_TASKS. */\n  tasks?: string[];\n  /** Evaluation substrate. Default 'real' (Test-Driven Repair). */\n  sandboxMode?: SandboxMode;\n  /** Per-variant test-command wall-clock budget (ms). Default Darwin's 120000. */\n  taskTimeoutMs?: number;\n}\n\nexport interface RepairResult {\n  /** True iff a child beat the baseline and was promoted. */\n  improved: boolean;\n  /** Winner variant id (lineage tail), or null when nothing beat the baseline. */\n  winnerId: string | null;\n  /** baseline → … → winner ids. */\n  winnerLineage: string[];\n  /** Baseline finalScore (0 if unevaluated). */\n  baselineScore: number;\n  /** Winner finalScore, or null. */\n  winnerScore: number | null;\n  /** winnerScore − baselineScore. */\n  deltaOverBaseline: number;\n  generations: number;\n  /** Total variants in the archive (baseline + descendants). */\n  variantsEvaluated: number;\n  /** Full Darwin result for callers that need the archive/traces. */\n  raw: EvolutionResult;\n}\n\n/** Default scoring tasks for a repair run. */\nexport const REPAIR_TASKS: readonly string[] = [\n  'run repository test suite',\n  'verify generated harness safety',\n  'check trace quality',\n];\n\nfunction finalScoreOf(record: ArchiveRecord | null): number | null {\n  return record?.score?.finalScore ?? null;\n}\n\n/**\n * Run an autonomous repair/evolution pass over a repo and return a friendly\n * summary. Defaults to Test-Driven Repair ('real' sandbox): the repo's own\n * tests gate every promotion.\n */\nexport async function repair(opts: RepairOptions): Promise<RepairResult> {\n  const repoRoot = resolve(opts.repoRoot);\n  const config: EvolutionConfig = {\n    repoRoot,\n    workRoot: opts.workRoot ? resolve(opts.workRoot) : resolve(repoRoot, '.metaharness'),\n    generations: opts.generations ?? 3,\n    childrenPerGeneration: opts.childrenPerGeneration ?? 4,\n    tasks: opts.tasks ? [...opts.tasks] : [...REPAIR_TASKS],\n    promotionDelta: opts.promotionDelta ?? 0.05,\n    concurrency: opts.concurrency ?? 4,\n    seed: opts.seed ?? 0,\n    sandboxMode: opts.sandboxMode ?? 'real',\n    ...(opts.taskTimeoutMs ? { taskTimeoutMs: opts.taskTimeoutMs } : {}),\n  };\n\n  const result = await evolve(config);\n\n  const baselineScore = finalScoreOf(result.baseline) ?? 0;\n  const winnerScore = finalScoreOf(result.winner);\n  const delta = (winnerScore ?? baselineScore) - baselineScore;\n  // winnerLineage is baseline → … → winner, so the tail is the winner id.\n  const winnerId = result.winner ? result.winnerLineage[result.winnerLineage.length - 1] ?? null : null;\n\n  return {\n    improved: result.winner != null && delta > 0,\n    winnerId,\n    winnerLineage: result.winnerLineage,\n    baselineScore,\n    winnerScore,\n    deltaOverBaseline: delta,\n    generations: result.generations,\n    variantsEvaluated: result.records.length,\n    raw: result,\n  };\n}\n\n/** Reusable repair runner with bound defaults (e.g. a fixed sandbox mode). */\nexport class DarwinRepair {\n  constructor(private readonly defaults: Partial<RepairOptions> = {}) {}\n\n  repair(opts: RepairOptions): Promise<RepairResult> {\n    return repair({ ...this.defaults, ...opts });\n  }\n}\n"]}

package/dist/router/cost-optimal-router.d.ts

@@ -0,0 +1,89 @@
+/**
+ * Cost-Optimal Model Router (ADR-073)
+ *
+ * Routes each query to the *cheapest model predicted to clear a quality bar*,
+ * learned from eval logs — the productized DRACO Phase-2 finding. Wraps
+ * `@metaharness/router` (dependency-free k-NN / kernel-ridge, optional native
+ * FastGRNN via the already-present `@ruvector/tiny-dancer`).
+ *
+ * This is additive to `ModelRouter`'s existing config-rule routing: it selects
+ * a *model* by predicted cost-quality rather than a provider by static rule.
+ * It degrades gracefully — with no labelled examples it falls back to the
+ * best-predicted candidate (effectively the caller's prior behavior), so a
+ * cold start never breaks routing.
+ *
+ * @see docs/adr/ADR-073-metaharness-router-cost-optimal-model-routing.md
+ */
+import { type RouteResult, type RouterCandidate } from '@metaharness/router';
+import type { ProviderType } from './types.js';
+/** One labelled observation: a query embedding and the quality a model achieved on it. */
+export interface RoutingExample {
+    embedding: number[];
+    quality: number;
+}
+/** A flat training row: query embedding → quality each model achieved on that query. */
+export interface RoutingDatasetRow {
+    embedding: number[];
+    scores: Record<string, number>;
+}
+/** Maps a router model id (e.g. "anthropic/claude-haiku-4.5") to a concrete provider + model. */
+export interface ModelBinding {
+    provider: ProviderType;
+    model: string;
+}
+export interface CostOptimalRouterConfig {
+    /**
+     * Quality bar (0..1). When set, route() returns the CHEAPEST candidate
+     * predicted to clear it; when none clear it, the best-predicted; when unset,
+     * always the best-predicted.
+     */
+    qualityBar?: number;
+    /** k-NN neighbours used to predict quality on a new query (default 5). */
+    k?: number;
+    /**
+     * Optional explicit model id → provider/model bindings. When a routed id is
+     * absent here, it is parsed as `"<provider>/<model>"` (the OpenRouter-style
+     * convention), falling back to treating the whole id as the model name on the
+     * default provider.
+     */
+    modelMap?: Record<string, ModelBinding>;
+    /** Provider to assume when a model id has no `"<provider>/"` prefix. */
+    defaultProvider?: ProviderType;
+}
+/** A routing decision: the underlying RouteResult plus the resolved provider/model. */
+export interface CostOptimalDecision extends RouteResult {
+    provider: ProviderType;
+    model: string;
+}
+/**
+ * Parse a router model id into a provider/model binding. `"anthropic/claude-..."`
+ * → `{ provider: 'anthropic', model: 'claude-...' }`. Ids without a known
+ * provider prefix bind to `defaultProvider` with the id as the model name.
+ */
+export declare function parseModelId(id: string, modelMap: Record<string, ModelBinding> | undefined, defaultProvider: ProviderType): ModelBinding;
+export declare class CostOptimalRouter {
+    private readonly router;
+    private readonly modelMap?;
+    private readonly defaultProvider;
+    private constructor();
+    /**
+     * Build from explicit candidates (each with its own labelled examples + price).
+     * Use when you already have per-model example sets.
+     */
+    static fromCandidates(candidates: RouterCandidate[], config?: CostOptimalRouterConfig): CostOptimalRouter;
+    /**
+     * Build from a flat routing dataset — rows of (query embedding → quality each
+     * model achieved) plus a per-model price table. This is the shape eval logs
+     * and the DRACO benchmark emit, and the same shape that seeds a native
+     * tiny-dancer trainer.
+     */
+    static fromDataset(rows: RoutingDatasetRow[], prices: Record<string, number>, config?: CostOptimalRouterConfig): CostOptimalRouter;
+    /** Route a pre-embedded query to the cost-optimal model + provider. */
+    route(queryEmbedding: number[]): CostOptimalDecision;
+    /**
+     * Route using an injected embedder (e.g. the ONNX/ruvector embedding service).
+     * Keeps this module decoupled from any specific embedding backend.
+     */
+    routeText(text: string, embed: (text: string) => Promise<number[]> | number[]): Promise<CostOptimalDecision>;
+}
+//# sourceMappingURL=cost-optimal-router.d.ts.map

package/dist/router/cost-optimal-router.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cost-optimal-router.d.ts","sourceRoot":"","sources":["../../src/router/cost-optimal-router.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAU,KAAK,WAAW,EAAE,KAAK,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACrF,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAE/C,0FAA0F;AAC1F,MAAM,WAAW,cAAc;IAC7B,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,wFAAwF;AACxF,MAAM,WAAW,iBAAiB;IAChC,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAChC;AAED,iGAAiG;AACjG,MAAM,WAAW,YAAY;IAC3B,QAAQ,EAAE,YAAY,CAAC;IACvB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,uBAAuB;IACtC;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,0EAA0E;IAC1E,CAAC,CAAC,EAAE,MAAM,CAAC;IACX;;;;;OAKG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;IACxC,wEAAwE;IACxE,eAAe,CAAC,EAAE,YAAY,CAAC;CAChC;AAED,uFAAuF;AACvF,MAAM,WAAW,mBAAoB,SAAQ,WAAW;IACtD,QAAQ,EAAE,YAAY,CAAC;IACvB,KAAK,EAAE,MAAM,CAAC;CACf;AAED;;;;GAIG;AACH,wBAAgB,YAAY,CAC1B,EAAE,EAAE,MAAM,EACV,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,GAAG,SAAS,EAClD,eAAe,EAAE,YAAY,GAC5B,YAAY,CAoBd;AAED,qBAAa,iBAAiB;IAC5B,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAS;IAChC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAA+B;IACzD,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAe;IAE/C,OAAO;IAMP;;;OAGG;IACH,MAAM,CAAC,cAAc,CACnB,UAAU,EAAE,eAAe,EAAE,EAC7B,MAAM,GAAE,uBAA4B,GACnC,iBAAiB;IASpB;;;;;OAKG;IACH,MAAM,CAAC,WAAW,CAChB,IAAI,EAAE,iBAAiB,EAAE,EACzB,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAC9B,MAAM,GAAE,uBAA4B,GACnC,iBAAiB;IAQpB,uEAAuE;IACvE,KAAK,CAAC,cAAc,EAAE,MAAM,EAAE,GAAG,mBAAmB;IAMpD;;;OAGG;IACG,SAAS,CACb,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,EAAE,CAAC,GAAG,MAAM,EAAE,GACpD,OAAO,CAAC,mBAAmB,CAAC;CAIhC"}

package/dist/router/cost-optimal-router.js

@@ -0,0 +1,94 @@
+/**
+ * Cost-Optimal Model Router (ADR-073)
+ *
+ * Routes each query to the *cheapest model predicted to clear a quality bar*,
+ * learned from eval logs — the productized DRACO Phase-2 finding. Wraps
+ * `@metaharness/router` (dependency-free k-NN / kernel-ridge, optional native
+ * FastGRNN via the already-present `@ruvector/tiny-dancer`).
+ *
+ * This is additive to `ModelRouter`'s existing config-rule routing: it selects
+ * a *model* by predicted cost-quality rather than a provider by static rule.
+ * It degrades gracefully — with no labelled examples it falls back to the
+ * best-predicted candidate (effectively the caller's prior behavior), so a
+ * cold start never breaks routing.
+ *
+ * @see docs/adr/ADR-073-metaharness-router-cost-optimal-model-routing.md
+ */
+import { Router } from '@metaharness/router';
+/**
+ * Parse a router model id into a provider/model binding. `"anthropic/claude-..."`
+ * → `{ provider: 'anthropic', model: 'claude-...' }`. Ids without a known
+ * provider prefix bind to `defaultProvider` with the id as the model name.
+ */
+export function parseModelId(id, modelMap, defaultProvider) {
+    if (modelMap && modelMap[id])
+        return modelMap[id];
+    const slash = id.indexOf('/');
+    if (slash > 0) {
+        const maybeProvider = id.slice(0, slash);
+        const known = [
+            'anthropic',
+            'openai',
+            'openrouter',
+            'ollama',
+            'litellm',
+            'onnx',
+            'gemini',
+            'custom',
+        ];
+        if (known.includes(maybeProvider)) {
+            return { provider: maybeProvider, model: id.slice(slash + 1) };
+        }
+    }
+    return { provider: defaultProvider, model: id };
+}
+export class CostOptimalRouter {
+    router;
+    modelMap;
+    defaultProvider;
+    constructor(router, config) {
+        this.router = router;
+        this.modelMap = config.modelMap;
+        this.defaultProvider = config.defaultProvider ?? 'anthropic';
+    }
+    /**
+     * Build from explicit candidates (each with its own labelled examples + price).
+     * Use when you already have per-model example sets.
+     */
+    static fromCandidates(candidates, config = {}) {
+        const router = new Router({
+            candidates,
+            k: config.k,
+            qualityBar: config.qualityBar,
+        });
+        return new CostOptimalRouter(router, config);
+    }
+    /**
+     * Build from a flat routing dataset — rows of (query embedding → quality each
+     * model achieved) plus a per-model price table. This is the shape eval logs
+     * and the DRACO benchmark emit, and the same shape that seeds a native
+     * tiny-dancer trainer.
+     */
+    static fromDataset(rows, prices, config = {}) {
+        const router = Router.fromExamples(rows, prices, {
+            k: config.k,
+            qualityBar: config.qualityBar,
+        });
+        return new CostOptimalRouter(router, config);
+    }
+    /** Route a pre-embedded query to the cost-optimal model + provider. */
+    route(queryEmbedding) {
+        const result = this.router.route(queryEmbedding);
+        const { provider, model } = parseModelId(result.id, this.modelMap, this.defaultProvider);
+        return { ...result, provider, model };
+    }
+    /**
+     * Route using an injected embedder (e.g. the ONNX/ruvector embedding service).
+     * Keeps this module decoupled from any specific embedding backend.
+     */
+    async routeText(text, embed) {
+        const embedding = await embed(text);
+        return this.route(embedding);
+    }
+}
+//# sourceMappingURL=cost-optimal-router.js.map

package/dist/router/cost-optimal-router.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cost-optimal-router.js","sourceRoot":"","sources":["../../src/router/cost-optimal-router.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAE,MAAM,EAA0C,MAAM,qBAAqB,CAAC;AA+CrF;;;;GAIG;AACH,MAAM,UAAU,YAAY,CAC1B,EAAU,EACV,QAAkD,EAClD,eAA6B;IAE7B,IAAI,QAAQ,IAAI,QAAQ,CAAC,EAAE,CAAC;QAAE,OAAO,QAAQ,CAAC,EAAE,CAAC,CAAC;IAClD,MAAM,KAAK,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IAC9B,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;QACd,MAAM,aAAa,GAAG,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAiB,CAAC;QACzD,MAAM,KAAK,GAAmB;YAC5B,WAAW;YACX,QAAQ;YACR,YAAY;YACZ,QAAQ;YACR,SAAS;YACT,MAAM;YACN,QAAQ;YACR,QAAQ;SACT,CAAC;QACF,IAAI,KAAK,CAAC,QAAQ,CAAC,aAAa,CAAC,EAAE,CAAC;YAClC,OAAO,EAAE,QAAQ,EAAE,aAAa,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,EAAE,CAAC;QACjE,CAAC;IACH,CAAC;IACD,OAAO,EAAE,QAAQ,EAAE,eAAe,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;AAClD,CAAC;AAED,MAAM,OAAO,iBAAiB;IACX,MAAM,CAAS;IACf,QAAQ,CAAgC;IACxC,eAAe,CAAe;IAE/C,YAAoB,MAAc,EAAE,MAA+B;QACjE,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,QAAQ,GAAG,MAAM,CAAC,QAAQ,CAAC;QAChC,IAAI,CAAC,eAAe,GAAG,MAAM,CAAC,eAAe,IAAI,WAAW,CAAC;IAC/D,CAAC;IAED;;;OAGG;IACH,MAAM,CAAC,cAAc,CACnB,UAA6B,EAC7B,SAAkC,EAAE;QAEpC,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC;YACxB,UAAU;YACV,CAAC,EAAE,MAAM,CAAC,CAAC;YACX,UAAU,EAAE,MAAM,CAAC,UAAU;SAC9B,CAAC,CAAC;QACH,OAAO,IAAI,iBAAiB,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC/C,CAAC;IAED;;;;;OAKG;IACH,MAAM,CAAC,WAAW,CAChB,IAAyB,EACzB,MAA8B,EAC9B,SAAkC,EAAE;QAEpC,MAAM,MAAM,GAAG,MAAM,CAAC,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE;YAC/C,CAAC,EAAE,MAAM,CAAC,CAAC;YACX,UAAU,EAAE,MAAM,CAAC,UAAU;SAC9B,CAAC,CAAC;QACH,OAAO,IAAI,iBAAiB,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC/C,CAAC;IAED,uEAAuE;IACvE,KAAK,CAAC,cAAwB;QAC5B,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC;QACjD,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,GAAG,YAAY,CAAC,MAAM,CAAC,EAAE,EAAE,IAAI,CAAC,QAAQ,EAAE,IAAI,CAAC,eAAe,CAAC,CAAC;QACzF,OAAO,EAAE,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC;IACxC,CAAC;IAED;;;OAGG;IACH,KAAK,CAAC,SAAS,CACb,IAAY,EACZ,KAAqD;QAErD,MAAM,SAAS,GAAG,MAAM,KAAK,CAAC,IAAI,CAAC,CAAC;QACpC,OAAO,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAC/B,CAAC;CACF","sourcesContent":["/**\n * Cost-Optimal Model Router (ADR-073)\n *\n * Routes each query to the *cheapest model predicted to clear a quality bar*,\n * learned from eval logs — the productized DRACO Phase-2 finding. Wraps\n * `@metaharness/router` (dependency-free k-NN / kernel-ridge, optional native\n * FastGRNN via the already-present `@ruvector/tiny-dancer`).\n *\n * This is additive to `ModelRouter`'s existing config-rule routing: it selects\n * a *model* by predicted cost-quality rather than a provider by static rule.\n * It degrades gracefully — with no labelled examples it falls back to the\n * best-predicted candidate (effectively the caller's prior behavior), so a\n * cold start never breaks routing.\n *\n * @see docs/adr/ADR-073-metaharness-router-cost-optimal-model-routing.md\n */\n\nimport { Router, type RouteResult, type RouterCandidate } from '@metaharness/router';\nimport type { ProviderType } from './types.js';\n\n/** One labelled observation: a query embedding and the quality a model achieved on it. */\nexport interface RoutingExample {\n  embedding: number[];\n  quality: number; // 0..1\n}\n\n/** A flat training row: query embedding → quality each model achieved on that query. */\nexport interface RoutingDatasetRow {\n  embedding: number[];\n  scores: Record<string, number>; // modelId → quality (0..1)\n}\n\n/** Maps a router model id (e.g. \"anthropic/claude-haiku-4.5\") to a concrete provider + model. */\nexport interface ModelBinding {\n  provider: ProviderType;\n  model: string;\n}\n\nexport interface CostOptimalRouterConfig {\n  /**\n   * Quality bar (0..1). When set, route() returns the CHEAPEST candidate\n   * predicted to clear it; when none clear it, the best-predicted; when unset,\n   * always the best-predicted.\n   */\n  qualityBar?: number;\n  /** k-NN neighbours used to predict quality on a new query (default 5). */\n  k?: number;\n  /**\n   * Optional explicit model id → provider/model bindings. When a routed id is\n   * absent here, it is parsed as `\"<provider>/<model>\"` (the OpenRouter-style\n   * convention), falling back to treating the whole id as the model name on the\n   * default provider.\n   */\n  modelMap?: Record<string, ModelBinding>;\n  /** Provider to assume when a model id has no `\"<provider>/\"` prefix. */\n  defaultProvider?: ProviderType;\n}\n\n/** A routing decision: the underlying RouteResult plus the resolved provider/model. */\nexport interface CostOptimalDecision extends RouteResult {\n  provider: ProviderType;\n  model: string;\n}\n\n/**\n * Parse a router model id into a provider/model binding. `\"anthropic/claude-...\"`\n * → `{ provider: 'anthropic', model: 'claude-...' }`. Ids without a known\n * provider prefix bind to `defaultProvider` with the id as the model name.\n */\nexport function parseModelId(\n  id: string,\n  modelMap: Record<string, ModelBinding> | undefined,\n  defaultProvider: ProviderType,\n): ModelBinding {\n  if (modelMap && modelMap[id]) return modelMap[id];\n  const slash = id.indexOf('/');\n  if (slash > 0) {\n    const maybeProvider = id.slice(0, slash) as ProviderType;\n    const known: ProviderType[] = [\n      'anthropic',\n      'openai',\n      'openrouter',\n      'ollama',\n      'litellm',\n      'onnx',\n      'gemini',\n      'custom',\n    ];\n    if (known.includes(maybeProvider)) {\n      return { provider: maybeProvider, model: id.slice(slash + 1) };\n    }\n  }\n  return { provider: defaultProvider, model: id };\n}\n\nexport class CostOptimalRouter {\n  private readonly router: Router;\n  private readonly modelMap?: Record<string, ModelBinding>;\n  private readonly defaultProvider: ProviderType;\n\n  private constructor(router: Router, config: CostOptimalRouterConfig) {\n    this.router = router;\n    this.modelMap = config.modelMap;\n    this.defaultProvider = config.defaultProvider ?? 'anthropic';\n  }\n\n  /**\n   * Build from explicit candidates (each with its own labelled examples + price).\n   * Use when you already have per-model example sets.\n   */\n  static fromCandidates(\n    candidates: RouterCandidate[],\n    config: CostOptimalRouterConfig = {},\n  ): CostOptimalRouter {\n    const router = new Router({\n      candidates,\n      k: config.k,\n      qualityBar: config.qualityBar,\n    });\n    return new CostOptimalRouter(router, config);\n  }\n\n  /**\n   * Build from a flat routing dataset — rows of (query embedding → quality each\n   * model achieved) plus a per-model price table. This is the shape eval logs\n   * and the DRACO benchmark emit, and the same shape that seeds a native\n   * tiny-dancer trainer.\n   */\n  static fromDataset(\n    rows: RoutingDatasetRow[],\n    prices: Record<string, number>,\n    config: CostOptimalRouterConfig = {},\n  ): CostOptimalRouter {\n    const router = Router.fromExamples(rows, prices, {\n      k: config.k,\n      qualityBar: config.qualityBar,\n    });\n    return new CostOptimalRouter(router, config);\n  }\n\n  /** Route a pre-embedded query to the cost-optimal model + provider. */\n  route(queryEmbedding: number[]): CostOptimalDecision {\n    const result = this.router.route(queryEmbedding);\n    const { provider, model } = parseModelId(result.id, this.modelMap, this.defaultProvider);\n    return { ...result, provider, model };\n  }\n\n  /**\n   * Route using an injected embedder (e.g. the ONNX/ruvector embedding service).\n   * Keeps this module decoupled from any specific embedding backend.\n   */\n  async routeText(\n    text: string,\n    embed: (text: string) => Promise<number[]> | number[],\n  ): Promise<CostOptimalDecision> {\n    const embedding = await embed(text);\n    return this.route(embedding);\n  }\n}\n"]}