@sverka/arena 0.1.8 → 0.1.10

package/dist/bin.d.mts

@@ -1 +1,13 @@
-export {}
+//#region src/bin.d.ts
+/**
+ * sverka-arena — CLI for the agent arena: run the benchmark matrix, render
+ * saved results, check the environment. Spec: specs/49-arena-cli.
+ */
+interface Io {
+  out(s: string): void;
+  err(s: string): void;
+}
+/** Entry point — exported for tests; bin calls it with process.argv. */
+export declare function main(argv: readonly string[], io?: Io): Promise<number>;
+//#endregion
+//# sourceMappingURL=bin.d.mts.map

package/dist/bin.d.mts.map

@@ -0,0 +1 @@
+{"version":3,"file":"bin.d.mts","names":[],"sources":["../src/bin.ts"],"mappings":";;;;;UAgBU;EACR,IAAI;EACJ,IAAI;;;wBA+JgB,KACpB,yBACA,KAAI,KACH"}

package/dist/bin.mjs

@@ -1,13 +1,163 @@
 #!/usr/bin/env node
+import { i as loadArenaConfig, l as runArena, n as ArenaError, t as renderReport } from "./report-DM4LfVAm.mjs";
+import { readFile } from "node:fs/promises";
+import { existsSync } from "node:fs";
+import { delimiter, join, resolve } from "node:path";
 import process from "node:process";
 //#region src/bin.ts
-async function main() {
-	process.stderr.write("arena CLI — not yet implemented\n");
-	return 1;
+/**
+* sverka-arena — CLI for the agent arena: run the benchmark matrix, render
+* saved results, check the environment. Spec: specs/49-arena-cli.
+*/
+const USAGE = `usage: sverka-arena <command> [options]
+
+commands:
+  run      [--config <path>] [--out <dir>] [--format json|text]
+  report   <results.json>               [--format json|text]
+  doctor   [--config <path>]            [--format json|text]
+`;
+function flagValue(argv, i) {
+	const val = argv[i + 1];
+	if (val === void 0 || val.startsWith("--")) throw new ArenaError(`${argv[i]} requires a value`, "CONFIG_INVALID");
+	return val;
+}
+function parseFormat(val) {
+	if (val !== "json" && val !== "text") throw new ArenaError(`invalid --format '${val}'`, "CONFIG_INVALID");
+	return val;
+}
+function parseArgs(argv) {
+	const positional = [];
+	let config = "arena.config.ts";
+	let out;
+	let format = "text";
+	let command;
+	for (let i = 0; i < argv.length; i++) {
+		const arg = argv[i];
+		switch (arg) {
+			case "--config":
+				config = flagValue(argv, i);
+				i += 1;
+				break;
+			case "--out":
+				out = flagValue(argv, i);
+				i += 1;
+				break;
+			case "--format":
+				format = parseFormat(flagValue(argv, i));
+				i += 1;
+				break;
+			default:
+				if (arg.startsWith("--")) throw new ArenaError(`unknown option '${arg}'`, "CONFIG_INVALID");
+				if (command === void 0) command = arg;
+				else positional.push(arg);
+		}
+	}
+	return {
+		command,
+		positional,
+		config,
+		out,
+		format
+	};
+}
+function which(bin) {
+	const exts = process.platform === "win32" ? ["", ...(process.env.PATHEXT ?? ".EXE;.CMD;.BAT").split(";")] : [""];
+	for (const dir of (process.env.PATH ?? "").split(delimiter)) if (exts.some((ext) => existsSync(join(dir, bin + ext)))) return true;
+	return false;
+}
+async function cmdReport(args, io) {
+	const file = args.positional[0];
+	if (file === void 0) {
+		io.err("report: missing <results.json> path");
+		return 2;
+	}
+	let result;
+	try {
+		result = JSON.parse(await readFile(file, "utf8"));
+	} catch (err) {
+		io.err(`report: cannot read '${file}': ${err instanceof Error ? err.message : String(err)}`);
+		return 2;
+	}
+	if (!Array.isArray(result.aggregates) || !Array.isArray(result.analysis) || !Array.isArray(result.results)) {
+		io.err(`report: '${file}' is not a sverka-arena results file\n`);
+		return 2;
+	}
+	if (args.format === "json") io.out(JSON.stringify(result, null, 2) + "\n");
+	else io.out(renderReport(result) + "\n");
+	return 0;
+}
+async function cmdRun(args, io) {
+	const config = await loadArenaConfig(args.config);
+	if (args.out !== void 0) config.outputDir = resolve(args.out);
+	const result = await runArena(config);
+	if (args.format === "json") io.out(JSON.stringify(result, null, 2) + "\n");
+	else io.out(renderReport(result) + "\n");
+	return 0;
+}
+async function cmdDoctor(args, io) {
+	const config = await loadArenaConfig(args.config);
+	const checks = [];
+	const agentFound = which(config.agent.id);
+	checks.push({
+		name: `agent binary: ${config.agent.id}`,
+		ok: agentFound,
+		detail: agentFound ? "on PATH" : "not found on PATH"
+	});
+	const models = config.judge ? [...config.models, config.judge.model] : config.models;
+	for (const m of models) {
+		if (m.envVar === void 0) continue;
+		checks.push({
+			name: `env: ${m.envVar} (model ${m.id})`,
+			ok: process.env[m.envVar] !== void 0,
+			detail: process.env[m.envVar] !== void 0 ? "set" : "not set"
+		});
+	}
+	if (args.format === "json") io.out(JSON.stringify({ checks }, null, 2) + "\n");
+	else for (const chk of checks) io.out(`${chk.ok ? "ok" : "FAIL"}  ${chk.name} — ${chk.detail}\n`);
+	return checks.every((chk) => chk.ok) ? 0 : 1;
+}
+const defaultIo = {
+	out: (s) => process.stdout.write(s),
+	err: (s) => process.stderr.write(s)
+};
+/** Entry point — exported for tests; bin calls it with process.argv. */
+async function main(argv, io = defaultIo) {
+	let args;
+	try {
+		args = parseArgs(argv);
+	} catch (err) {
+		io.err((err instanceof Error ? err.message : String(err)) + "\n");
+		io.err(USAGE);
+		return 2;
+	}
+	if (args.command === void 0) {
+		io.err(USAGE);
+		return 2;
+	}
+	try {
+		switch (args.command) {
+			case "run": return await cmdRun(args, io);
+			case "report": return await cmdReport(args, io);
+			case "doctor": return await cmdDoctor(args, io);
+			default:
+				io.err(`unknown command '${args.command}'\n`);
+				io.err(USAGE);
+				return 2;
+		}
+	} catch (err) {
+		if (err instanceof ArenaError) {
+			io.err(`error: ${err.message}\n`);
+			return 2;
+		}
+		io.err(`error: ${err instanceof Error ? err.message : String(err)}\n`);
+		return 3;
+	}
+}
+if (import.meta.url === `file://${process.argv[1]}`) {
+	const code = await main(process.argv.slice(2));
+	process.exit(code);
 }
-const code = await main();
-process.exit(code);
 //#endregion
-export {};
+export { main };
 
 //# sourceMappingURL=bin.mjs.map

package/dist/bin.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"bin.mjs","names":[],"sources":["../src/bin.ts"],"sourcesContent":["#!/usr/bin/env node\nimport process from \"node:process\";\n// Stub — will be implemented in next phase\nasync function main(): Promise<number> {\n  process.stderr.write(\"arena CLI — not yet implemented\\n\");\n  return 1;\n}\n\nconst code = await main();\nprocess.exit(code);\n"],"mappings":";;;AAGA,eAAe,OAAwB;CACrC,QAAQ,OAAO,MAAM,mCAAmC;CACxD,OAAO;AACT;AAEA,MAAM,OAAO,MAAM,KAAK;AACxB,QAAQ,KAAK,IAAI"}
+{"version":3,"file":"bin.mjs","names":[],"sources":["../src/bin.ts"],"sourcesContent":["#!/usr/bin/env node\n/**\n * sverka-arena — CLI for the agent arena: run the benchmark matrix, render\n * saved results, check the environment. Spec: specs/49-arena-cli.\n */\n\nimport process from \"node:process\";\nimport { readFile } from \"node:fs/promises\";\nimport { existsSync } from \"node:fs\";\nimport { delimiter, join, resolve } from \"node:path\";\n\nimport type { ArenaResult } from \"./types.js\";\nimport { ArenaError, loadArenaConfig } from \"./config.js\";\nimport { renderReport } from \"./report.js\";\nimport { runArena } from \"./runner.js\";\n\ninterface Io {\n  out(s: string): void;\n  err(s: string): void;\n}\n\ninterface ParsedArgs {\n  command: string | undefined;\n  positional: string[];\n  config: string;\n  out: string | undefined;\n  format: \"text\" | \"json\";\n}\n\nconst USAGE = `usage: sverka-arena <command> [options]\n\ncommands:\n  run      [--config <path>] [--out <dir>] [--format json|text]\n  report   <results.json>               [--format json|text]\n  doctor   [--config <path>]            [--format json|text]\n`;\n\nfunction flagValue(argv: readonly string[], i: number): string {\n  const val = argv[i + 1];\n  if (val === undefined || val.startsWith(\"--\")) {\n    throw new ArenaError(`${argv[i]} requires a value`, \"CONFIG_INVALID\");\n  }\n  return val;\n}\n\nfunction parseFormat(val: string): \"text\" | \"json\" {\n  if (val !== \"json\" && val !== \"text\") {\n    throw new ArenaError(`invalid --format '${val}'`, \"CONFIG_INVALID\");\n  }\n  return val;\n}\n\nfunction parseArgs(argv: readonly string[]): ParsedArgs {\n  const positional: string[] = [];\n  let config = \"arena.config.ts\";\n  let out: string | undefined;\n  let format: \"text\" | \"json\" = \"text\";\n  let command: string | undefined;\n  for (let i = 0; i < argv.length; i++) {\n    const arg = argv[i]!;\n    switch (arg) {\n      case \"--config\":\n        config = flagValue(argv, i);\n        i += 1;\n        break;\n      case \"--out\":\n        out = flagValue(argv, i);\n        i += 1;\n        break;\n      case \"--format\":\n        format = parseFormat(flagValue(argv, i));\n        i += 1;\n        break;\n      default:\n        if (arg.startsWith(\"--\")) {\n          throw new ArenaError(`unknown option '${arg}'`, \"CONFIG_INVALID\");\n        }\n        if (command === undefined) command = arg;\n        else positional.push(arg);\n    }\n  }\n  return { command, positional, config, out, format };\n}\n\nfunction which(bin: string): boolean {\n  const exts =\n    process.platform === \"win32\"\n      ? [\"\", ...(process.env.PATHEXT ?? \".EXE;.CMD;.BAT\").split(\";\")]\n      : [\"\"];\n  for (const dir of (process.env.PATH ?? \"\").split(delimiter)) {\n    if (exts.some((ext) => existsSync(join(dir, bin + ext)))) return true;\n  }\n  return false;\n}\n\nasync function cmdReport(args: ParsedArgs, io: Io): Promise<number> {\n  const file = args.positional[0];\n  if (file === undefined) {\n    io.err(\"report: missing <results.json> path\");\n    return 2;\n  }\n  let result: ArenaResult;\n  try {\n    result = JSON.parse(await readFile(file, \"utf8\")) as ArenaResult;\n  } catch (err) {\n    io.err(\n      `report: cannot read '${file}': ${err instanceof Error ? err.message : String(err)}`,\n    );\n    return 2;\n  }\n  if (\n    !Array.isArray(result.aggregates) ||\n    !Array.isArray(result.analysis) ||\n    !Array.isArray(result.results)\n  ) {\n    io.err(`report: '${file}' is not a sverka-arena results file\\n`);\n    return 2;\n  }\n  if (args.format === \"json\") {\n    io.out(JSON.stringify(result, null, 2) + \"\\n\");\n  } else {\n    io.out(renderReport(result) + \"\\n\");\n  }\n  return 0;\n}\n\nasync function cmdRun(args: ParsedArgs, io: Io): Promise<number> {\n  const config = await loadArenaConfig(args.config);\n  if (args.out !== undefined) config.outputDir = resolve(args.out);\n  const result = await runArena(config);\n  if (args.format === \"json\") {\n    io.out(JSON.stringify(result, null, 2) + \"\\n\");\n  } else {\n    io.out(renderReport(result) + \"\\n\");\n  }\n  return 0;\n}\n\nasync function cmdDoctor(args: ParsedArgs, io: Io): Promise<number> {\n  const config = await loadArenaConfig(args.config);\n  const checks: { name: string; ok: boolean; detail: string }[] = [];\n\n  const agentFound = which(config.agent.id);\n  checks.push({\n    name: `agent binary: ${config.agent.id}`,\n    ok: agentFound,\n    detail: agentFound ? \"on PATH\" : \"not found on PATH\",\n  });\n\n  const models = config.judge\n    ? [...config.models, config.judge.model]\n    : config.models;\n  for (const m of models) {\n    if (m.envVar === undefined) continue;\n    checks.push({\n      name: `env: ${m.envVar} (model ${m.id})`,\n      ok: process.env[m.envVar] !== undefined,\n      detail: process.env[m.envVar] !== undefined ? \"set\" : \"not set\",\n    });\n  }\n\n  if (args.format === \"json\") {\n    io.out(JSON.stringify({ checks }, null, 2) + \"\\n\");\n  } else {\n    for (const chk of checks) {\n      io.out(`${chk.ok ? \"ok\" : \"FAIL\"}  ${chk.name} — ${chk.detail}\\n`);\n    }\n  }\n  return checks.every((chk) => chk.ok) ? 0 : 1;\n}\n\nconst defaultIo: Io = {\n  out: (s) => process.stdout.write(s),\n  err: (s) => process.stderr.write(s),\n};\n\n/** Entry point — exported for tests; bin calls it with process.argv. */\nexport async function main(\n  argv: readonly string[],\n  io: Io = defaultIo,\n): Promise<number> {\n  let args: ParsedArgs;\n  try {\n    args = parseArgs(argv);\n  } catch (err) {\n    io.err((err instanceof Error ? err.message : String(err)) + \"\\n\");\n    io.err(USAGE);\n    return 2;\n  }\n\n  if (args.command === undefined) {\n    io.err(USAGE);\n    return 2;\n  }\n\n  try {\n    switch (args.command) {\n      case \"run\":\n        return await cmdRun(args, io);\n      case \"report\":\n        return await cmdReport(args, io);\n      case \"doctor\":\n        return await cmdDoctor(args, io);\n      default:\n        io.err(`unknown command '${args.command}'\\n`);\n        io.err(USAGE);\n        return 2;\n    }\n  } catch (err) {\n    if (err instanceof ArenaError) {\n      io.err(`error: ${err.message}\\n`);\n      return 2;\n    }\n    io.err(`error: ${err instanceof Error ? err.message : String(err)}\\n`);\n    return 3;\n  }\n}\n\n// Run when invoked directly (not imported by tests).\nif (import.meta.url === `file://${process.argv[1]}`) {\n  const code = await main(process.argv.slice(2));\n  process.exit(code);\n}\n"],"mappings":";;;;;;;;;;;AA6BA,MAAM,QAAQ;;;;;;;AAQd,SAAS,UAAU,MAAyB,GAAmB;CAC7D,MAAM,MAAM,KAAK,IAAI;CACrB,IAAI,QAAQ,KAAA,KAAa,IAAI,WAAW,IAAI,GAC1C,MAAM,IAAI,WAAW,GAAG,KAAK,GAAG,oBAAoB,gBAAgB;CAEtE,OAAO;AACT;AAEA,SAAS,YAAY,KAA8B;CACjD,IAAI,QAAQ,UAAU,QAAQ,QAC5B,MAAM,IAAI,WAAW,qBAAqB,IAAI,IAAI,gBAAgB;CAEpE,OAAO;AACT;AAEA,SAAS,UAAU,MAAqC;CACtD,MAAM,aAAuB,CAAC;CAC9B,IAAI,SAAS;CACb,IAAI;CACJ,IAAI,SAA0B;CAC9B,IAAI;CACJ,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;EACpC,MAAM,MAAM,KAAK;EACjB,QAAQ,KAAR;GACE,KAAK;IACH,SAAS,UAAU,MAAM,CAAC;IAC1B,KAAK;IACL;GACF,KAAK;IACH,MAAM,UAAU,MAAM,CAAC;IACvB,KAAK;IACL;GACF,KAAK;IACH,SAAS,YAAY,UAAU,MAAM,CAAC,CAAC;IACvC,KAAK;IACL;GACF;IACE,IAAI,IAAI,WAAW,IAAI,GACrB,MAAM,IAAI,WAAW,mBAAmB,IAAI,IAAI,gBAAgB;IAElE,IAAI,YAAY,KAAA,GAAW,UAAU;SAChC,WAAW,KAAK,GAAG;EAC5B;CACF;CACA,OAAO;EAAE;EAAS;EAAY;EAAQ;EAAK;CAAO;AACpD;AAEA,SAAS,MAAM,KAAsB;CACnC,MAAM,OACJ,QAAQ,aAAa,UACjB,CAAC,IAAI,IAAI,QAAQ,IAAI,WAAW,iBAAA,CAAkB,MAAM,GAAG,CAAC,IAC5D,CAAC,EAAE;CACT,KAAK,MAAM,QAAQ,QAAQ,IAAI,QAAQ,GAAA,CAAI,MAAM,SAAS,GACxD,IAAI,KAAK,MAAM,QAAQ,WAAW,KAAK,KAAK,MAAM,GAAG,CAAC,CAAC,GAAG,OAAO;CAEnE,OAAO;AACT;AAEA,eAAe,UAAU,MAAkB,IAAyB;CAClE,MAAM,OAAO,KAAK,WAAW;CAC7B,IAAI,SAAS,KAAA,GAAW;EACtB,GAAG,IAAI,qCAAqC;EAC5C,OAAO;CACT;CACA,IAAI;CACJ,IAAI;EACF,SAAS,KAAK,MAAM,MAAM,SAAS,MAAM,MAAM,CAAC;CAClD,SAAS,KAAK;EACZ,GAAG,IACD,wBAAwB,KAAK,KAAK,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,GACnF;EACA,OAAO;CACT;CACA,IACE,CAAC,MAAM,QAAQ,OAAO,UAAU,KAChC,CAAC,MAAM,QAAQ,OAAO,QAAQ,KAC9B,CAAC,MAAM,QAAQ,OAAO,OAAO,GAC7B;EACA,GAAG,IAAI,YAAY,KAAK,uCAAuC;EAC/D,OAAO;CACT;CACA,IAAI,KAAK,WAAW,QAClB,GAAG,IAAI,KAAK,UAAU,QAAQ,MAAM,CAAC,IAAI,IAAI;MAE7C,GAAG,IAAI,aAAa,MAAM,IAAI,IAAI;CAEpC,OAAO;AACT;AAEA,eAAe,OAAO,MAAkB,IAAyB;CAC/D,MAAM,SAAS,MAAM,gBAAgB,KAAK,MAAM;CAChD,IAAI,KAAK,QAAQ,KAAA,GAAW,OAAO,YAAY,QAAQ,KAAK,GAAG;CAC/D,MAAM,SAAS,MAAM,SAAS,MAAM;CACpC,IAAI,KAAK,WAAW,QAClB,GAAG,IAAI,KAAK,UAAU,QAAQ,MAAM,CAAC,IAAI,IAAI;MAE7C,GAAG,IAAI,aAAa,MAAM,IAAI,IAAI;CAEpC,OAAO;AACT;AAEA,eAAe,UAAU,MAAkB,IAAyB;CAClE,MAAM,SAAS,MAAM,gBAAgB,KAAK,MAAM;CAChD,MAAM,SAA0D,CAAC;CAEjE,MAAM,aAAa,MAAM,OAAO,MAAM,EAAE;CACxC,OAAO,KAAK;EACV,MAAM,iBAAiB,OAAO,MAAM;EACpC,IAAI;EACJ,QAAQ,aAAa,YAAY;CACnC,CAAC;CAED,MAAM,SAAS,OAAO,QAClB,CAAC,GAAG,OAAO,QAAQ,OAAO,MAAM,KAAK,IACrC,OAAO;CACX,KAAK,MAAM,KAAK,QAAQ;EACtB,IAAI,EAAE,WAAW,KAAA,GAAW;EAC5B,OAAO,KAAK;GACV,MAAM,QAAQ,EAAE,OAAO,UAAU,EAAE,GAAG;GACtC,IAAI,QAAQ,IAAI,EAAE,YAAY,KAAA;GAC9B,QAAQ,QAAQ,IAAI,EAAE,YAAY,KAAA,IAAY,QAAQ;EACxD,CAAC;CACH;CAEA,IAAI,KAAK,WAAW,QAClB,GAAG,IAAI,KAAK,UAAU,EAAE,OAAO,GAAG,MAAM,CAAC,IAAI,IAAI;MAEjD,KAAK,MAAM,OAAO,QAChB,GAAG,IAAI,GAAG,IAAI,KAAK,OAAO,OAAO,IAAI,IAAI,KAAK,KAAK,IAAI,OAAO,GAAG;CAGrE,OAAO,OAAO,OAAO,QAAQ,IAAI,EAAE,IAAI,IAAI;AAC7C;AAEA,MAAM,YAAgB;CACpB,MAAM,MAAM,QAAQ,OAAO,MAAM,CAAC;CAClC,MAAM,MAAM,QAAQ,OAAO,MAAM,CAAC;AACpC;;AAGA,eAAsB,KACpB,MACA,KAAS,WACQ;CACjB,IAAI;CACJ,IAAI;EACF,OAAO,UAAU,IAAI;CACvB,SAAS,KAAK;EACZ,GAAG,KAAK,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,KAAK,IAAI;EAChE,GAAG,IAAI,KAAK;EACZ,OAAO;CACT;CAEA,IAAI,KAAK,YAAY,KAAA,GAAW;EAC9B,GAAG,IAAI,KAAK;EACZ,OAAO;CACT;CAEA,IAAI;EACF,QAAQ,KAAK,SAAb;GACE,KAAK,OACH,OAAO,MAAM,OAAO,MAAM,EAAE;GAC9B,KAAK,UACH,OAAO,MAAM,UAAU,MAAM,EAAE;GACjC,KAAK,UACH,OAAO,MAAM,UAAU,MAAM,EAAE;GACjC;IACE,GAAG,IAAI,oBAAoB,KAAK,QAAQ,IAAI;IAC5C,GAAG,IAAI,KAAK;IACZ,OAAO;EACX;CACF,SAAS,KAAK;EACZ,IAAI,eAAe,YAAY;GAC7B,GAAG,IAAI,UAAU,IAAI,QAAQ,GAAG;GAChC,OAAO;EACT;EACA,GAAG,IAAI,UAAU,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,EAAE,GAAG;EACrE,OAAO;CACT;AACF;AAGA,IAAI,YAAY,QAAQ,UAAU,QAAQ,KAAK,MAAM;CACnD,MAAM,OAAO,MAAM,KAAK,QAAQ,KAAK,MAAM,CAAC,CAAC;CAC7C,QAAQ,KAAK,IAAI;AACnB"}

package/dist/index.d.mts

@@ -1,6 +1,7 @@
 import { ChildProcess } from "node:child_process";
 import * as acp from "@agentclientprotocol/sdk";
 import { PromptResponse, SessionUpdate, Usage } from "@agentclientprotocol/sdk";
+import { z } from "zod";
 //#region src/types.d.ts
 /** Agent adapter — pluggable interface for different agent CLIs */
 interface AgentAdapter {
@@ -43,7 +44,8 @@ interface Task {
   /** Optional expected output for deterministic comparison */
   expectedOutput?: string;
   /** Fixture project to copy into the workspace for this task.
-   * Path relative to the arena package root (e.g. "fixtures/ts-fix-test").
+   * Relative paths resolve against the arena package root (programmatic
+   * use) — `loadArenaConfig` anchors them to the config file's directory.
    * If omitted, the runner uses an empty temp directory. */
   fixture?: string;
   /** Optional deterministic checks to run after the agent finishes.
@@ -62,6 +64,7 @@ interface ArenaConfig {
   agent: AgentAdapter;
   models: ModelConfig[];
   plugins: PluginConfig[];
+  /** Base dir for the per-run isolated workspaces (default: system temp). */
   workspace?: string;
   repetitions?: number;
   outputDir: string;
@@ -422,5 +425,68 @@ export declare function judgeAllRuns(results: RunResult[], tasks: Task[], config
  */
 export declare function compareCombos(baseline: RunResult, candidate: RunResult): ComboComparison;
 //#endregion
-export type { AgentAdapter, AgentProcess, AgentSpawnConfig, AggregateMetrics, ArenaConfig, ArenaResult, CaseAnalysis, CheckResult, ComboComparison, DeterministicCheck, JudgeConfig, JudgeVerdict, ModelConfig, Observation, PluginConfig, RunMetrics, RunResult, Task, ToolCall, ToolCallMessage, ToolCallUpdateMessage, TraceData, TraceStep };
+//#region src/config.d.ts
+export declare class ArenaError extends Error {
+  readonly code: "CONFIG_INVALID" | "CONFIG_NOT_FOUND" | "UNKNOWN_AGENT";
+  constructor(message: string, code: "CONFIG_INVALID" | "CONFIG_NOT_FOUND" | "UNKNOWN_AGENT");
+}
+export declare function resolveAdapter(name: string): AgentAdapter;
+declare const configFileSchema: z.ZodObject<{
+  agent: z.ZodString;
+  models: z.ZodArray<z.ZodObject<{
+    id: z.ZodString;
+    name: z.ZodString;
+    envVar: z.ZodOptional<z.ZodString>;
+  }, z.core.$strip>>;
+  plugins: z.ZodOptional<z.ZodArray<z.ZodObject<{
+    id: z.ZodString;
+    name: z.ZodString;
+    path: z.ZodString;
+    enabled: z.ZodOptional<z.ZodBoolean>;
+    installPath: z.ZodOptional<z.ZodString>;
+  }, z.core.$strip>>>;
+  tasks: z.ZodArray<z.ZodObject<{
+    id: z.ZodString;
+    name: z.ZodString;
+    prompt: z.ZodString;
+    timeoutMs: z.ZodOptional<z.ZodNumber>;
+    successCriteria: z.ZodOptional<z.ZodString>;
+    expectedOutput: z.ZodOptional<z.ZodString>;
+    fixture: z.ZodOptional<z.ZodString>;
+    checks: z.ZodOptional<z.ZodArray<z.ZodObject<{
+      id: z.ZodString;
+      command: z.ZodString;
+      description: z.ZodString;
+    }, z.core.$strip>>>;
+  }, z.core.$strip>>;
+  workspace: z.ZodOptional<z.ZodString>;
+  repetitions: z.ZodOptional<z.ZodNumber>;
+  outputDir: z.ZodString;
+  judge: z.ZodOptional<z.ZodObject<{
+    model: z.ZodObject<{
+      id: z.ZodString;
+      name: z.ZodString;
+      envVar: z.ZodOptional<z.ZodString>;
+    }, z.core.$strip>;
+    agent: z.ZodOptional<z.ZodString>;
+    repetitions: z.ZodOptional<z.ZodNumber>;
+    revealPlugins: z.ZodOptional<z.ZodBoolean>;
+    systemPrompt: z.ZodOptional<z.ZodString>;
+  }, z.core.$strip>>;
+}, z.core.$strip>;
+/** Declarative config shape authored in `arena.config.ts`. */
+type ArenaConfigFile = z.input<typeof configFileSchema>;
+/** Type-only identity helper — gives configs full typechecking. */
+export declare function defineConfig(config: ArenaConfigFile): ArenaConfigFile;
+/**
+ * Load and validate an arena config file, resolving the named agent
+ * (and judge agent, defaulting to the same adapter) to live adapters.
+ */
+export declare function loadArenaConfig(configPath: string): Promise<ArenaConfig>;
+//#endregion
+//#region src/report.d.ts
+/** Render an ArenaResult as human-readable text. */
+export declare function renderReport(result: ArenaResult): string;
+//#endregion
+export type { AgentAdapter, AgentProcess, AgentSpawnConfig, AggregateMetrics, ArenaConfig, ArenaConfigFile, ArenaResult, CaseAnalysis, CheckResult, ComboComparison, DeterministicCheck, JudgeConfig, JudgeVerdict, ModelConfig, Observation, PluginConfig, RunMetrics, RunResult, Task, ToolCall, ToolCallMessage, ToolCallUpdateMessage, TraceData, TraceStep };
 //# sourceMappingURL=index.d.mts.map

package/dist/index.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.mts","names":[],"sources":["../src/types.ts","../src/adapters/devin.ts","../src/acp.ts","../src/runner.ts","../src/judge.ts"],"mappings":";;;;;UACiB;EACf;EACA,MAAM,QAAQ,mBAAmB;;UAGlB;EACf,OAAO;EACP;EACA,SAAS;EACT;EACA,MAAM;;UAGS;EACf,IAAI,gBAAgB,oBAAoB,QAAQ;EAChD;;;UAIe;EACf;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;;EAEA;;EAEA;;;;EAIA;;;EAGA,SAAS;;;UAIM;EACf;EACA;EACA;;;UAIe;EACf,OAAO;EACP,OAAO;EACP,QAAQ;EACR,SAAS;EACT;EACA;EACA;;EAEA,QAAQ;;;UAIO;;EAEf,OAAO;;EAEP,OAAO;;EAEP;;EAEA;;EAEA;;;UAIe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA,SAAS;EACT,OAAO;;EAEP;EACA;;EAEA,cAAc;EACd;;EAEA,UAAU;;;UAIK;EACf;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA,OAAO;EACP;IACE;IACA;IACA;IACA;;;;UAKa;EACf;EACA;EACA;EACA;EACA,YAAY;EACZ,eAAe;EACf;;;UAIe;EACf;EACA,WAAW;EACX;;;UAIe;EACf;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;;UAIe;EACf;EACA;IACE;IACA;IACA;IACA;IACA;;EAEF,SAAS;EACT,YAAY;;EAEZ,UAAU;;;UAIK;EACf;EACA;EACA;;EAEA,aAAa;;EAEb;;;UAIe;;EAEf;EACA;;EAEA;EACA;EACA;EACA;EACA;;EAEA;;;;;UCnMQ;EACR;EACA;EACA;EACA;EACA;IACE;IACA;MACE;MACA;;;;;UAMI;EACR;EACA;EACA;EACA;;;UAIQ;EACR;EACA;EACA;IACE;IACA;IACA;;EAEF,eAAe;EACf,OAAO;;;wBAMO;;wBAeA,cAAc,YAAY;;wBAmC1B,mBAAmB,OAAO;EACxC;EACA;EACA;EACA;;;;;;;;;;;wBA0DoB,eACpB,mBACA,SAAS,iBACR;;;;;;;;;;;;qBAwCU,wBAAwB;WAC1B;EAET,MAAM,QAAQ,mBAAmB;;;;;KCrNvB,kBAAkB,QAC5B;EACE;;;KAIQ,wBAAwB,QAClC;EACE;;;;;;;;;;;;wBA2EY,YAAY,QAAQ,mBAAmB;;;;;;;;wBAgBvC,2BAA2B,IAAI,qBAC7C,IAAI,0BACJ,IAAI;;;;;;;;;;;;;;;wBAqFgB,cACpB,MAAM,cACN,mBACA,gBACA,mBACA,cAAc,QAAQ,0BACtB,gBAAgB,QAAQ,gCACxB,aAAa,6BACZ,QAAQ;;;;;;;;;;;;;wBC3KK,mBAAmB,SAAS,iBAAiB;;;;;;;;;wBA0BvC,SAAS,QAAQ,cAAc,QAAQ;;;;;;;;wBA4Q7C,iBACd,SAAS,aACT,SAAS,GAAG,uBACZ,iBACC;;;;;;;;;;wBAoIa,gBACd,SAAS,aACT,OAAO,SACN;;;;qBC5cU;;;;;;;;;wBAkBG,iBACd,MAAM,MACN,mBACA,QAAQ,aACR;;;;;;;;wBA0Cc,mBACd,kBACA,kBACA,KAAK,YACJ;;;;;wBAoDmB,SACpB,KAAK,WACL,MAAM,MACN,QAAQ,aACR,mBACC,QAAQ;;;;;;;wBAmCW,aACpB,SAAS,aACT,OAAO,QACP,QAAQ,cACP,QAAQ;;;;;;;;;;;;wBAoCK,cACd,UAAU,WACV,WAAW,YACV"}
+{"version":3,"file":"index.d.mts","names":[],"sources":["../src/types.ts","../src/adapters/devin.ts","../src/acp.ts","../src/runner.ts","../src/judge.ts","../src/config.ts","../src/report.ts"],"mappings":";;;;;;UACiB;EACf;EACA,MAAM,QAAQ,mBAAmB;;UAGlB;EACf,OAAO;EACP;EACA,SAAS;EACT;EACA,MAAM;;UAGS;EACf,IAAI,gBAAgB,oBAAoB,QAAQ;EAChD;;;UAIe;EACf;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;;EAEA;;EAEA;;;;;EAKA;;;EAGA,SAAS;;;UAIM;EACf;EACA;EACA;;;UAIe;EACf,OAAO;EACP,OAAO;EACP,QAAQ;EACR,SAAS;;EAET;EACA;EACA;;EAEA,QAAQ;;;UAIO;;EAEf,OAAO;;EAEP,OAAO;;EAEP;;EAEA;;EAEA;;;UAIe;;EAEf;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA,SAAS;EACT,OAAO;;EAEP;EACA;;EAEA,cAAc;EACd;;EAEA,UAAU;;;UAIK;EACf;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;UAIe;EACf;EACA;EACA,OAAO;EACP;IACE;IACA;IACA;IACA;;;;UAKa;EACf;EACA;EACA;EACA;EACA,YAAY;EACZ,eAAe;EACf;;;UAIe;EACf;EACA,WAAW;EACX;;;UAIe;EACf;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;;EAEA;;EAEA;;;UAIe;EACf;EACA;IACE;IACA;IACA;IACA;IACA;;EAEF,SAAS;EACT,YAAY;;EAEZ,UAAU;;;UAIK;EACf;EACA;EACA;;EAEA,aAAa;;EAEb;;;UAIe;;EAEf;EACA;;EAEA;EACA;EACA;EACA;EACA;;EAEA;;;;;UCrMQ;EACR;EACA;EACA;EACA;EACA;IACE;IACA;MACE;MACA;;;;;UAMI;EACR;EACA;EACA;EACA;;;UAIQ;EACR;EACA;EACA;IACE;IACA;IACA;;EAEF,eAAe;EACf,OAAO;;;wBAMO;;wBAeA,cAAc,YAAY;;wBAmC1B,mBAAmB,OAAO;EACxC;EACA;EACA;EACA;;;;;;;;;;;wBA0DoB,eACpB,mBACA,SAAS,iBACR;;;;;;;;;;;;qBAwCU,wBAAwB;WAC1B;EAET,MAAM,QAAQ,mBAAmB;;;;;KCrNvB,kBAAkB,QAC5B;EACE;;;KAIQ,wBAAwB,QAClC;EACE;;;;;;;;;;;;wBA2EY,YAAY,QAAQ,mBAAmB;;;;;;;;wBAgBvC,2BAA2B,IAAI,qBAC7C,IAAI,0BACJ,IAAI;;;;;;;;;;;;;;;wBAqFgB,cACpB,MAAM,cACN,mBACA,gBACA,mBACA,cAAc,QAAQ,0BACtB,gBAAgB,QAAQ,gCACxB,aAAa,6BACZ,QAAQ;;;;;;;;;;;;;wBC3KK,mBAAmB,SAAS,iBAAiB;;;;;;;;;wBA0BvC,SAAS,QAAQ,cAAc,QAAQ;;;;;;;;wBAkS7C,iBACd,SAAS,aACT,SAAS,GAAG,uBACZ,iBACC;;;;;;;;;;wBAoIa,gBACd,SAAS,aACT,OAAO,SACN;;;;qBCleU;;;;;;;;;wBAkBG,iBACd,MAAM,MACN,mBACA,QAAQ,aACR;;;;;;;;wBA0Cc,mBACd,kBACA,kBACA,KAAK,YACJ;;;;;wBAoDmB,SACpB,KAAK,WACL,MAAM,MACN,QAAQ,aACR,mBACC,QAAQ;;;;;;;wBAmCW,aACpB,SAAS,aACT,OAAO,QACP,QAAQ,cACP,QAAQ;;;;;;;;;;;;wBAoCK,cACd,UAAU,WACV,WAAW,YACV;;;qBCnNU,mBAAmB;WAGnB;EADT,YAAA,iBACS;;wBAaG,eAAe,eAAe;cA2CxC,kBAAgB,EAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiBpB,EAAA,KAAA;;KAGU,kBAAkB,EAAE,aAAa;;wBAG7B,aAAa,QAAQ,kBAAkB;;;;;wBAUjC,gBACpB,qBACC,QAAQ;;;;wBCxDK,aAAa,QAAQ"}