@garuda-agent/garuda 0.17.0 → 0.18.0

package/dist/{chunk-ZAE25JM2.js → chunk-QANRZW7W.js}

@@ -57,6 +57,12 @@ async function loadMcpConfig({ home, root } = { home: homedir(), root: process.c
       problems.push(`${file}: invalid JSON: ${error.message}`);
       return [];
     }
+    if (Array.isArray(json?.servers)) {
+      problems.push(
+        `${file}: "servers" must be an object of name \u2192 server, not a list: { "servers": { "graphify": { "command": "\u2026" }, "linear": { "url": "\u2026" } } }`
+      );
+      return [];
+    }
     const parsed = fileSchema.safeParse(json);
     if (!parsed.success) {
       problems.push(`${file}: ${z.prettifyError(parsed.error)}`);
@@ -147,6 +153,7 @@ function commandLine(def) {
 }
 
 export {
+  MCP_FILE,
   SERVER_NAME,
   isHttp,
   loadMcpConfig,
@@ -155,4 +162,4 @@ export {
   expandEnv,
   commandLine
 };
-//# sourceMappingURL=chunk-ZAE25JM2.js.map
+//# sourceMappingURL=chunk-QANRZW7W.js.map

package/dist/{chunk-735BKABA.js → chunk-UKIE4AAM.js}

@@ -3,7 +3,7 @@ import {
   Redactor,
   contextSize,
   totalTokens
-} from "./chunk-J4346IDH.js";
+} from "./chunk-7KBOIPUT.js";
 import {
   cleanLine
 } from "./chunk-PZS3V6P3.js";
@@ -469,4 +469,4 @@ export {
   addSideUsage,
   gateProjectSettings
 };
-//# sourceMappingURL=chunk-735BKABA.js.map
+//# sourceMappingURL=chunk-UKIE4AAM.js.map

package/dist/cli/index.js

@@ -15,7 +15,11 @@ import {
   runCommand,
   runTurnInTerminal,
   stopMessage
-} from "../chunk-Z66BWTMD.js";
+} from "../chunk-OQ26W5L4.js";
+import {
+  SwitchApprover,
+  TerminalApprover
+} from "../chunk-4JSUUDFQ.js";
 import {
   PlainRenderer
 } from "../chunk-FARS4R67.js";
@@ -23,13 +27,13 @@ import {
   AutoApprover
 } from "../chunk-NXJ4XLB5.js";
 import {
-  Notifier,
-  pickChannel
-} from "../chunk-N4I6YS3E.js";
-import {
-  SwitchApprover,
-  TerminalApprover
-} from "../chunk-4JSUUDFQ.js";
+  NO_MODEL,
+  chooseModel,
+  keepProviderKeys,
+  loadModelsConfig,
+  providerKeySource,
+  resolveModel
+} from "../chunk-2FIV2VZP.js";
 import {
   DEFAULT_MAX_TOKENS,
   Runtime,
@@ -39,9 +43,16 @@ import {
   rebuildState,
   runAgent,
   signature
-} from "../chunk-2DYMJMRQ.js";
+} from "../chunk-NHWHG3SW.js";
+import {
+  loadCatalog
+} from "../chunk-65NUEOI7.js";
 import "../chunk-G2AX7MGS.js";
 import "../chunk-A2QAIRQL.js";
+import {
+  Notifier,
+  pickChannel
+} from "../chunk-N4I6YS3E.js";
 import {
   FileSessionStore,
   closeOpenToolCalls,
@@ -52,19 +63,8 @@ import {
   popPoint,
   pushPoint,
   redoPoint
-} from "../chunk-735BKABA.js";
-import {
-  NO_MODEL,
-  chooseModel,
-  keepProviderKeys,
-  loadModelsConfig,
-  providerKeySource,
-  resolveModel
-} from "../chunk-JRVWH27M.js";
-import {
-  loadCatalog
-} from "../chunk-U65DQZDB.js";
-import "../chunk-J4346IDH.js";
+} from "../chunk-UKIE4AAM.js";
+import "../chunk-7KBOIPUT.js";
 import {
   loadSearchConfig
 } from "../chunk-B25VL4RU.js";
@@ -74,11 +74,11 @@ import {
 import "../chunk-EU7PA45O.js";
 import {
   VERSION
-} from "../chunk-BSK5T2ZG.js";
+} from "../chunk-C57GHLKP.js";
 import "../chunk-5DEWP3YX.js";
-import "../chunk-ZAE25JM2.js";
-import "../chunk-ATW7BNIG.js";
 import "../chunk-PZS3V6P3.js";
+import "../chunk-QANRZW7W.js";
+import "../chunk-ATW7BNIG.js";
 import "../chunk-AOGFYUKG.js";
 import {
   JOB_DENIAL
@@ -784,10 +784,23 @@ async function main() {
   ).option("--verbose", "with json output: also show the tool activity on stderr").action(async (options) => {
     process.exitCode = await start(options, program);
   });
-  program.command("eval").description("run the eval tasks in scratch folders and report pass/fail, steps and cost").option("-m, --model <id>", "model id (or set GARUDA_MODEL)").option("-s, --suite <name>", "basic (default), hard, java, python, all or repo").option("-t, --task <ids...>", "run only these tasks (from any suite)").option("--max-steps <n>", "step limit per task", (v) => Number.parseInt(v, 10)).option(
+  const evalCommand = program.command("eval").description("run the eval tasks in scratch folders and report pass/fail, steps and cost").option("-m, --model <id>", "model id (or set GARUDA_MODEL)").option("-s, --suite <name>", "basic (default), hard, java, python, medium, all or repo").option(
+    "-t, --task <ids...>",
+    "run only these tasks (from any dev suite, or the held-out set with --split)"
+  ).option("--max-steps <n>", "step limit per task", (v) => Number.parseInt(v, 10)).option(
     "--repeat <n>",
     "run each task n times and show the mean",
     (v) => Number.parseInt(v, 10)
+  ).option("--agent <name>", "garuda (default) or claude-code: the baseline on the same tasks").option(
+    "--split <name>",
+    "dev (default), heldout or all: the held-out tasks run only on request"
+  ).option("--addendum <file>", "add this text (4000 characters at most) to the system prompt").option("--mcp <server>", "allow this one server from ~/.garuda/mcp.json (an MCP A/B)").option(
+    "--mcp-prepare <command>",
+    "with --mcp: run this in each scratch folder before the agent (e.g. graphify update .)"
+  ).option("--shuffle", "run the tasks and their repeats in a random order from --seed").option(
+    "--seed <n>",
+    "seed for --shuffle and the report's interval (default 1)",
+    (v) => Number(v)
   ).option("--index <mode>", "code index tools for the model: off (default), lookup or all").option("--executor <name>", "auto (default), os or host").option("--keep", "keep the scratch folders").option("--list", "list the tasks and exit").option("--subagents <mode>", "the explore subagent: off (default) or on, for A/B runs").option("--subagent-model <id>", "model of the explore subagent; default: the main model").option("--todo <mode>", "the todo_write tool: off (default) or on, for A/B runs").option("--lsp <mode>", "language server errors in edit results: off (default) or on").option(
     "--format <mode>",
     "the project's formatter after edits: off (default) or on, for A/B runs"
@@ -819,13 +832,33 @@ async function main() {
     "--test-command <cmd>",
     "with --from-git: the test command; {files} takes the test files (default: detected)"
   ).action(async (options) => {
-    const { runEvalCommand } = await import("../evalCommand-YVQAWUZE.js");
+    const { runEvalCommand } = await import("../evalCommand-SKJQ6CW2.js");
     process.exitCode = await runEvalCommand({
       ...options,
       model: options.model ?? process.env.GARUDA_MODEL,
       subagentModel: options.subagentModel ?? process.env.GARUDA_SUBAGENT_MODEL
     });
   });
+  evalCommand.command("tune").description(
+    "improve a prompt addendum: a proposer model edits it, an edit stays when the dev score improves, the held-out set is measured once"
+  ).requiredOption(
+    "--addendum <file>",
+    "the starting addendum (a missing file starts empty); never edited"
+  ).option("--proposer <model>", "the model that proposes edits (default claude-opus-5-5)").option("-m, --model <id>", "the agent's model (or set GARUDA_MODEL)").option("-s, --suite <names...>", "dev suites (default: basic hard medium)").option("--rounds <n>", "proposals (default 5)", (v) => Number(v)).option("--repeat <n>", "runs of each dev task per score (default 3)", (v) => Number(v)).option("--heldout-repeat <n>", "runs of each held-out task (default 5)", (v) => Number(v)).option("--margin <n>", "more passing runs that keep an edit (default 2)", (v) => Number(v)).option(
+    "--max-cost <usd>",
+    "stop before a round that would pass this (default 20)",
+    (v) => Number(v)
+  ).option("--seed <n>", "seed for the run order and the intervals (default 1)", (v) => Number(v)).option("--yes", "start without the question").option("--parallel <n>", "runs at the same time in each score (default 4)", (v) => Number(v)).action(async (options) => {
+    const { runEvalTune } = await import("../evalTune-CZRCPSZ3.js");
+    process.exitCode = await runEvalTune({
+      ...options,
+      model: options.model ?? process.env.GARUDA_MODEL
+    });
+  });
+  evalCommand.command("compare").description("compare two eval reports of the same tasks (pass rates with a paired interval)").argument("<a>", "report.json, or its run folder (.garuda/evals/<run-id>)").argument("<b>", "the second report").option("--seed <n>", "seed for the bootstrap (default 1)", (v) => Number(v)).action(async (a, b, options) => {
+    const { runEvalCompare } = await import("../evalCommand-SKJQ6CW2.js");
+    process.exitCode = await runEvalCompare(a, b, options);
+  });
   program.command("init").description(
     "set up this folder (AGENTS.md; commands, MCP servers and rules from Claude Code, OpenCode, Codex, Gemini CLI, Tabnine, Cursor, Copilot), then chat"
   ).option("-m, --model <id>", "model id (or set GARUDA_MODEL)").action(async (options) => {
@@ -841,19 +874,19 @@ async function main() {
   program.command("night").description(
     "run this project's night queue (jobs from /schedule), up to 3 at a time, then write one digest"
   ).option("--at <HH:MM>", "wait until this local time first (keep the terminal open)").option("--parallel <n>", "jobs at the same time (1 to 10, default 3)").addOption(new Option("--from-launchd").hideHelp()).action(async (options) => {
-    const { nightCommand } = await import("../nightCommand-OODJNWDT.js");
+    const { nightCommand } = await import("../nightCommand-WJU6VZA6.js");
     process.exitCode = await nightCommand(options, new PlainRenderer());
   });
   program.command("setup").description(
     "choose a provider and model and store its key (~/.garuda/credentials, only you can read it); editors run it to sign in"
   ).option("--show", "show the model and where each key comes from (keys masked)").option("--forget [key-name]", "remove a stored key, or all of them").action(async (options) => {
-    const { setupCommand, terminalIO } = await import("../setupCommand-GBQIVYJC.js");
+    const { setupCommand, terminalIO } = await import("../setupCommand-YMF6I3MD.js");
     process.exitCode = await setupCommand(options, { io: await terminalIO() });
   });
   program.command("acp").description(
     "run as an agent for an editor over the Agent Client Protocol (VS Code with an extension, Zed, JetBrains, Neovim, Emacs); the editor starts it"
   ).argument("[action]", "setup: run garuda setup (the editor's sign-in, ACP Terminal Auth)").option("-m, --model <id>", "model id (or set GARUDA_MODEL)").action(async (action, options) => {
-    const { acpCommand } = await import("../acpCommand-7EY3NHPG.js");
+    const { acpCommand } = await import("../acpCommand-IUZPLPLI.js");
     process.exitCode = await acpCommand({
       ...options,
       ...action === void 0 ? {} : { action }
@@ -876,7 +909,7 @@ async function start(options, program) {
   let job;
   let jobGuard;
   if (options.job !== void 0) {
-    const { prepareJob } = await import("../jobCommand-FE6XN3YD.js");
+    const { prepareJob } = await import("../jobCommand-2ODQLHB3.js");
     const prepared = await prepareJob(
       mainRoot,
       options.job,
@@ -1115,7 +1148,7 @@ async function start(options, program) {
       if (resolved2.kind === "prompt") prompt = resolved2.prompt;
     }
     if (job !== void 0) {
-      const { testsBefore } = await import("../jobCommand-FE6XN3YD.js");
+      const { testsBefore } = await import("../jobCommand-2ODQLHB3.js");
       await testsBefore(job, runtime, renderer);
     }
     const outcome = await runTurnInTerminal(
@@ -1128,7 +1161,7 @@ async function start(options, program) {
     await runtime.close();
     finish(outcome);
     if (job !== void 0) {
-      const { finishJob } = await import("../jobCommand-FE6XN3YD.js");
+      const { finishJob } = await import("../jobCommand-2ODQLHB3.js");
       await finishJob(job, outcome, runtime, renderer, void 0, {
         fromLaunchd: options.fromLaunchd === true
       });
@@ -1214,7 +1247,7 @@ function wantsInk() {
 }
 async function loadInk(renderer) {
   try {
-    return await import("../inkChat-FNUBK43U.js");
+    return await import("../inkChat-WYYAFZXD.js");
   } catch (error) {
     renderer.info(`Plain chat (Ink is not available: ${error.message.split("\n")[0]}).`);
     return void 0;

package/dist/config-WV2H64TF.js

@@ -0,0 +1,23 @@
+#!/usr/bin/env node
+import {
+  MCP_FILE,
+  SERVER_NAME,
+  checkServerUrl,
+  commandLine,
+  defHash,
+  expandEnv,
+  isHttp,
+  loadMcpConfig
+} from "./chunk-QANRZW7W.js";
+import "./chunk-ATW7BNIG.js";
+export {
+  MCP_FILE,
+  SERVER_NAME,
+  checkServerUrl,
+  commandLine,
+  defHash,
+  expandEnv,
+  isHttp,
+  loadMcpConfig
+};
+//# sourceMappingURL=config-WV2H64TF.js.map