@garuda-agent/garuda 0.17.0 → 0.18.0

package/dist/evalTune-CZRCPSZ3.js

@@ -0,0 +1,295 @@
+#!/usr/bin/env node
+import {
+  PROPOSER_SYSTEM,
+  parseProposal,
+  readAddendum,
+  tune
+} from "./chunk-FOPBOB7W.js";
+import {
+  EVAL_SUITES,
+  HELD_OUT_TASKS,
+  TOOLCHAINS,
+  checkToolchains,
+  formatReport,
+  requiredToolchains,
+  runEvals
+} from "./chunk-Q4P3W657.js";
+import {
+  DEFAULT_SEED,
+  reportV2,
+  shuffled
+} from "./chunk-DOCDCQVM.js";
+import "./chunk-NXJ4XLB5.js";
+import {
+  NO_MODEL,
+  chooseModel,
+  keepProviderKeys,
+  loadModelsConfig,
+  resolveModel
+} from "./chunk-2FIV2VZP.js";
+import "./chunk-NHWHG3SW.js";
+import "./chunk-65NUEOI7.js";
+import "./chunk-G2AX7MGS.js";
+import "./chunk-A2QAIRQL.js";
+import {
+  newSessionId
+} from "./chunk-UKIE4AAM.js";
+import {
+  costOf
+} from "./chunk-7KBOIPUT.js";
+import "./chunk-B25VL4RU.js";
+import "./chunk-23FXBNBB.js";
+import "./chunk-C57GHLKP.js";
+import "./chunk-5DEWP3YX.js";
+import "./chunk-PZS3V6P3.js";
+import "./chunk-QANRZW7W.js";
+import "./chunk-ATW7BNIG.js";
+import "./chunk-55CFEEIN.js";
+import "./chunk-AOGFYUKG.js";
+import "./chunk-6XHON3Q6.js";
+import {
+  createExecutor
+} from "./chunk-B4SJY7KT.js";
+import "./chunk-BHZ6G65V.js";
+import "./chunk-P6XWWIEK.js";
+import "./chunk-UIRJOQQS.js";
+import "./chunk-4LFJFDCF.js";
+import "./chunk-3NJNVZTK.js";
+
+// src/cli/evalTune.ts
+import { mkdirSync, writeFileSync } from "fs";
+import { writeFile } from "fs/promises";
+import { join } from "path";
+import { createInterface } from "readline/promises";
+var DEFAULT_TUNE_PARALLEL = 4;
+function progressLine(label, done, total, r) {
+  return `[${label} ${done}/${total}] ${r.passed ? "\u2713" : "\u2717"} ${r.id} (${r.steps} steps)${r.passed ? "" : `: ${(r.reason ?? r.stopReason).split("\n")[0]}`}`;
+}
+var DEFAULT_PROPOSER = "claude-opus-5-5";
+var DEFAULT_TUNE_SUITES = ["basic", "hard", "medium"];
+var whole = (value, fallback, min) => {
+  const v = value ?? fallback;
+  return Number.isInteger(v) && v >= min ? v : void 0;
+};
+async function askProposer(proposer, input) {
+  const client = await proposer.create();
+  let response;
+  for await (const event of client.stream({
+    system: PROPOSER_SYSTEM,
+    messages: [{ role: "user", content: [{ type: "text", text: input }] }],
+    tools: [],
+    maxTokens: 8e3
+  })) {
+    if (event.type === "response") response = event.response;
+  }
+  if (response === void 0) return "the proposer gave no answer";
+  const text = response.content.filter((b) => b.type === "text").map((b) => b.text).join("");
+  const parsed = parseProposal(text);
+  if (parsed === void 0) return "the proposer's answer has no <addendum> block";
+  const price = proposer.info.price;
+  return {
+    ...parsed,
+    costUsd: price === void 0 ? void 0 : costOf(response.usage, price, response.priceFactor ?? 1)
+  };
+}
+async function runEvalTune(options) {
+  const rounds = whole(options.rounds, 5, 1);
+  const repeat = whole(options.repeat, 3, 1);
+  const heldoutRepeat = whole(options.heldoutRepeat, 5, 1);
+  const margin = whole(options.margin, 2, 1);
+  const seed = whole(options.seed, DEFAULT_SEED, 0);
+  const parallel = whole(options.parallel, DEFAULT_TUNE_PARALLEL, 1);
+  const maxCost = options.maxCost ?? 20;
+  if (rounds === void 0 || repeat === void 0 || heldoutRepeat === void 0 || margin === void 0 || seed === void 0 || parallel === void 0 || !(maxCost > 0)) {
+    process.stderr.write(
+      "--rounds, --repeat, --heldout-repeat, --margin and --parallel must be whole numbers of 1 or more, --seed 0 or more, and --max-cost above 0.\n"
+    );
+    return 1;
+  }
+  const suites = options.suite ?? [...DEFAULT_TUNE_SUITES];
+  const devTasks = [];
+  for (const name of suites) {
+    const tasks = EVAL_SUITES[name];
+    if (tasks === void 0) {
+      process.stderr.write(`Unknown suite ${name}. Use: ${Object.keys(EVAL_SUITES).join(", ")}.
+`);
+      return 1;
+    }
+    devTasks.push(...tasks);
+  }
+  const start = await readAddendum(options.addendum, true);
+  if ("problem" in start) {
+    process.stderr.write(`${start.problem}
+`);
+    return 1;
+  }
+  const models = await loadModelsConfig();
+  if (models.problem !== void 0) {
+    process.stderr.write(`${models.problem}
+`);
+    return 1;
+  }
+  const modelId = chooseModel(models.config, options.model);
+  if (!modelId) {
+    process.stderr.write(`${NO_MODEL}
+`);
+    return 1;
+  }
+  let agent;
+  let proposer;
+  try {
+    agent = resolveModel(modelId, models.config);
+    proposer = resolveModel(options.proposer ?? DEFAULT_PROPOSER, models.config);
+  } catch (error) {
+    process.stderr.write(`${error.message}
+`);
+    return 1;
+  }
+  const { assertModelAllowedByPolicy, loadTeamPolicy } = await import("./policy-ZJ7RMOFN.js");
+  let team;
+  try {
+    team = await loadTeamPolicy();
+    assertModelAllowedByPolicy(team?.policy, agent.spec);
+    assertModelAllowedByPolicy(team?.policy, proposer.spec);
+  } catch (error) {
+    process.stderr.write(`Team policy: ${error.message}
+`);
+    return 1;
+  }
+  const warning = await keepProviderKeys([agent.def.apiKeyEnv, proposer.def.apiKeyEnv]);
+  if (warning !== void 0) process.stderr.write(`${warning}
+`);
+  const missing = (await checkToolchains(requiredToolchains([...devTasks, ...HELD_OUT_TASKS]))).filter((s) => !s.ok);
+  if (missing.length > 0) {
+    for (const status of missing) process.stderr.write(`${TOOLCHAINS[status.id].hint}
+`);
+    return 1;
+  }
+  const perScore = devTasks.length * repeat;
+  const heldOutRuns = 2 * HELD_OUT_TASKS.length * heldoutRepeat;
+  process.stderr.write(
+    [
+      `Tuning ${options.addendum} (${start.text.length} characters) with ${agent.spec}; ${proposer.spec} proposes.`,
+      `Dev: ${devTasks.length} task(s) from ${suites.join(", ")} \xD7 ${repeat} = ${perScore} runs a score, ${parallel} at a time; up to ${rounds + 1} scores.`,
+      `Held-out: ${HELD_OUT_TASKS.length} task(s) \xD7 ${heldoutRepeat}, twice, at the end (${heldOutRuns} runs), if a version is kept.`,
+      `Keep rule: +${margin} passing runs, or as many with a lower cost per accepted change (interval below \xD71.0).`,
+      `Budget: --max-cost $${maxCost}; the loop stops before a round that would pass it.`,
+      `Garuda changes no file outside the run's folder; ${options.addendum} stays as it is.`,
+      ""
+    ].join("\n")
+  );
+  if (options.yes !== true) {
+    if (!process.stdin.isTTY) {
+      process.stderr.write("Add --yes to start without a question.\n");
+      return 1;
+    }
+    const rl = createInterface({ input: process.stdin, output: process.stderr });
+    const answer = await rl.question("Start? (y/N) ");
+    rl.close();
+    if (!/^y(es)?$/i.test(answer.trim())) return 1;
+  }
+  const { executor, notice } = createExecutor("auto");
+  if (notice !== void 0) process.stderr.write(`${notice}
+`);
+  const dir = join(process.cwd(), ".garuda", "evals", `tune-${newSessionId()}`);
+  mkdirSync(dir, { recursive: true });
+  const score = async (addendum, tasks, k, label) => {
+    const listed = tasks.flatMap((t) => Array.from({ length: k }, () => t));
+    const outDir = join(dir, label);
+    let done = 0;
+    const results = await runEvals(
+      shuffled(listed, seed),
+      {
+        parallel,
+        modelId: agent.spec,
+        model: () => agent.create(),
+        modelInfo: agent.info,
+        ...agent.maxTokens === void 0 ? {} : { maxTokens: agent.maxTokens },
+        ...team === void 0 ? {} : { policy: team.policy },
+        outDir,
+        executor: executor.name === "host" ? "host" : "os",
+        ...addendum === "" ? {} : { addendum }
+      },
+      (r) => {
+        done++;
+        process.stderr.write(`${progressLine(label, done, listed.length, r)}
+`);
+      }
+    );
+    writeFileSync(
+      join(outDir, "report.json"),
+      `${JSON.stringify(
+        {
+          ...await reportV2(
+            {
+              model: agent.spec,
+              provider: agent.provider,
+              executor: executor.name,
+              isolation: executor.isolation,
+              seed,
+              shuffle: true,
+              split: label.startsWith("heldout") ? "heldout" : "dev",
+              ...addendum === "" ? {} : { addendum: { file: `${label}.md`, text: addendum } }
+            },
+            tasks,
+            results
+          ),
+          results
+        },
+        null,
+        2
+      )}
+`
+    );
+    process.stderr.write(`
+${label}:
+${formatReport(results, { seed })}
+
+`);
+    return results;
+  };
+  const result = await tune({
+    start: start.text,
+    devTasks,
+    heldOutTasks: HELD_OUT_TASKS,
+    score,
+    propose: async (input) => {
+      const answer = await askProposer(proposer, input);
+      if (typeof answer === "string") {
+        return { addendum: "", reason: answer, costUsd: 0 };
+      }
+      return answer;
+    },
+    rounds,
+    repeat,
+    heldoutRepeat,
+    margin,
+    maxCostUsd: maxCost,
+    seed,
+    save: (name, text) => writeFile(join(dir, name), text),
+    log: (line) => process.stderr.write(`${line}
+`)
+  });
+  const kept = result.history.filter((h) => h.decision?.kept === true);
+  process.stdout.write(
+    [
+      "",
+      `Rounds: ${result.history.length}; kept: ${kept.length === 0 ? "none" : kept.map((h) => h.version).join(", ")}. Spent about $${result.spentUsd.toFixed(2)}${result.stopped === "rounds" ? "" : ` (stopped: ${result.stopped})`}.`,
+      ...result.heldOut === void 0 ? [] : ["", "Held-out (the only number to quote):", result.heldOut],
+      "",
+      `The final addendum: ${join(dir, "final.md")}. To use it, copy it into GARUDA.md yourself.`,
+      `Every version, the proposer's inputs and history.json: ${dir}`,
+      ""
+    ].join("\n")
+  );
+  return 0;
+}
+export {
+  DEFAULT_PROPOSER,
+  DEFAULT_TUNE_PARALLEL,
+  DEFAULT_TUNE_SUITES,
+  askProposer,
+  progressLine,
+  runEvalTune
+};
+//# sourceMappingURL=evalTune-CZRCPSZ3.js.map

package/dist/{inkChat-FNUBK43U.js → inkChat-WYYAFZXD.js}

@@ -6,7 +6,11 @@ import {
   fuzzyFiles,
   runChat,
   statusOf
-} from "./chunk-Z66BWTMD.js";
+} from "./chunk-OQ26W5L4.js";
+import {
+  colorPreview,
+  header
+} from "./chunk-4JSUUDFQ.js";
 import {
   retryText,
   serverToolOutput,
@@ -15,24 +19,20 @@ import {
   summariseResult,
   todoLines
 } from "./chunk-FARS4R67.js";
-import {
-  colorPreview,
-  header
-} from "./chunk-4JSUUDFQ.js";
-import "./chunk-2DYMJMRQ.js";
+import "./chunk-NHWHG3SW.js";
+import "./chunk-65NUEOI7.js";
 import "./chunk-G2AX7MGS.js";
 import "./chunk-A2QAIRQL.js";
-import "./chunk-735BKABA.js";
-import "./chunk-U65DQZDB.js";
-import "./chunk-J4346IDH.js";
+import "./chunk-UKIE4AAM.js";
+import "./chunk-7KBOIPUT.js";
 import "./chunk-B25VL4RU.js";
 import "./chunk-23FXBNBB.js";
 import "./chunk-EU7PA45O.js";
-import "./chunk-BSK5T2ZG.js";
+import "./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 "./chunk-B4SJY7KT.js";
 import "./chunk-BHZ6G65V.js";
@@ -1194,4 +1194,4 @@ async function runInkChat(runtime, approver, setEventTarget, banner, sessionPath
 export {
   runInkChat
 };
-//# sourceMappingURL=inkChat-FNUBK43U.js.map
+//# sourceMappingURL=inkChat-WYYAFZXD.js.map

package/dist/{jobCommand-FE6XN3YD.js → jobCommand-2ODQLHB3.js}

@@ -5,10 +5,10 @@ import {
   msUntil,
   prepareJob,
   testsBefore
-} from "./chunk-TJIWD7R7.js";
+} from "./chunk-4BVCYCQT.js";
 import "./chunk-N4I6YS3E.js";
-import "./chunk-735BKABA.js";
-import "./chunk-J4346IDH.js";
+import "./chunk-UKIE4AAM.js";
+import "./chunk-7KBOIPUT.js";
 import "./chunk-5DEWP3YX.js";
 import "./chunk-PZS3V6P3.js";
 import "./chunk-55CFEEIN.js";
@@ -30,4 +30,4 @@ export {
   prepareJob,
   testsBefore
 };
-//# sourceMappingURL=jobCommand-FE6XN3YD.js.map
+//# sourceMappingURL=jobCommand-2ODQLHB3.js.map

package/dist/{manager-73MPYFTP.js → manager-TCYTA5XU.js}

@@ -11,26 +11,26 @@ import {
 } from "./chunk-4P26OKTP.js";
 import {
   VERSION
-} from "./chunk-BSK5T2ZG.js";
+} from "./chunk-C57GHLKP.js";
 import {
   USER_SCOPE
 } from "./chunk-5DEWP3YX.js";
+import {
+  capText,
+  cleanJson,
+  cleanLine,
+  cleanText,
+  neutralizeTags
+} from "./chunk-PZS3V6P3.js";
 import {
   commandLine,
   defHash,
   expandEnv,
   isHttp
-} from "./chunk-ZAE25JM2.js";
+} from "./chunk-QANRZW7W.js";
 import {
   isLoopbackHost
 } from "./chunk-ATW7BNIG.js";
-import {
-  capText,
-  cleanJson,
-  cleanLine,
-  cleanText,
-  neutralizeTags
-} from "./chunk-PZS3V6P3.js";
 import {
   DEFAULT_ENV_ALLOWLIST,
   sandboxPaths
@@ -846,4 +846,4 @@ export {
   changeLines,
   warnings
 };
-//# sourceMappingURL=manager-73MPYFTP.js.map
+//# sourceMappingURL=manager-TCYTA5XU.js.map

package/dist/{nightCommand-OODJNWDT.js → nightCommand-WJU6VZA6.js}

@@ -1,13 +1,13 @@
 #!/usr/bin/env node
 import {
   msUntil
-} from "./chunk-TJIWD7R7.js";
+} from "./chunk-4BVCYCQT.js";
 import {
   notificationBytes,
   pickChannel
 } from "./chunk-N4I6YS3E.js";
-import "./chunk-735BKABA.js";
-import "./chunk-J4346IDH.js";
+import "./chunk-UKIE4AAM.js";
+import "./chunk-7KBOIPUT.js";
 import "./chunk-5DEWP3YX.js";
 import "./chunk-PZS3V6P3.js";
 import "./chunk-55CFEEIN.js";
@@ -252,4 +252,4 @@ Digest: ${file}`);
 export {
   nightCommand
 };
-//# sourceMappingURL=nightCommand-OODJNWDT.js.map
+//# sourceMappingURL=nightCommand-WJU6VZA6.js.map

package/dist/{run-ML6GSPN4.js → run-PTMRQCIU.js}

@@ -1,4 +1,7 @@
 #!/usr/bin/env node
+import {
+  folderKind
+} from "./chunk-42DSA4EA.js";
 import {
   loadSearchConfig,
   providerLabel,
@@ -10,11 +13,8 @@ import {
 import {
   SERVER_NAME,
   checkServerUrl
-} from "./chunk-ZAE25JM2.js";
+} from "./chunk-QANRZW7W.js";
 import "./chunk-ATW7BNIG.js";
-import {
-  folderKind
-} from "./chunk-42DSA4EA.js";
 
 // src/init/run.ts
 import { existsSync as existsSync3 } from "fs";
@@ -885,4 +885,4 @@ export {
   initPrompt,
   runInit
 };
-//# sourceMappingURL=run-ML6GSPN4.js.map
+//# sourceMappingURL=run-PTMRQCIU.js.map

package/dist/{setupCommand-GBQIVYJC.js → setupCommand-YMF6I3MD.js}

@@ -11,14 +11,14 @@ import {
   resolveModel,
   setDefaultModel,
   writeCredentials
-} from "./chunk-JRVWH27M.js";
+} from "./chunk-2FIV2VZP.js";
 import {
   fetchCatalog,
   loadCatalog,
   mergedModels,
   saveCatalog
-} from "./chunk-U65DQZDB.js";
-import "./chunk-J4346IDH.js";
+} from "./chunk-65NUEOI7.js";
+import "./chunk-7KBOIPUT.js";
 import "./chunk-ATW7BNIG.js";
 import "./chunk-BHZ6G65V.js";
 
@@ -332,4 +332,4 @@ export {
   setupCommand,
   terminalIO
 };
-//# sourceMappingURL=setupCommand-GBQIVYJC.js.map
+//# sourceMappingURL=setupCommand-YMF6I3MD.js.map

package/dist/tune-YVI2KWQF.js

@@ -0,0 +1,31 @@
+#!/usr/bin/env node
+import {
+  ADDENDUM_MAX_CHARS,
+  PROPOSER_SYSTEM,
+  decide,
+  parseProposal,
+  proposerInput,
+  readAddendum,
+  rejectReason,
+  tune
+} from "./chunk-FOPBOB7W.js";
+import "./chunk-DOCDCQVM.js";
+import "./chunk-C57GHLKP.js";
+import "./chunk-6XHON3Q6.js";
+import "./chunk-B4SJY7KT.js";
+import "./chunk-BHZ6G65V.js";
+import "./chunk-P6XWWIEK.js";
+import "./chunk-UIRJOQQS.js";
+import "./chunk-4LFJFDCF.js";
+import "./chunk-3NJNVZTK.js";
+export {
+  ADDENDUM_MAX_CHARS,
+  PROPOSER_SYSTEM,
+  decide,
+  parseProposal,
+  proposerInput,
+  readAddendum,
+  rejectReason,
+  tune
+};
+//# sourceMappingURL=tune-YVI2KWQF.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@garuda-agent/garuda",
-  "version": "0.17.0",
+  "version": "0.18.0",
   "description": "Garuda: a terminal coding agent with an OS sandbox, approvals and an audit log.",
   "license": "Apache-2.0",
   "author": "Madhusudhan Vasanth Kumar",

package/dist/chunk-BSK5T2ZG.js

@@ -1,9 +0,0 @@
-#!/usr/bin/env node
-
-// src/version.ts
-var VERSION = "0.17.0";
-
-export {
-  VERSION
-};
-//# sourceMappingURL=chunk-BSK5T2ZG.js.map