@garuda-agent/garuda 0.17.0 → 0.18.1

package/dist/evalTune-W53KAD4V.js

@@ -0,0 +1,302 @@
+#!/usr/bin/env node
+import {
+  PROPOSER_SYSTEM,
+  parseProposal,
+  readAddendum,
+  tune
+} from "./chunk-IKPGHM54.js";
+import {
+  EVAL_SUITES,
+  HELD_OUT_TASKS,
+  requiredToolchains
+} from "./chunk-GKU62MHQ.js";
+import {
+  formatReport,
+  reportV2,
+  runEvals
+} from "./chunk-DUYLVGPY.js";
+import {
+  DEFAULT_SEED,
+  TOOLCHAINS,
+  checkToolchains,
+  shuffled
+} from "./chunk-CWVXOSAA.js";
+import "./chunk-H7LIBGZT.js";
+import "./chunk-YAQ6QLKB.js";
+import "./chunk-NXJ4XLB5.js";
+import "./chunk-5UDYQ5MK.js";
+import "./chunk-G2AX7MGS.js";
+import "./chunk-ULAJPMY7.js";
+import "./chunk-JMH5W5DG.js";
+import {
+  NO_MODEL,
+  chooseModel,
+  keepProviderKeys,
+  loadModelsConfig,
+  resolveModel
+} from "./chunk-QLBCM2YQ.js";
+import "./chunk-NH72AELP.js";
+import "./chunk-4QRTTRVK.js";
+import "./chunk-A2QAIRQL.js";
+import {
+  newSessionId
+} from "./chunk-NLS7GMF7.js";
+import {
+  costOf
+} from "./chunk-7KBOIPUT.js";
+import "./chunk-B25VL4RU.js";
+import "./chunk-MNE7HHX2.js";
+import "./chunk-IYGJWX65.js";
+import "./chunk-5DEWP3YX.js";
+import "./chunk-PZS3V6P3.js";
+import "./chunk-QANRZW7W.js";
+import "./chunk-ATW7BNIG.js";
+import "./chunk-WK5NE6RY.js";
+import "./chunk-AOGFYUKG.js";
+import "./chunk-QXTHEFKP.js";
+import {
+  createExecutor
+} from "./chunk-LMKGSYJC.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-W53KAD4V.js.map

package/dist/{export-SJ3DUP6W.js → export-E677HIQS.js}

@@ -2,11 +2,20 @@
 import {
   summariseCall,
   summariseResult
-} from "./chunk-FARS4R67.js";
+} from "./chunk-DQSCCE77.js";
+import {
+  hasReceipt
+} from "./chunk-4QRTTRVK.js";
 import {
   serverCallText,
   serverResultSummary
 } from "./chunk-A2QAIRQL.js";
+import {
+  receiptText
+} from "./chunk-2OYTNNUJ.js";
+import "./chunk-NLS7GMF7.js";
+import "./chunk-7KBOIPUT.js";
+import "./chunk-BHZ6G65V.js";
 import {
   PathOutsideRootError,
   displayPath,
@@ -98,6 +107,20 @@ function sessionMarkdown(records, sessionId) {
       case "compaction":
         out.push("_The conversation was compacted here to fit the context window._", "");
         break;
+      case "fork":
+        out.push(
+          `_This session starts as a fork of session ${record.from.sessionId}, at turn ${record.from.turn}._`,
+          ""
+        );
+        break;
+      case "receipt":
+        if (hasReceipt(record.receipt)) {
+          out.push(
+            ...receiptText(record.receipt).split("\n").map((line) => `> ${line}`),
+            ""
+          );
+        }
+        break;
       case "model":
         out.push(`_The model changed to ${record.model}._`, "");
         break;
@@ -129,8 +152,8 @@ function when(iso) {
   const pad = (n) => String(n).padStart(2, "0");
   return `${d.getFullYear()}-${pad(d.getMonth() + 1)}-${pad(d.getDate())} ${pad(d.getHours())}:${pad(d.getMinutes())}`;
 }
-async function writeExport(root, sessionId, markdown, file = "") {
-  const name = file === "" ? `garuda-${sessionId}.md` : file;
+async function writeExport(root, sessionId, markdown, file = "", ext = ".md") {
+  const name = file === "" ? `garuda-${sessionId}${ext}` : file;
   let absolute;
   try {
     absolute = await resolveInRoot(root, name);
@@ -155,4 +178,4 @@ export {
   sessionMarkdown,
   writeExport
 };
-//# sourceMappingURL=export-SJ3DUP6W.js.map
+//# sourceMappingURL=export-E677HIQS.js.map

package/dist/exportHtml-BMV73YRO.js

@@ -0,0 +1,21 @@
+#!/usr/bin/env node
+import {
+  OUTPUT_CHARS,
+  callChange,
+  escapeHtml,
+  sessionHtml
+} from "./chunk-TLA2PDGM.js";
+import "./chunk-DQSCCE77.js";
+import "./chunk-4QRTTRVK.js";
+import "./chunk-A2QAIRQL.js";
+import "./chunk-2OYTNNUJ.js";
+import "./chunk-NLS7GMF7.js";
+import "./chunk-7KBOIPUT.js";
+import "./chunk-BHZ6G65V.js";
+export {
+  OUTPUT_CHARS,
+  callChange,
+  escapeHtml,
+  sessionHtml
+};
+//# sourceMappingURL=exportHtml-BMV73YRO.js.map