@garuda-agent/garuda 0.18.0 → 0.18.2

package/dist/{evalCommand-SKJQ6CW2.js → evalCommand-MRO6ENCA.js}

@@ -1,37 +1,42 @@
 #!/usr/bin/env node
 import {
-  DEFAULT_COMMITS,
   EVAL_SUITES,
   HELD_OUT_TASKS,
+  chooseTasks,
+  requiredToolchains
+} from "./chunk-LWNMEGDF.js";
+import {
+  reportV2
+} from "./chunk-6JQTPW3P.js";
+import {
+  garudaCommit,
+  staleBuildNote
+} from "./chunk-YAQ6QLKB.js";
+import {
+  BUILD_COMMIT
+} from "./chunk-E6YRCDSF.js";
+import {
+  DEFAULT_COMMITS,
+  DEFAULT_SEED,
   TOOLCHAINS,
   buildRepoSuite,
   checkToolchains,
-  chooseTasks,
+  compareReports,
+  formatComparison,
   formatReport,
   isAuthFailure,
   isMcpFailure,
   isSetupFailure,
   loadRepoSuite,
+  loadReport,
   prepareToolchain,
   repoEvalTasks,
-  requiredToolchains,
   runEvals,
   saveRepoSuite,
+  shuffled,
   summarize,
   toolchainId
-} from "./chunk-Q4P3W657.js";
-import {
-  BUILD_COMMIT,
-  DEFAULT_SEED,
-  compareReports,
-  formatComparison,
-  garudaCommit,
-  loadReport,
-  reportV2,
-  shuffled,
-  staleBuildNote
-} from "./chunk-DOCDCQVM.js";
-import "./chunk-NXJ4XLB5.js";
+} from "./chunk-A5DLAXB4.js";
 import {
   NO_MODEL,
   chooseModel,
@@ -40,38 +45,41 @@ import {
   providerKeyValue,
   resolveModel
 } from "./chunk-2FIV2VZP.js";
+import "./chunk-NXJ4XLB5.js";
 import {
   CODE_INDEX_MODES,
   DEFAULT_CODE_INDEX_MODE,
   detectProfiles
-} from "./chunk-NHWHG3SW.js";
-import "./chunk-65NUEOI7.js";
+} from "./chunk-STVFWSIL.js";
 import "./chunk-G2AX7MGS.js";
+import "./chunk-ULAJPMY7.js";
+import "./chunk-65NUEOI7.js";
+import "./chunk-4QRTTRVK.js";
 import "./chunk-A2QAIRQL.js";
 import {
   newSessionId
-} from "./chunk-UKIE4AAM.js";
+} from "./chunk-NLS7GMF7.js";
 import "./chunk-7KBOIPUT.js";
 import "./chunk-B25VL4RU.js";
-import "./chunk-23FXBNBB.js";
-import "./chunk-C57GHLKP.js";
+import "./chunk-MNE7HHX2.js";
+import "./chunk-CFDRP6XW.js";
 import "./chunk-5DEWP3YX.js";
 import "./chunk-PZS3V6P3.js";
 import "./chunk-QANRZW7W.js";
 import "./chunk-ATW7BNIG.js";
-import "./chunk-55CFEEIN.js";
+import "./chunk-WK5NE6RY.js";
 import {
   detectTestCommand
 } from "./chunk-AOGFYUKG.js";
 import {
   git
-} from "./chunk-6XHON3Q6.js";
+} from "./chunk-QXTHEFKP.js";
 import {
   EXECUTOR_NAMES,
   HostExecutor,
   createExecutor,
   expandAllowlist
-} from "./chunk-B4SJY7KT.js";
+} from "./chunk-LMKGSYJC.js";
 import "./chunk-BHZ6G65V.js";
 import "./chunk-P6XWWIEK.js";
 import "./chunk-UIRJOQQS.js";
@@ -83,6 +91,7 @@ import { existsSync, mkdirSync, realpathSync, writeFileSync } from "fs";
 import { createRequire } from "module";
 import { homedir } from "os";
 import { delimiter, dirname, join } from "path";
+var progressHead = (label, done, total) => label === void 0 ? "" : `[${label} ${done}/${total}] `;
 var BATCH_TASK_TIMEOUT_MS = 12 * 60 * 6e4;
 async function runEvalCommand(options) {
   if (options.fromGit) return buildFromGit(options);
@@ -316,7 +325,7 @@ ${detail === "" ? "" : `    ${detail}
   }
   let addendum;
   if (options.addendum !== void 0) {
-    const { readAddendum } = await import("./tune-YVI2KWQF.js");
+    const { readAddendum } = await import("./tune-KSNKBPNJ.js");
     const read = await readAddendum(options.addendum);
     if ("problem" in read) {
       process.stderr.write(`${read.problem}
@@ -359,7 +368,7 @@ ${detail === "" ? "" : `    ${detail}
     1,
     options.parallel ?? (batchMode === "on" ? Math.min(tasks.length, 20) : 1)
   );
-  const outDir = join(process.cwd(), ".garuda", "evals", newSessionId());
+  const outDir = options.outDir ?? join(process.cwd(), ".garuda", "evals", newSessionId());
   mkdirSync(outDir, { recursive: true });
   const stale = staleBuildNote(BUILD_COMMIT, await garudaCommit());
   if (stale !== void 0) process.stderr.write(`Warning: ${stale}
@@ -378,6 +387,7 @@ ${where}, in scratch folders.
     );
   }
   const started = Date.now();
+  let done = 0;
   const results = await runEvals(
     tasks,
     {
@@ -405,8 +415,10 @@ ${where}, in scratch folders.
       executor: executor.name === "host" ? "host" : "os",
       ...options.keep ? { keep: true } : {},
       onEvent: (taskId, event) => {
-        if (event.type === "tool_call") process.stderr.write(`  [${taskId}] ${event.call.name}
+        if (event.type === "tool_call" && options.progress === void 0) {
+          process.stderr.write(`  [${taskId}] ${event.call.name}
 `);
+        }
         if (event.type === "notice") process.stderr.write(`  [${taskId}] ${event.text}
 `);
         if (event.type === "model_retry")
@@ -416,11 +428,15 @@ ${where}, in scratch folders.
           );
       }
     },
-    (r) => process.stderr.write(
-      `${r.passed ? "\u2713" : "\u2717"} ${r.id}: ${r.passed ? "pass" : "fail"} (${r.steps} steps)${r.reason ? `
+    (r) => {
+      done++;
+      const head = progressHead(options.progress, done, tasks.length);
+      process.stderr.write(
+        `${head}${r.passed ? "\u2713" : "\u2717"} ${r.id}: ${r.passed ? "pass" : "fail"} (${r.steps} steps)${r.reason ? `
     ${r.reason.split("\n").slice(0, 3).join("\n    ")}` : ""}
 `
-    ),
+      );
+    },
     // A refused key or an unreachable API fails every run: stop at the first one (0.17).
     isSetupFailure
   );
@@ -495,20 +511,35 @@ async function buildFromGit(options) {
       return 1;
     }
   }
+  const { loadTeamPolicy } = await import("./policy-ZJ7RMOFN.js");
+  let team;
+  try {
+    team = await loadTeamPolicy();
+  } catch (error) {
+    process.stderr.write(`Team policy: ${error.message}
+`);
+    return 1;
+  }
+  const { checkAsEval } = await import("./runner-EBF6EOYM.js");
   process.stderr.write(
     `Building the repo suite from ${options.commits ?? DEFAULT_COMMITS} recent commits${options.since === void 0 ? "" : ` since ${options.since}`} in ${root}.
-Each candidate runs its own test files with \`${testCommand}\` twice on this machine (no sandbox), in a git worktree; your checkout does not change.
+Each candidate runs its own test files with \`${testCommand}\` twice on this machine (no sandbox), in a git worktree; your checkout does not change. A kept one runs a third time as an eval runs it: in a copy, in the sandbox.
 
 `
   );
-  const { suite, checked, skipped, failure } = await buildRepoSuite(new HostExecutor(), root, {
-    testCommand,
-    ...options.commits === void 0 ? {} : { commits: options.commits },
-    ...options.since === void 0 ? {} : { since: options.since },
-    ...options.maxTasks === void 0 ? {} : { maxTasks: options.maxTasks },
-    onProgress: (line) => process.stderr.write(`  ${line}
+  const { suite, checked, skipped, failure, evalFailure } = await buildRepoSuite(
+    new HostExecutor(),
+    root,
+    {
+      testCommand,
+      evalCheck: (sha, check) => checkAsEval(root, sha, check, team === void 0 ? {} : { policy: team.policy }),
+      ...options.commits === void 0 ? {} : { commits: options.commits },
+      ...options.since === void 0 ? {} : { since: options.since },
+      ...options.maxTasks === void 0 ? {} : { maxTasks: options.maxTasks },
+      onProgress: (line) => process.stderr.write(`  ${line}
 `)
-  });
+    }
+  );
   const file = await saveRepoSuite(suite);
   const reasons = Object.entries(skipped).sort((a, b) => b[1] - a[1]).map(([why, n]) => `  ${String(n).padStart(4)}  ${why}`);
   process.stdout.write(
@@ -521,6 +552,11 @@ ${suite.tasks.length} task(s) kept; ${checked} candidate(s) checked with the tes
         ...failure.tail.split("\n").map((line) => `  ${line}`),
         'If every commit fails like this, the tests do not run in a worktree on this machine; give a working command with --test-command ("{files}" takes the test files).'
       ],
+      ...evalFailure === void 0 ? [] : [
+        `The first commit whose tests pass here but fail as an eval runs them (${evalFailure.sha.slice(0, 7)}: \`${evalFailure.command}\`):`,
+        ...evalFailure.tail.split("\n").map((line) => `  ${line}`),
+        "No model could pass such a task, so it is left out. A test that needs the network, or starts its own sandbox, fails like this."
+      ],
       `Suite: ${file}`,
       ...suite.tasks.length === 0 ? ["No task: try more commits (--commits 500) or an older --since."] : ["Run it: garuda eval -s repo -m <model> (add --list to see the tasks)"]
     ].join("\n")}
@@ -680,7 +716,9 @@ export {
   BATCH_TASK_TIMEOUT_MS,
   chooseMcp,
   claudeCodeAgent,
+  progressHead,
+  repoSuite,
   runEvalCommand,
   runEvalCompare
 };
-//# sourceMappingURL=evalCommand-SKJQ6CW2.js.map
+//# sourceMappingURL=evalCommand-MRO6ENCA.js.map

package/dist/{evalTune-CZRCPSZ3.js → evalTune-VL5RHYOG.js}

@@ -4,22 +4,25 @@ import {
   parseProposal,
   readAddendum,
   tune
-} from "./chunk-FOPBOB7W.js";
+} from "./chunk-YEH2OONM.js";
 import {
   EVAL_SUITES,
   HELD_OUT_TASKS,
+  requiredToolchains
+} from "./chunk-LWNMEGDF.js";
+import {
+  reportV2
+} from "./chunk-6JQTPW3P.js";
+import "./chunk-YAQ6QLKB.js";
+import "./chunk-E6YRCDSF.js";
+import {
+  DEFAULT_SEED,
   TOOLCHAINS,
   checkToolchains,
   formatReport,
-  requiredToolchains,
-  runEvals
-} from "./chunk-Q4P3W657.js";
-import {
-  DEFAULT_SEED,
-  reportV2,
+  runEvals,
   shuffled
-} from "./chunk-DOCDCQVM.js";
-import "./chunk-NXJ4XLB5.js";
+} from "./chunk-A5DLAXB4.js";
 import {
   NO_MODEL,
   chooseModel,
@@ -27,29 +30,32 @@ import {
   loadModelsConfig,
   resolveModel
 } from "./chunk-2FIV2VZP.js";
-import "./chunk-NHWHG3SW.js";
-import "./chunk-65NUEOI7.js";
+import "./chunk-NXJ4XLB5.js";
+import "./chunk-STVFWSIL.js";
 import "./chunk-G2AX7MGS.js";
+import "./chunk-ULAJPMY7.js";
+import "./chunk-65NUEOI7.js";
+import "./chunk-4QRTTRVK.js";
 import "./chunk-A2QAIRQL.js";
 import {
   newSessionId
-} from "./chunk-UKIE4AAM.js";
+} from "./chunk-NLS7GMF7.js";
 import {
   costOf
 } from "./chunk-7KBOIPUT.js";
 import "./chunk-B25VL4RU.js";
-import "./chunk-23FXBNBB.js";
-import "./chunk-C57GHLKP.js";
+import "./chunk-MNE7HHX2.js";
+import "./chunk-CFDRP6XW.js";
 import "./chunk-5DEWP3YX.js";
 import "./chunk-PZS3V6P3.js";
 import "./chunk-QANRZW7W.js";
 import "./chunk-ATW7BNIG.js";
-import "./chunk-55CFEEIN.js";
+import "./chunk-WK5NE6RY.js";
 import "./chunk-AOGFYUKG.js";
-import "./chunk-6XHON3Q6.js";
+import "./chunk-QXTHEFKP.js";
 import {
   createExecutor
-} from "./chunk-B4SJY7KT.js";
+} from "./chunk-LMKGSYJC.js";
 import "./chunk-BHZ6G65V.js";
 import "./chunk-P6XWWIEK.js";
 import "./chunk-UIRJOQQS.js";
@@ -292,4 +298,4 @@ export {
   progressLine,
   runEvalTune
 };
-//# sourceMappingURL=evalTune-CZRCPSZ3.js.map
+//# sourceMappingURL=evalTune-VL5RHYOG.js.map

package/dist/{export-SJ3DUP6W.js → export-ZPF3AMO6.js}

@@ -2,11 +2,20 @@
 import {
   summariseCall,
   summariseResult
-} from "./chunk-FARS4R67.js";
+} from "./chunk-AX3A5KPH.js";
+import {
+  receiptText
+} from "./chunk-2OYTNNUJ.js";
+import {
+  hasReceipt
+} from "./chunk-4QRTTRVK.js";
 import {
   serverCallText,
   serverResultSummary
 } from "./chunk-A2QAIRQL.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-ZPF3AMO6.js.map

package/dist/exportHtml-ISISG5HG.js

@@ -0,0 +1,21 @@
+#!/usr/bin/env node
+import {
+  OUTPUT_CHARS,
+  callChange,
+  escapeHtml,
+  sessionHtml
+} from "./chunk-4BELTWP4.js";
+import "./chunk-AX3A5KPH.js";
+import "./chunk-2OYTNNUJ.js";
+import "./chunk-4QRTTRVK.js";
+import "./chunk-A2QAIRQL.js";
+import "./chunk-NLS7GMF7.js";
+import "./chunk-7KBOIPUT.js";
+import "./chunk-BHZ6G65V.js";
+export {
+  OUTPUT_CHARS,
+  callChange,
+  escapeHtml,
+  sessionHtml
+};
+//# sourceMappingURL=exportHtml-ISISG5HG.js.map