@wrongstack/cli 1.0.28 → 1.0.32

package/dist/{bench-GTLPM22B.js → bench-MMZHFHDL.js}

@@ -5,7 +5,7 @@ import {
 // src/subcommands/handlers/bench.ts
 import * as fs from "node:fs/promises";
 import { createRequire } from "node:module";
-import * as path2 from "node:path";
+import * as path3 from "node:path";
 import {
   collectCellPredictions,
   compareReports,
@@ -20,7 +20,7 @@ import {
   gradeSwebench,
   loadBenchConfig as loadBenchConfig2,
   mineTranscript,
-  readRunDir,
+  readRunDir as readRunDir2,
   renderComparisonMarkdown,
   renderMarkdownReport,
   reportHeaderLine,
@@ -30,8 +30,93 @@ import {
 } from "@wrongstack/bench";
 import { color, toErrorMessage } from "@wrongstack/core/utils";
 
-// src/subcommands/handlers/bench-run-config.ts
+// src/subcommands/handlers/bench-experiments.ts
+import { createHash } from "node:crypto";
+import { open } from "node:fs/promises";
 import * as path from "node:path";
+import {
+  compareQualityExperiment,
+  readRunDir,
+  reviewTranscript,
+  routingExperiment
+} from "@wrongstack/bench";
+async function boundedFile(file, limit) {
+  const handle = await open(file, "r");
+  try {
+    const info = await handle.stat();
+    if (!info.isFile() || info.size > limit)
+      throw new Error("Experiment input exceeds the size limit");
+    const buffer = Buffer.alloc(limit + 1);
+    const { bytesRead } = await handle.read(buffer, 0, buffer.length, 0);
+    if (bytesRead > limit) throw new Error("Experiment input exceeds the size limit");
+    return buffer.subarray(0, bytesRead).toString("utf8");
+  } finally {
+    await handle.close();
+  }
+}
+async function benchExperimentCommand(action, args, deps) {
+  const flag = (name) => typeof deps.flags?.[name] === "string" ? deps.flags[name] : void 0;
+  try {
+    const policyFile = flag("policy");
+    const policy = policyFile ? JSON.parse(await boundedFile(path.resolve(deps.cwd, policyFile), 128 * 1024)) : void 0;
+    let report;
+    if (action === "review") {
+      const transcript = flag("transcript");
+      if (!transcript)
+        throw new Error("Use bench review --transcript <session.jsonl> [--policy <json>]");
+      const raw = await boundedFile(path.resolve(deps.cwd, transcript), 32 * 1024 * 1024);
+      const events = [];
+      for (const line of raw.split("\n").filter((line2) => line2.trim())) {
+        const event = JSON.parse(line);
+        if (!event || typeof event !== "object" || Array.isArray(event))
+          throw new Error("Transcript contains a non-event");
+        events.push(event);
+      }
+      if (policy !== void 0 && (!policy || typeof policy !== "object" || Array.isArray(policy)))
+        throw new Error("Invalid review policy");
+      report = reviewTranscript(
+        events,
+        createHash("sha256").update(raw).digest("hex"),
+        policy
+      );
+    } else if (action === "route") {
+      if (!args[0] || !policy || typeof policy !== "object" || Array.isArray(policy))
+        throw new Error("Use bench route <run-dir> --policy <routing.json>");
+      report = routingExperiment(
+        await readRunDir(path.resolve(deps.cwd, args[0])),
+        policy
+      );
+    } else {
+      const dimension = flag("dimension");
+      if (!args[0] || !args[1] || dimension !== "compaction" && dimension !== "behavior")
+        throw new Error(
+          "Use bench experiment <baseline-dir> <candidate-dir> --dimension compaction|behavior"
+        );
+      const [baseline, candidate] = await Promise.all([
+        readRunDir(path.resolve(deps.cwd, args[0])),
+        readRunDir(path.resolve(deps.cwd, args[1]))
+      ]);
+      report = compareQualityExperiment(baseline, candidate, dimension);
+    }
+    deps.renderer.write(`${JSON.stringify(report, null, 2)}
+`);
+    return 0;
+  } catch (error) {
+    deps.renderer.writeError(error instanceof Error ? error.message : "Experiment failed");
+    return 1;
+  }
+}
+async function readBenchReviewPolicy(file) {
+  const raw = JSON.parse(await boundedFile(file, 128 * 1024));
+  if (!raw || typeof raw !== "object" || Array.isArray(raw))
+    throw new Error("Invalid review policy");
+  const policy = raw;
+  reviewTranscript([], "0".repeat(64), policy);
+  return policy;
+}
+
+// src/subcommands/handlers/bench-run-config.ts
+import * as path2 from "node:path";
 import {
   configFromCells,
   loadBenchConfig,
@@ -56,7 +141,7 @@ async function resolveBenchRunConfig(input) {
   }
   const modelsFlag = stringFlag(input.flags, "models");
   const modelsPath = modelsFlag ?? DEFAULT_BENCH_MODELS_FILE;
-  const resolvedModels = path.resolve(input.cwd, modelsPath);
+  const resolvedModels = path2.resolve(input.cwd, modelsPath);
   const customModelsFile = modelsFlag !== void 0 && modelsFlag !== DEFAULT_BENCH_MODELS_FILE;
   try {
     return await loadBenchConfig(resolvedModels);
@@ -95,6 +180,10 @@ var benchCmd = async (args, deps) => {
   const sub = args[0];
   const rest = args.slice(1);
   switch (sub) {
+    case "review":
+    case "route":
+    case "experiment":
+      return benchExperimentCommand(sub, rest, deps);
     case "run":
       return benchRun(rest, deps);
     case "report":
@@ -120,6 +209,9 @@ function printUsage(deps) {
       "  mine     Copy a real session transcript and draft trace-eval cases",
       "  report   Re-render report.md from a finished run directory",
       "  list     Show available suites and configured model cells",
+      "  review   Review a transcript with advisory behavior/compaction evidence",
+      "  route    Recommend models in shadow mode from a finished run and category policy",
+      "  experiment Compare paired behavior/compaction runs on an identical corpus",
       "",
       color.dim("Examples:"),
       color.dim("  wstack bench run --cell anthropic/claude-sonnet-4-6,openai/gpt-5.4"),
@@ -143,11 +235,20 @@ function flagBool(deps, name) {
   const v = deps.flags?.[name];
   return v === true || v === "true";
 }
+function positiveIntegerFlag(deps, name) {
+  const raw = flagStr(deps, name);
+  if (raw === void 0) return void 0;
+  const value = Number(raw);
+  if (!/^\d+$/.test(raw.trim()) || !Number.isSafeInteger(value) || value < 1) {
+    throw new Error(`--${name} must be a positive safe integer.`);
+  }
+  return value;
+}
 async function resolveWstackEntry() {
   try {
     const req = createRequire(import.meta.url);
     const pkgPath = req.resolve("@wrongstack/cli/package.json");
-    const entry = path2.join(path2.dirname(pkgPath), "dist", "index.js");
+    const entry = path3.join(path3.dirname(pkgPath), "dist", "index.js");
     await fs.access(entry);
     return entry;
   } catch {
@@ -155,9 +256,28 @@ async function resolveWstackEntry() {
   }
 }
 async function benchRun(_args, deps) {
+  let reviewPolicy;
+  try {
+    if (flagStr(deps, "review-policy"))
+      reviewPolicy = await readBenchReviewPolicy(
+        path3.resolve(deps.cwd, flagStr(deps, "review-policy"))
+      );
+  } catch (error) {
+    deps.renderer.writeError(toErrorMessage(error));
+    return 1;
+  }
   const suiteId = flagStr(deps, "suite") ?? "core";
-  const limitRaw = flagStr(deps, "limit");
-  const limit = limitRaw ? Math.max(1, Number.parseInt(limitRaw, 10)) : void 0;
+  let limit;
+  let concurrency;
+  let repeats;
+  try {
+    limit = positiveIntegerFlag(deps, "limit");
+    concurrency = positiveIntegerFlag(deps, "concurrency");
+    repeats = positiveIntegerFlag(deps, "repeats");
+  } catch (error) {
+    deps.renderer.writeError(toErrorMessage(error));
+    return 1;
+  }
   const outBase = flagStr(deps, "out") ?? "bench-results";
   let config;
   try {
@@ -172,23 +292,11 @@ async function benchRun(_args, deps) {
     deps.renderer.writeError(toErrorMessage(err));
     return 1;
   }
-  const concurrencyRaw = flagStr(deps, "concurrency");
-  if (concurrencyRaw) {
-    const c = Number.parseInt(concurrencyRaw, 10);
-    if (c > 0) config.concurrency = c;
-  }
-  const repeatsRaw = flagStr(deps, "repeats");
-  if (repeatsRaw) {
-    const r = Number.parseInt(repeatsRaw, 10);
-    if (!Number.isFinite(r) || r <= 0) {
-      deps.renderer.writeError("--repeats must be a positive integer.");
-      return 1;
-    }
-    config.repeats = r;
-  }
+  if (concurrency !== void 0) config.concurrency = concurrency;
+  if (repeats !== void 0) config.repeats = repeats;
   const stamp = (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-");
-  const outDir = path2.resolve(deps.cwd, outBase, stamp);
-  const predictionsDir = path2.join(outDir, "predictions");
+  const outDir = path3.resolve(deps.cwd, outBase, stamp);
+  const predictionsDir = path3.join(outDir, "predictions");
   let suite;
   let grade;
   let isSwebench = false;
@@ -206,14 +314,14 @@ async function benchRun(_args, deps) {
     }
     const languagesRaw = flagStr(deps, "languages");
     const languages = languagesRaw ? languagesRaw.split(",").map((s) => s.trim()).filter(Boolean) : void 0;
-    suite = createPolyglotSuite({ polyglotDir: path2.resolve(deps.cwd, polyglotDir), languages });
+    suite = createPolyglotSuite({ polyglotDir: path3.resolve(deps.cwd, polyglotDir), languages });
     grade = (a) => gradePolyglot(a);
   } else if (suiteId === "swebench") {
     isSwebench = true;
     const datasetDir = flagStr(deps, "dataset-dir");
     const docker = flagBool(deps, "docker");
     suite = createSwebenchSuite({
-      datasetDir: datasetDir ? path2.resolve(deps.cwd, datasetDir) : void 0,
+      datasetDir: datasetDir ? path3.resolve(deps.cwd, datasetDir) : void 0,
       docker
     });
     grade = (a) => gradeSwebench({ ...a, predictionsDir });
@@ -226,8 +334,8 @@ async function benchRun(_args, deps) {
       );
       return 1;
     }
-    const manifestFile = manifestRaw ? path2.resolve(deps.cwd, manifestRaw) : void 0;
-    const suiteDir = suiteDirRaw ? path2.resolve(deps.cwd, suiteDirRaw) : path2.dirname(manifestFile);
+    const manifestFile = manifestRaw ? path3.resolve(deps.cwd, manifestRaw) : void 0;
+    const suiteDir = suiteDirRaw ? path3.resolve(deps.cwd, suiteDirRaw) : path3.dirname(manifestFile);
     suite = createLocalManifestSuite({ suiteDir, manifestFile });
     grade = (a) => gradeLocalManifest(a);
   } else {
@@ -242,7 +350,7 @@ async function benchRun(_args, deps) {
   const wstackEntry = await resolveWstackEntry();
   deps.renderer.writeInfo(`Running ${suiteId} across ${config.cells.length} model(s)\u2026`);
   await fs.mkdir(outDir, { recursive: true });
-  const partialPath = path2.join(outDir, "results.jsonl");
+  const partialPath = path3.join(outDir, "results.jsonl");
   await fs.writeFile(partialPath, "", "utf8");
   let report;
   try {
@@ -250,6 +358,7 @@ async function benchRun(_args, deps) {
       suite,
       grade,
       config,
+      reviewPolicy,
       cliVersion: CLI_VERSION,
       toolNames,
       toolManifestHash,
@@ -267,7 +376,7 @@ async function benchRun(_args, deps) {
   }
   await writeJsonArtifacts(outDir, report);
   const md = renderMarkdownReport(report);
-  await fs.writeFile(path2.join(outDir, "report.md"), md, "utf8");
+  await fs.writeFile(path3.join(outDir, "report.md"), md, "utf8");
   deps.renderer.write("\n" + md + "\n");
   if (isSwebench) {
     for (const cell of config.cells) {
@@ -280,7 +389,7 @@ async function benchRun(_args, deps) {
       "Grade with the official SWE-bench harness: python -m swebench.harness.run_evaluation --predictions_path <file> --run_id <id>"
     );
   }
-  deps.renderer.writeInfo(`Report written to ${path2.join(outDir, "report.md")}`);
+  deps.renderer.writeInfo(`Report written to ${path3.join(outDir, "report.md")}`);
   const spawned = report.results.filter((row) => row.run.status !== "crashed").length;
   if (report.results.length > 0 && spawned === 0) {
     deps.renderer.writeError(
@@ -296,16 +405,16 @@ async function benchReport(args, deps) {
     deps.renderer.writeError("Usage: wstack bench report <run-directory>");
     return 1;
   }
-  const outDir = path2.resolve(deps.cwd, dir);
+  const outDir = path3.resolve(deps.cwd, dir);
   let report;
   try {
-    report = await readRunDir(outDir);
+    report = await readRunDir2(outDir);
   } catch (err) {
     deps.renderer.writeError(`cannot read run artifacts in ${outDir}: ${toErrorMessage(err)}`);
     return 1;
   }
   const md = renderMarkdownReport(report);
-  await fs.writeFile(path2.join(outDir, "report.md"), md, "utf8");
+  await fs.writeFile(path3.join(outDir, "report.md"), md, "utf8");
   deps.renderer.write("\n" + md + "\n");
   return 0;
 }
@@ -317,12 +426,12 @@ async function benchCompare(args, deps) {
     deps.renderer.writeError("Usage: wstack bench compare <baseline-dir> <candidate-dir>");
     return 1;
   }
-  const baselineDir = path2.resolve(deps.cwd, baselineRaw);
-  const candidateDir = path2.resolve(deps.cwd, candidateRaw);
+  const baselineDir = path3.resolve(deps.cwd, baselineRaw);
+  const candidateDir = path3.resolve(deps.cwd, candidateRaw);
   let comparisonMd;
   try {
-    const baseline = await readRunDir(baselineDir);
-    const candidate = await readRunDir(candidateDir);
+    const baseline = await readRunDir2(baselineDir);
+    const candidate = await readRunDir2(candidateDir);
     const comparison = compareReports(baseline, candidate);
     comparisonMd = renderComparisonMarkdown(comparison);
   } catch (err) {
@@ -330,8 +439,8 @@ async function benchCompare(args, deps) {
     return 1;
   }
   const outRaw = flagStr(deps, "out");
-  const outFile = outRaw ? path2.resolve(deps.cwd, outRaw) : path2.join(candidateDir, "compare.md");
-  await fs.mkdir(path2.dirname(outFile), { recursive: true });
+  const outFile = outRaw ? path3.resolve(deps.cwd, outRaw) : path3.join(candidateDir, "compare.md");
+  await fs.mkdir(path3.dirname(outFile), { recursive: true });
   await fs.writeFile(outFile, comparisonMd, "utf8");
   deps.renderer.write("\n" + comparisonMd + "\n");
   deps.renderer.writeInfo(`Comparison written to ${outFile}`);
@@ -348,8 +457,8 @@ async function benchMine(args, deps) {
   const outRaw = flagStr(deps, "out") ?? "evals";
   try {
     const mined = await mineTranscript({
-      transcriptPath: path2.resolve(deps.cwd, transcriptRaw),
-      outDir: path2.resolve(deps.cwd, outRaw)
+      transcriptPath: path3.resolve(deps.cwd, transcriptRaw),
+      outDir: path3.resolve(deps.cwd, outRaw)
     });
     deps.renderer.writeInfo(
       `Mined ${mined.candidates.length} edit attempt(s) from ${mined.sessionId}.`
@@ -385,7 +494,7 @@ async function benchList(_args, deps) {
   const modelsPath = flagStr(deps, "models");
   if (modelsPath) {
     try {
-      const config = await loadBenchConfig2(path2.resolve(deps.cwd, modelsPath));
+      const config = await loadBenchConfig2(path3.resolve(deps.cwd, modelsPath));
       deps.renderer.write("\n" + color.bold("Model cells\n"));
       for (const cell of config.cells) {
         deps.renderer.write(
@@ -414,4 +523,4 @@ async function benchList(_args, deps) {
 export {
   benchCmd
 };
-//# sourceMappingURL=bench-GTLPM22B.js.map
+//# sourceMappingURL=bench-MMZHFHDL.js.map

package/dist/boot/brain-decision-log.d.ts

@@ -45,7 +45,7 @@ export interface BrainDecisionEntry {
  * cast at the call site because the EventBus generic constraint is
  * narrower than the dynamic string event names we register here.
  */
-export declare function subscribeBrainDecisionLog(events: any): {
+export declare function subscribeBrainDecisionLog(events: any, getMaxEntries?: () => number): {
     brainLog: BrainDecisionEntry[];
     pushBrainLog: (entry: BrainDecisionEntry) => void;
     /**

package/dist/boot/system-prompt-builder.d.ts

@@ -84,7 +84,7 @@ interface BindSystemPromptBuilderDeps {
     skillEagerMaxChars?: number | undefined;
     /** `config.features.tokenSavingMode` — forwarded so prompt guidance matches tool tiering. */
     tokenSavingMode?: TokenSavingTier | boolean | undefined;
-    /** `config.systemPrompt.variant` — selects system.md, system-lite.md, or system-pro.md. */
+    /** `config.systemPrompt.variant` — selects system.md, system-lite.md, system-pro.md, or system-scout.md. */
     systemPromptVariant?: SystemInstructionVariant | undefined;
     /** `--append-system-prompt[-file]` text; appended to the host prompt only. */
     appendedInstructions?: string | undefined;

package/dist/boot/system-prompt-menu.d.ts

@@ -1,8 +1,8 @@
 /**
- * Interactive system-prompt selection menu — Lite / Standard / Pro.
+ * Interactive system-prompt selection menu — Lite / Standard / Pro / Scout.
  *
  * Shown at startup when the user did not pin a variant via
- * `--system-lite`, `--system-pro`, or `--system-prompt <variant>`.
+ * `--system-lite`, `--system-pro`, `--system-scout`, or `--system-prompt <variant>`.
  * Each option displays the estimated token count of the identity block
  * that variant injects (resolved through the same bundled → global →
  * project instruction dirs the SystemPromptBuilder uses, so a project
@@ -56,6 +56,13 @@ export declare function runSystemPromptMenu(deps: {
     paths: SystemPromptMenuPaths;
     /** Saved variant from config — enables the summary gate. */
     lastVariant?: SystemInstructionVariant | undefined;
+    /**
+     * The launch folder is not a project (no manifest, no AGENTS.md, no git):
+     * the coding identities have nothing to work on, so Scout is preselected in
+     * the menu and the summary gate points at it. A saved choice is never
+     * replaced silently — the gate still defaults to it.
+     */
+    suggestScout?: boolean | undefined;
 }): Promise<SystemInstructionVariant>;
 /** Outcome of {@link maybeRunSystemPromptMenu}. */
 export interface SystemPromptMenuOutcome {
@@ -106,5 +113,7 @@ export declare function maybeRunSystemPromptMenu(opts: {
      * does not reliably honor for directory renames.
      */
     persist?: typeof persistSystemPromptVariant | undefined;
+    /** The launch folder is not a project — see `runSystemPromptMenu`. */
+    outsideProject?: boolean | undefined;
 }): Promise<SystemPromptMenuOutcome>;
 //# sourceMappingURL=system-prompt-menu.d.ts.map

package/dist/boot/tui-coordinator-setup.d.ts

@@ -1,4 +1,5 @@
 import type { Context } from '@wrongstack/core/agent';
+import type { BrainArbiter } from '@wrongstack/core/coordination';
 import { AutonomousCoordinator, type Director } from '@wrongstack/core/coordination';
 import type { EventBus } from '@wrongstack/core/kernel';
 import type { WstackPaths } from '@wrongstack/core/utils';
@@ -10,6 +11,7 @@ interface CoordinatorSetupContext {
     events: EventBus;
     /** The agent's Context — provides provider, model, session id, transcript path. */
     context: Context;
+    brain?: BrainArbiter | undefined;
     /** Resolved WrongStack paths — used for session-dir resolution. */
     wpaths: WstackPaths;
     /** The mailbox — cast to the core Mailbox interface for the coordinator. */

package/dist/boot/tui-startup-output.d.ts

@@ -0,0 +1,11 @@
+import type { Logger } from '@wrongstack/core/types';
+export declare function shouldCaptureTuiStartup(flags: Record<string, string | boolean>, positional: readonly string[], interactive?: boolean): boolean;
+/** Buffer diagnostics until the launch menu decides who owns the terminal.
+ * TUI logs go to its existing file logger; other surfaces replay normal output.
+ * Prompts, unstructured stderr and errors remain visible throughout boot.
+ */
+export declare function startTuiStartupOutput(enabled: boolean): {
+    acceptTui(fileLogger: Logger): void;
+    stop(replay?: boolean): void;
+};
+//# sourceMappingURL=tui-startup-output.d.ts.map

package/dist/{boot-auto-thin-6Q56FY3F.js → boot-auto-thin-K2DHHYIW.js}

@@ -1,10 +1,10 @@
 import {
   persistConfigSetting
-} from "./chunk-SFBFYIOW.js";
+} from "./chunk-4MDRHMX4.js";
 import {
   activeProfileConfigPath
 } from "./chunk-YMXXOOFN.js";
-import "./chunk-WSTVGC2C.js";
+import "./chunk-4BX2YDTQ.js";
 
 // src/boot/boot-auto-thin.ts
 import {
@@ -99,4 +99,4 @@ async function runBootAutoThin(deps) {
 export {
   runBootAutoThin
 };
-//# sourceMappingURL=boot-auto-thin-6Q56FY3F.js.map
+//# sourceMappingURL=boot-auto-thin-K2DHHYIW.js.map

package/dist/{chronicle-UTFETHP6.js → chronicle-PBJS7BN6.js}

@@ -82,9 +82,11 @@ var chronicleCmd = async (args, deps) => {
       const flags = deps.flags ?? {};
       const dryRun = args.includes("--dry-run") || args.includes("-n") || flags["dry-run"] === true || flags["dry-run"] === "true" || flags["n"] === true;
       const daysIndex = args.indexOf("--days");
+      const hasDaysFlag = daysIndex >= 0 || Object.hasOwn(flags, "days");
       const rawDays = daysIndex >= 0 && daysIndex + 1 < args.length ? args[daysIndex + 1] : typeof flags["days"] === "string" ? flags["days"] : void 0;
-      const days = rawDays ? parseInt(rawDays, 10) : 30;
-      if (!Number.isFinite(days) || days < 1) {
+      const trimmedDays = rawDays?.trim();
+      const days = !hasDaysFlag ? 30 : trimmedDays !== void 0 && /^\d+$/.test(trimmedDays) ? Number(trimmedDays) : Number.NaN;
+      if (!Number.isSafeInteger(days) || days < 1) {
         deps.renderer.writeError(
           `Invalid retention days: ${rawDays ?? ""}. Use --days N (positive integer).
 `
@@ -273,4 +275,4 @@ function tryStat(file) {
 export {
   chronicleCmd
 };
-//# sourceMappingURL=chronicle-UTFETHP6.js.map
+//# sourceMappingURL=chronicle-PBJS7BN6.js.map

package/dist/{chunk-EPR42SWM.js → chunk-2EZ2DHOW.js}

@@ -79,7 +79,7 @@ var BENCH_RUN_FLAGS = [
   {
     name: "limit",
     flag: "--limit <N>",
-    description: "Cap the number of tasks per cell (useful for smoke tests).",
+    description: "Cap the number of tasks per cell (positive safe integer; useful for smoke tests).",
     group: "control",
     kind: "value"
   },
@@ -94,14 +94,14 @@ var BENCH_RUN_FLAGS = [
   {
     name: "concurrency",
     flag: "--concurrency <N>",
-    description: "Override the per-cell concurrency from the model config.",
+    description: "Override the per-cell concurrency from the model config (positive safe integer).",
     group: "control",
     kind: "value"
   },
   {
     name: "repeats",
     flag: "--repeats <N>",
-    description: "Attempts per (task x model). >1 adds pass@k and a flakiness count so run-to-run noise is visible.",
+    description: "Attempts per (task x model), a positive safe integer. >1 adds pass@k and a flakiness count.",
     group: "control",
     kind: "value",
     defaultValue: "1"
@@ -706,6 +706,20 @@ var deepHelpTable = {
 
 // src/subcommands/handlers/per-subcommand-help-table.ts
 var helpTable = {
+  automation: {
+    name: "automation",
+    title: "wstack automation \u2014 persistent agent jobs",
+    description: "Manage persistent schedules, signed GitHub intake, isolated runs and run history.",
+    usage: "wstack automation <add|preview|templates|import|export|list|run|enable|disable|cancel|serve|prune> [id] [options]",
+    seeAlso: "docs/subcommands/automation.md"
+  },
+  sandbox: {
+    name: "sandbox",
+    title: "wstack sandbox \u2014 isolated Docker execution",
+    description: "Run the full agent in a copied project workspace and retain a patch without editing the host project.",
+    usage: "wstack sandbox docker --image <trusted-image> --prompt <task> [--dry-run] [--env NAME,...] [--provider id] [--model id] [--out directory]",
+    seeAlso: "docs/subcommands/sandbox.md"
+  },
   init: {
     name: "init",
     title: "wstack init \u2014 DEPRECATED (use wstack auth)",
@@ -795,10 +809,14 @@ var helpTable = {
     name: "auth",
     title: "wstack auth \u2014 manage API keys and provider credentials",
     description: "Add, view, and remove provider API keys. The interactive menu supports a custom-URL path for self-hosted / local servers, a quick-shortcut path for Ollama / vLLM / LM Studio, and a catalog path for the well-known providers.",
-    usage: "wstack auth [list|status|remove] [...] | wstack auth <provider> | wstack auth local [...]",
+    usage: "wstack auth [list|status|remove|cloud] [...] | wstack auth <provider> | wstack auth local [...]",
     subcommands: [
       { name: "list", description: "List saved providers and key status." },
       { name: "status <id>", description: "Show detail for one provider." },
+      {
+        name: "cloud <id>",
+        description: "Read or set native region/project/location/resource routing for a saved profile."
+      },
       { name: "remove <id>", description: "Remove a provider and its keys." },
       { name: "<provider>", description: "Add a key for a named provider (--label, --family, \u2026)." },
       {
@@ -995,8 +1013,12 @@ var helpTable = {
     name: "modeldiag",
     title: "wstack modeldiag \u2014 model benchmarks + heuristic diagnostics",
     description: "Read-only diagnostics for the configured model: key check, capability scan (vision / tools / context window), heuristic strengths/weaknesses (bestFor / avoidFor), sequential live provider/model smoke tests, and real benchmarks against a small prompt suite. Smoke tests use tiny prompts and never print credentials.",
-    usage: "wstack modeldiag [test|eval] [...]",
+    usage: "wstack modeldiag [preflight|test|eval] [...]",
     subcommands: [
+      {
+        name: "preflight [provider-profile]",
+        description: "Check native cloud provider setup without model requests or exposing credential values."
+      },
       {
         name: "<no subcommand>",
         description: "Print configured model capabilities, context limits, and recommendations."
@@ -1017,7 +1039,7 @@ var helpTable = {
     name: "bench",
     title: "wstack bench \u2014 run model-independent agentic benchmarks",
     description: "Run WrongStack against the bundled smoke suite, a local manifest, Aider polyglot, or SWE-bench Verified with deterministic graders. Default `bench run` is the bundled core edit-eval; pass --cell provider/model (or a saved model).",
-    usage: "wstack bench [run|compare|report|list] [...]",
+    usage: "wstack bench [run|compare|report|list|mine|review|route|experiment] [...]",
     subcommands: [
       { name: "run", description: "Run a suite. Default: bundled smoke + --cell or saved model." },
       {
@@ -1025,6 +1047,18 @@ var helpTable = {
         description: "Diff two finished run directories (fingerprint-aware)."
       },
       { name: "report <dir>", description: "Render the Markdown report for a prior run." },
+      {
+        name: "review --transcript <file>",
+        description: "Read-only transcript behavior/compaction evidence; optional --policy."
+      },
+      {
+        name: "route <dir> --policy <file>",
+        description: "Shadow routing recommendations from graded model-matrix evidence."
+      },
+      {
+        name: "experiment <a> <b> --dimension compaction|behavior",
+        description: "Compare paired task/model/attempt runs with explicit experiment provenance."
+      },
       { name: "list", description: "List available suites and the model configs in the catalog." }
     ],
     seeAlso: "wstack modeldiag (read-only diagnostics; bench actually runs the model)"
@@ -1142,6 +1176,25 @@ var helpTable = {
     ],
     seeAlso: "wstack mcp list; --mcp-config (one-off servers without importing)"
   },
+  sage: {
+    name: "sage",
+    title: "wstack sage \u2014 this project's SAGE memory in other coding agents",
+    description: "Wire SAGE memory into Claude Code, Codex, Cursor or Antigravity over MCP: writes the MCP server entry plus a skill/rule telling the agent when to recall and how to propose. The server attaches to the SAGE daemon while wstack is open in this project and never starts one. Tools are read-only plus memory_candidates (list/propose); a proposal becomes memory only through WrongStack review. Search is lexical: semantic recall runs inside a WrongStack host.",
+    usage: "wstack sage connect <claude-code|codex|cursor|antigravity|all|print> [--dry-run]",
+    subcommands: [
+      { name: "connect <client|all>", description: "Write the MCP entry and skill for a client." },
+      { name: "connect print", description: "Print JSON/TOML config for any other MCP client." },
+      { name: "disconnect <client|all>", description: "Remove what connect wrote." },
+      { name: "mcp [--origin <name>]", description: "The stdio MCP server clients run." },
+      {
+        name: "sync [--restart-service]",
+        description: "Run project memory sync with HQ; optionally restart this project\u2019s SAGE owner."
+      },
+      { name: "--dry-run", description: "Show the file changes without writing." },
+      { name: "--command <path>", description: "Executable to launch instead of wstack." }
+    ],
+    seeAlso: "wstack mcp serve (all WrongStack tools over MCP)"
+  },
   // -- Chronicle ──────────────────────────────────────────────────────
   chronicle: {
     name: "chronicle",
@@ -1243,4 +1296,4 @@ export {
   subcommandsWithFocusedHelp,
   renderGenericHelp
 };
-//# sourceMappingURL=chunk-EPR42SWM.js.map
+//# sourceMappingURL=chunk-2EZ2DHOW.js.map