skillgauntlet 1.3.1 → 1.3.2

package/bin/skillgauntlet.js

@@ -13,10 +13,11 @@ import { fingerprint, cacheFileFor, readCachedQueries, writeCachedQueries } from
 import { openRun, querySetHash, readRun, readAnswers, validRecords } from '../lib/runs.js';
 import { DEPTHS, minutesFor } from '../lib/depth.js';
 import { ask, choose, select, confirm, say, closeUi, hasTerminal, setBack, Back } from '../lib/ui.js';
-import { hostModels, unknownModel } from '../lib/models.js';
+import { hostModels, unknownModel, codebuddyCatalog } from '../lib/models.js';
 import { browseDir } from '../lib/browse.js';
 import { withTicker } from '../lib/ticker.js';
-import { renderResult, bar } from '../lib/chart.js';
+import { renderResult } from '../lib/chart.js';
+import { makeProgress } from '../lib/progress.js';
 import { setLang, detectLang, errorText, t } from '../lib/i18n.js';
 import { resolveBin, backend, backendFor, usesOwnDeepseek, setNativeModels, HOSTS, checkModel, draftQueries, draftQuality, runTrial, requireHost, pingHost, setHostBin, binInstalled } from '../lib/trial.js';
 import { HOST_CLI, searchHost, installCommand, installHost } from '../lib/findhost.js';
@@ -127,6 +128,13 @@ function notListed(model, known, label) {
   say(`\n${t('modelUnknown', { model, host: label, list: list.slice(0, 8).join(', ') })}`);
   return true;
 }
+// Models in CodeBuddy's cached catalog that its --help does not list for this account right now.
+function catalogOnly(host) {
+  if (host !== 'codebuddy') return [];
+  const listed = new Set(modelsOf('codebuddy', 'CodeBuddy'));
+  return codebuddyCatalog().filter((m) => !listed.has(m) && !m.startsWith('custom-local:'));
+}
+
 function modelsOf(host, label) {
   if (!modelCache.has(host)) {
     const list = withTicker(waitText(t('waitModels', { host: label }), { slowAfter: 600 }), () => hostModels(host, resolveBin(host)));
@@ -205,13 +213,15 @@ async function runTriggerLayer({ skillDir, skillName, depth, model, host = 'clau
     queries: { schema: 'skillgauntlet/queries/1', queries: qs.queries },
   });
 
+  const prog = progress(trials.length);
   try {
     const outcome = await executeTrials({
       run,
       trials,
       doTrial: ({ q }) => runTrial({ bin, skillDir, skillName, query: q.query, model, maxExplore: depth.maxExplore, host }),
-      onTrial: progress(),
+      onTrial: prog,
     });
+    prog.finish();
     const result = finishRun(run, { model, backend: backendFor(host, model), host, runs: depth.runs, expected: trials.length }, outcome.stopped);
     say(`  ${t('runSaved', { path: run.dir })}`);
     if (outcome.stopped) reportStopped(outcome.stopped, run.dir, model);
@@ -219,6 +229,7 @@ async function runTriggerLayer({ skillDir, skillName, depth, model, host = 'clau
     else finishedRuns.push(run.dir);
     return result && { ...result, runDir: run.dir, queries: qs.queries };
   } catch (err) {
+    prog.finish();
     run.close(null, { interrupted: true });
     throw err;
   }
@@ -250,17 +261,7 @@ function archiveProbe({ skillDir, skillName, model, probe }) {
   say(`  ${t('runSaved', { path: run.dir })}`);
 }
 
-function progress() {
-  return ({ row, done, total }) => {
-    // On a terminal one line is redrawn in place; the verdicts of single trials belong to the result, not the wait.
-    if (process.stdout.isTTY) {
-      process.stdout.write(`\r\x1b[2K  ${t('progressLine')} ${bar(done / total, 24)} ${done}/${total}${done >= total ? '\n' : ''}`);
-      return;
-    }
-    const mark = { triggered: '✓', not_triggered: '·', other_skill: '↷' }[row.verdict] || '!';
-    say(`  [${done}/${total}] ${mark} ${row.queryId}#${row.run} → ${row.verdict}`);
-  };
-}
+const progress = makeProgress;
 
 function reportIncomplete(n, dir, model) {
   say(`
@@ -286,12 +287,14 @@ async function cmdResume(args) {
   const runMeta = JSON.parse(fs.readFileSync(path.join(full, 'meta.json'), 'utf8'));
   const runHost = runMeta.host || 'claude-code';
   requireHost(runHost);
+  const resumeProgress = progress();
   const result = await resumeRun({
     dir: full,
     bin: resolveBin((HOSTS[runHost] || HOSTS['claude-code']).bin),
     want: { model: flag('--model'), runs: flag('--runs'), maxExplore: flag('--max-explore'), backend: backendFor(runHost, flag('--model') ?? runMeta.model) },
-    onTrial: progress(),
+    onTrial: resumeProgress,
   });
+  resumeProgress.finish();
   if (result.nothing) {
     say(t('resumeNothing'));
     return;
@@ -320,12 +323,14 @@ async function pickInstalled() {
 }
 
 // A model for hosts that have an account default: pick it, a known name, or type one. Returns null for the default.
-async function pickModel(known = []) {
+async function pickModel(known = [], more = []) {
   if (!hasTerminal()) return (await ask(t('modelOptionalQ'), '')).trim() || null;
   const other = '\u0000other';
+  // `known` is what the account can use now; `more` is in the host's catalog but not promised to the account
   const v = await select(t('modelQ'), [
     { label: t('modelDefault'), value: '' },
-    ...known.map((m) => ({ label: m, value: m })),
+    ...known.map((m) => ({ label: m, hint: more.length ? t('modelAccountHint') : '', value: m })),
+    ...more.map((m) => ({ label: m, hint: t('modelCatalogHint'), value: m })),
     { label: t('modelOther'), value: other },
   ], 0);
   if (v === other) return (await ask(t('modelOptionalQ'), '')).trim() || null;
@@ -540,8 +545,8 @@ async function interactive() {
         { label: t('depthStandard'), hint: t('depthStandardHint'), value: 'standard' },
       ], 1);
       S.depth = DEPTHS[depthId];
-      S.model = await pickModel(modelsOf('codebuddy', 'CodeBuddy'));
-      if (notListed(S.model, modelsOf('codebuddy', 'CodeBuddy'), 'CodeBuddy')) return AGAIN;
+      S.model = await pickModel(modelsOf('codebuddy', 'CodeBuddy'), catalogOnly('codebuddy'));
+      if (notListed(S.model, [...modelsOf('codebuddy', 'CodeBuddy'), ...catalogOnly('codebuddy')], 'CodeBuddy')) return AGAIN;
       if (usesOwnDeepseek('codebuddy', S.model) && !process.env.DEEPSEEK_API_KEY) {
         say(`\n${t('needDeepseekKey')}`);
         return END;

package/lib/i18n.js

@@ -131,6 +131,8 @@ const MESSAGES = {
     'chart.otherSkill': '改用别的技能：{name}',
     'chart.more': '…还有 {n} 题，见完整报告',
     'chart.wrongTitle': '不该触发却触发的 {n} 次',
+    modelAccountHint: '你的账号现在可用',
+    modelCatalogHint: 'CodeBuddy 目录里有，账号不一定能用',
     modelUnknown: '{host} 的模型清单里没有“{model}”。它现在支持：{list}。请重新选一个，或选“默认模型”。',
     updateAvail: '有新版本 {latest}（你现在用的是 {current}）。',
     updateQ: '要更新吗？',
@@ -527,6 +529,8 @@ const MESSAGES = {
     'chart.otherSkill': 'used another skill: {name}',
     'chart.more': '…{n} more, see the full report',
     'chart.wrongTitle': '{n} times it triggered when it should not have',
+    modelAccountHint: 'usable with your account now',
+    modelCatalogHint: 'in CodeBuddy\'s catalog; your account may not have it',
     modelUnknown: '{host} does not list a model called "{model}". It currently supports: {list}. Pick another, or choose the default model.',
     updateAvail: 'A newer version {latest} is available (you are on {current}).',
     updateQ: 'Update?',
@@ -971,6 +975,8 @@ Layer 0 always runs first: it is free, and it disqualifies most problems before
     'chart.otherSkill': '別のスキルを使用：{name}',
     'chart.more': '…ほか {n} 件は詳細レポートへ',
     'chart.wrongTitle': '発動すべきでないのに発動した {n} 回',
+    modelAccountHint: 'このアカウントで今使えます',
+    modelCatalogHint: 'CodeBuddy のカタログにあります（アカウントで使えるとは限りません）',
     modelUnknown: '{host} のモデル一覧に「{model}」はありません。現在対応：{list}。別のものを選ぶか、既定のモデルを選んでください。',
     updateAvail: '新しいバージョン {latest} があります（現在は {current}）。',
     updateQ: '更新しますか？',

package/lib/models.js

@@ -1,5 +1,8 @@
 // Model names a host will accept, read from the host itself so the list is the user's own (it depends on the
 // account and the custom models they set up) and never goes stale inside this package. Any failure gives [].
+import fs from 'node:fs';
+import os from 'node:os';
+import path from 'node:path';
 import { execFileSync, execSync } from 'node:child_process';
 import { launchSpec } from './spawn.js';
 
@@ -44,3 +47,35 @@ export function unknownModel(model, known) {
   if (!model || !known?.length || known.includes(model) || /^deepseek/i.test(model)) return null;
   return known;
 }
+
+// `codebuddy --help` lists only what the account is signed in for right now (it can shrink to two entries), while the
+// model catalog CodeBuddy last fetched stays in its local cache. Every string array called "models" inside those cache
+// files is part of that catalog; the union with the help list is what the user remembers having been offered. A name
+// from the catalog only is not promised to work for the account, so the list marks it and a failure shows the CLI's words.
+export function catalogFromTexts(texts) {
+  const names = new Set();
+  const walk = (node, depth) => {
+    if (!node || depth > 8) return;
+    if (Array.isArray(node)) { node.forEach((n) => walk(n, depth + 1)); return; }
+    if (typeof node !== 'object') return;
+    for (const [k, v] of Object.entries(node)) {
+      if (k === 'models' && Array.isArray(v)) v.filter((x) => typeof x === 'string' && /^[\w.:-]{2,80}$/.test(x)).forEach((x) => names.add(x));
+      else walk(v, depth + 1);
+    }
+  };
+  for (const text of texts) { try { walk(JSON.parse(text), 0); } catch { /* not JSON */ } }
+  return [...names];
+}
+
+export function codebuddyCatalog(dir = path.join(os.homedir(), '.codebuddy', 'local_storage')) {
+  const texts = [];
+  try {
+    for (const f of fs.readdirSync(dir)) {
+      if (!f.endsWith('.info')) continue;
+      const file = path.join(dir, f);
+      if (fs.statSync(file).size > 2 * 1024 * 1024) continue;
+      texts.push(fs.readFileSync(file, 'utf8'));
+    }
+  } catch { /* no cache: only the help list */ }
+  return catalogFromTexts(texts);
+}

package/lib/progress.js

@@ -0,0 +1,48 @@
+// The wait for the trials. On a terminal the bar is drawn at once (0 of N) and redrawn every second with the seconds
+// elapsed, so there is something on screen from the moment the trials start rather than after the first one finishes;
+// a trial can take many seconds. `total` is known when a run starts; a resumed run learns it from its first trial.
+// Without a terminal each finished trial is one line. finish() stops the redraw and ends the line.
+import { t } from './i18n.js';
+import { bar } from './chart.js';
+
+export function makeProgress(total = 0, {
+  tty = Boolean(process.stdout.isTTY),
+  write = (s) => process.stdout.write(s),
+  line = (s) => process.stdout.write(`${s}\n`),
+  now = Date.now,
+  every = setInterval,
+  stopEvery = clearInterval,
+} = {}) {
+  const started = now();
+  let finished = 0;
+  let of = total;
+  let timer = null;
+  let open = false;
+  const draw = () => {
+    open = true;
+    write(`\r\x1b[2K  ${t('progressLine')} ${bar(of ? finished / of : 0, 24)} ${finished}/${of || '?'}  ${t('waitElapsed', { s: Math.floor((now() - started) / 1000) })}`);
+  };
+  const stop = () => {
+    if (timer) stopEvery(timer);
+    timer = null;
+    if (tty && open) { write('\n'); open = false; }
+  };
+  if (tty && total) {
+    draw();
+    timer = every(draw, 1000);
+  }
+  const onTrial = ({ row, done, total: n }) => {
+    finished = done;
+    of = n;
+    if (tty) {
+      if (!timer) timer = every(draw, 1000);
+      draw();
+      if (done >= n) stop();
+      return;
+    }
+    const mark = { triggered: '✓', not_triggered: '·', other_skill: '↷' }[row.verdict] || '!';
+    line(`  [${done}/${n}] ${mark} ${row.queryId}#${row.run} → ${row.verdict}`);
+  };
+  onTrial.finish = stop;
+  return onTrial;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "skillgauntlet",
-  "version": "1.3.1",
+  "version": "1.3.2",
   "description": "Check an agent skill before you install it: context cost, permission surface, code risk, instruction hijacking, description quality, trigger rate and gain.",
   "type": "module",
   "bin": {