@curdx/flow 7.1.19 → 7.1.21

package/CHANGELOG.md

@@ -2,6 +2,45 @@
 
 All notable changes to `@curdx/flow` are documented here. Format follows [Keep a Changelog](https://keepachangelog.com/) and the project follows [Semantic Versioning](https://semver.org/).
 
+## 7.1.21 — 2026-05-11
+
+### Added
+
+- **Plugin-local runtime CLI and workflow snapshot surface.** `plugins/curdx-flow/bin/curdx-flow` now exposes deterministic `route`, `snapshot`, `specs`, `state`, `tasks`, `verify-blocks`, and `doctor` commands that skills and smoke tests can call without reaching into bundled hook internals.
+- **True Claude Code end-to-end flow test.** New `npm run test:claudecc:e2e` creates a temporary failing Node project, runs `/curdx-flow:start` through the real Claude Code CLI with the local plugin, verifies `npm test`, validates spec/task/snapshot invariants, captures `--debug-file`, and writes an `analyze` report.
+- **Prompt/batch runtime context hooks.** Added generated hook entrypoints for user-prompt expansion and post-tool-batch snapshots so Claude sees active-spec gates and next actions while a real session is running.
+
+### Changed
+
+- **Quick lite-spec is now a first-class completed workflow.** Runtime snapshots accept completed `spec-lite` quick specs with `tasks.md`, `.progress.md`, and `.curdx-state.json` without requiring full-spec `research.md` / `requirements.md` / `design.md` artifacts.
+- **Active spec marker contract is explicit.** `/curdx-flow:start` now instructs Claude to write `$defaultDir/.current-spec` with a bare name for default-root specs, and never to write a project-root `.current-spec`.
+- **Hook freshness checking is dirty-worktree friendly.** `check-hooks-fresh` now compares generated hook tree hashes before/after rebuild, preserving CI stale-bundle detection without failing merely because fresh bundles are already uncommitted.
+
+### Fixed
+
+- **`.current-spec` path resolution.** Runtime resolvers now treat relative marker values containing `/` as paths rather than double-prefixing them under `specs/`, and defensively accept a project-root marker if an older run wrote one.
+- **Small `npm test` implementation goals no longer route as publish-critical work.** Auto-policy no longer treats the word `npm` itself as high/critical release risk, while release/publish/tag/plugin/hook surfaces remain protected.
+- **Claude Code transcript lookup follows current project-dir encoding.** `analyze` now resolves `~/.claude/projects` directories where Claude Code hyphenates non-alphanumeric path characters, with legacy slash-only encoding retained as fallback.
+- **`analyze --out` writes the report file.** Markdown/JSON report output now respects `--out` in both fresh and cached-report paths.
+- **`analyze` scopes sidecar hook errors to the active project.** Global `~/.claude/curdx-flow/errors.jsonl` rows are now included only when their `cwd` or `transcript_path` matches the analyzed source, preventing stale errors from other sessions from polluting E2E reports.
+
+### Tests
+
+- Added regression coverage for quick lite-spec snapshots, active-spec marker placement, transcript encoding, scoped sidecar hook errors, `analyze --out`, route classification for `npm test`, and the new real Claude Code E2E script.
+- Verified with `npm run verify`, `npm run test:claudecc`, and `CURDX_FLOW_E2E_KEEP_TMP=1 npm run test:claudecc:e2e`.
+
+## 7.1.20 — 2026-05-11
+
+### Changed
+
+- **Companion capabilities are now the default required bundle.** `pua`, `claude-mem`, `chrome-devtools-mcp`, `frontend-design`, `sequential-thinking`, and `context7` are marked `required` in the installer alongside `curdx-flow`, so interactive installs show them as always installed and `--ids` installs auto-include missing required companions.
+- **Official plugin dependency metadata added.** `curdx-flow` now declares plugin dependencies for `pua`, `claude-mem`, `chrome-devtools-mcp`, and `frontend-design`, and the root marketplace explicitly allows those cross-marketplace dependencies.
+- **Capability routing marks bundled companions as `core-required`.** Smart-route recommendations still trigger only when relevant, but recommended bundled capabilities are no longer suppressed by narrow availability filters.
+
+### Tests
+
+- Added registry and manifest regression coverage for the required bundle, expanded capability-router expectations, rebuilt hook bundles, and tightened `claudecc` smoke checks for `core-required` recommendations.
+
 ## 7.1.19 — 2026-05-11
 
 ### Added

package/README.md

@@ -123,7 +123,7 @@ The same npm package delivers both products. The installer reads its descriptor
 
 - **Node.js** ≥ 20.12 ([download](https://nodejs.org))
 - **Claude Code CLI** installed and on `PATH` ([install instructions](https://docs.anthropic.com/en/docs/claude-code))
-- *(Optional)* **Bun** ≥ 1.0 — auto-detected; the installer offers to install it if you choose `claude-mem`
+- *(Optional)* **Bun** ≥ 1.0 — auto-detected; the installer offers to install it for the default `claude-mem` companion
 
 ### Install
 
@@ -131,7 +131,7 @@ The same npm package delivers both products. The installer reads its descriptor
 npx @curdx/flow
 ```
 
-On first run, you pick a language (中文 / English), then select what to install. The bundled `curdx-flow` plugin (the spec workflow itself) is always installed. Everything else is opt-in.
+On first run, you pick a language (中文 / English). The full curdx-flow bundle is installed by default: `curdx-flow`, `pua`, `claude-mem`, `chrome-devtools-mcp`, `frontend-design`, `sequential-thinking`, and `context7`. If a companion has a local prerequisite that is missing, the installer reports the skip reason and continues so the core workflow remains usable.
 
 That language choice also controls the managed `~/.claude/CLAUDE.md` block that flow writes. The block is rendered in the selected language, and when you choose `zh` it additionally injects a language policy telling Claude to keep tool/model interaction in English while replying to the user in Simplified Chinese.
 
@@ -428,9 +428,9 @@ Most workflow frameworks address this by adding more agents, more orchestration,
 
 No. flow runs **inside** Claude Code as a plugin. You still use the Claude Code CLI / IDE extension to chat; flow adds the `/curdx-flow:*` commands, subagents, hooks, and the bundled marketplace.
 
-#### Can I use flow without installing the marketplace plugins?
+#### Can I use flow without the companion capabilities?
 
-Yes. The bundled `curdx-flow` plugin is the spec workflow itself; everything else (claude-mem, chrome-devtools-mcp, …) is opt-in. `npx @curdx/flow install` lets you pick.
+The default installer treats the companion capabilities as required because routing depends on them for memory, current docs, frontend design, browser verification, structured reasoning, and retry/parallel recovery. If a local prerequisite is missing, that item may be skipped with a visible reason, but the next `npx @curdx/flow install` run will try to restore it.
 
 #### Do I need to commit the spec files?
 

package/dist/{analyze-FX2PCSL6.mjs → analyze-I4TXK4S7.mjs}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 
 // src/analyze/index.ts
-import { existsSync, mkdirSync, readFileSync as readFileSync2, readdirSync as readdirSync2, statSync as statSync3, writeFileSync } from "fs";
+import { existsSync, mkdirSync, readFileSync as readFileSync2, readdirSync as readdirSync2, realpathSync as realpathSync2, statSync as statSync3, writeFileSync } from "fs";
 import { homedir as homedir2 } from "os";
 import path4 from "path";
 
@@ -1617,8 +1617,17 @@ function resolveRealCwd(cwd) {
   return real;
 }
 function encodeCwd(realCwd) {
+  return realCwd.replace(/[^A-Za-z0-9]/g, "-");
+}
+function encodeLegacyCwd(realCwd) {
   return realCwd.replace(/\//g, "-");
 }
+function candidateProjectDirs(home, realCwd) {
+  const encoded = encodeCwd(realCwd);
+  const legacy = encodeLegacyCwd(realCwd);
+  const names = encoded === legacy ? [encoded] : [encoded, legacy];
+  return names.map((name) => path3.join(home, ".claude", "projects", name));
+}
 function resolveTranscriptSource(opts = {}) {
   const cwd = opts.cwd ?? process.cwd();
   if (opts.fixtureOverride) {
@@ -1642,12 +1651,20 @@ function resolveTranscriptSource(opts = {}) {
   }
   const home = opts.homedir ?? homedir();
   const realCwd = resolveRealCwd(cwd);
-  const encoded = encodeCwd(realCwd);
-  const encodedDir = path3.join(home, ".claude", "projects", encoded);
+  const candidates = candidateProjectDirs(home, realCwd);
+  let encodedDir = candidates[0];
   let entries = [];
-  try {
-    entries = readdirSync(encodedDir, { withFileTypes: true });
-  } catch {
+  let foundDir = false;
+  for (const candidate of candidates) {
+    try {
+      entries = readdirSync(candidate, { withFileTypes: true });
+      encodedDir = candidate;
+      foundDir = true;
+      break;
+    } catch {
+    }
+  }
+  if (!foundDir) {
     throw new TranscriptNotFoundError(
       encodedDir,
       `no Claude Code project dir for cwd ${cwd} \u2014 run \`claude\` here at least once, or pass CURDX_TRANSCRIPT_FIXTURE=\u2026 for tests`
@@ -1684,6 +1701,15 @@ function writeState(state) {
   if (!existsSync(STATE_DIR)) mkdirSync(STATE_DIR, { recursive: true });
   writeFileSync(STATE_PATH, JSON.stringify(state, null, 2), "utf8");
 }
+function emitReport(bytes, out) {
+  if (!out) {
+    process.stdout.write(bytes);
+    return;
+  }
+  const target = path4.resolve(out);
+  mkdirSync(path4.dirname(target), { recursive: true });
+  writeFileSync(target, bytes, "utf8");
+}
 function cleanupOrphanState(state, currentPaths) {
   const now = Date.now();
   const THIRTY_DAYS_MS = 30 * 24 * 60 * 60 * 1e3;
@@ -1740,7 +1766,26 @@ function loadSpecStates() {
   }
   return out;
 }
-function loadErrorEntries() {
+function normalizePathForScope(p) {
+  try {
+    return realpathSync2(p);
+  } catch {
+    return path4.resolve(p);
+  }
+}
+function isErrorEntryInScope(entry, source) {
+  const transcriptPaths = new Set(source.paths.map((p) => normalizePathForScope(p)));
+  if (entry.transcript_path) {
+    if (transcriptPaths.has(normalizePathForScope(entry.transcript_path))) return true;
+  }
+  if (entry.cwd) {
+    const sourceCwds = /* @__PURE__ */ new Set([normalizePathForScope(source.cwd)]);
+    if (source.kind === "real") sourceCwds.add(normalizePathForScope(source.realCwd));
+    if (sourceCwds.has(normalizePathForScope(entry.cwd))) return true;
+  }
+  return false;
+}
+function loadErrorEntries(source) {
   if (!existsSync(ERRORS_LOG_PATH)) return [];
   let raw;
   try {
@@ -1757,7 +1802,7 @@ function loadErrorEntries() {
       const kind = typeof parsed.kind === "string" ? parsed.kind : "unknown";
       const payload = parsed.payload && typeof parsed.payload === "object" && !Array.isArray(parsed.payload) ? parsed.payload : void 0;
       const correlationId = typeof parsed.correlationId === "string" ? parsed.correlationId : void 0;
-      out.push({
+      const entry = {
         ts: typeof parsed.ts === "string" ? parsed.ts : "",
         ...typeof parsed.hook === "string" ? { hook: parsed.hook } : {},
         ...typeof parsed.event === "string" ? { event: parsed.event } : {},
@@ -1768,7 +1813,8 @@ function loadErrorEntries() {
         kind,
         ...payload !== void 0 ? { payload } : {},
         ...correlationId !== void 0 ? { correlationId } : {}
-      });
+      };
+      if (isErrorEntryInScope(entry, source)) out.push(entry);
     } catch {
       continue;
     }
@@ -1797,12 +1843,12 @@ async function runAnalyzeInner(opts) {
   const cacheCompatible = (state.lastIncludePrompts ?? false) === includePrompts && (state.lastCostSummary ?? false) === costSummary;
   if (allCachedReady && pathStats.length > 0 && cacheCompatible && (state.lastReportJson || state.lastReportMarkdown)) {
     if (opts.json && state.lastReportJson) {
-      process.stdout.write(state.lastReportJson);
+      emitReport(state.lastReportJson, opts.out);
       writeState(state);
       return;
     }
     if (!opts.json && state.lastReportMarkdown) {
-      process.stdout.write(state.lastReportMarkdown);
+      emitReport(state.lastReportMarkdown, opts.out);
       writeState(state);
       return;
     }
@@ -1822,7 +1868,7 @@ async function runAnalyzeInner(opts) {
       if (r) redacted.push(r);
     }
     const filtered = filterEvents(redacted, { ...opts, limit });
-    const errorEntries = loadErrorEntries();
+    const errorEntries = loadErrorEntries(source);
     const specStates = loadSpecStates();
     let costBreakdown;
     let recommendations = [];
@@ -1890,7 +1936,6 @@ async function runAnalyzeInner(opts) {
       ...recommendations.length > 0 ? { recommendations } : {}
     });
     const safeJson = redactReportFields(json, { includePrompts });
-    void opts.out;
     const markdownStr = markdown;
     let jsonObj = safeJson;
     if (opts.costSummary === true && costBreakdown) {
@@ -1908,9 +1953,9 @@ async function runAnalyzeInner(opts) {
     state.lastIncludePrompts = includePrompts;
     state.lastCostSummary = costSummary;
     if (opts.json) {
-      process.stdout.write(jsonStr);
+      emitReport(jsonStr, opts.out);
     } else {
-      process.stdout.write(markdownStr);
+      emitReport(markdownStr, opts.out);
     }
     void safeJson;
     if (counters.parse_error || counters.unknown_type) {

package/dist/index.mjs

@@ -514,6 +514,7 @@ var pua = {
   name: "pua",
   description: "tanweai/pua \u2014 Chinese Claude Code skills bundle",
   type: "plugin",
+  required: true,
   slashNamespace: "/pua:*",
   whenToUse: "auto-fires on 2+ failures or user frustration; sub-modes p7 / p9 / pro / loop. Skip on first-attempt failures or when a known fix is executing.",
   marketplaces: () => [MARKETPLACE_NAME],
@@ -609,6 +610,7 @@ var claudeMem = {
   name: "claude-mem",
   description: "thedotmack/claude-mem \u2014 persistent cross-session memory for Claude Code",
   type: "plugin",
+  required: true,
   slashNamespace: "/claude-mem:*",
   whenToUse: 'for cross-session memory search ("did we solve this before?"), phased planning (`make-plan`), or phased execution (`do`).',
   marketplaces: () => [MARKETPLACE_NAME2],
@@ -668,6 +670,7 @@ var chromeDevtoolsMcp = {
   name: "chrome-devtools-mcp",
   description: "ChromeDevTools/chrome-devtools-mcp \u2014 drive a real Chrome from Claude Code",
   type: "plugin",
+  required: true,
   whenToUse: "when debugging code that runs in a browser: perf traces, network / console inspection, DOM / CSS issues. Prefer snapshot over screenshot.",
   prereqCheck: async (t2) => {
     const major = Number(process.versions.node.split(".")[0] ?? "0");
@@ -697,6 +700,7 @@ var frontendDesign = {
   name: "frontend-design",
   description: "Anthropic official \u2014 UI/frontend design helpers",
   type: "plugin",
+  required: true,
   whenToUse: "auto-fires when building UI / web components / pages. Best where visual personality matters (landing, marketing, portfolio).",
   isInstalled: () => isPluginInstalled(PLUGIN_ID4),
   install: (ctx) => installPluginById(PLUGIN_ID4, ctx),
@@ -742,6 +746,7 @@ var sequentialThinking = {
   name: "sequential-thinking",
   description: "modelcontextprotocol/server-sequential-thinking \u2014 structured reasoning helper",
   type: "mcp",
+  required: true,
   whenToUse: "for complex multi-step problems where assumptions may shift (architecture comparison, risk-assessed migrations, prod-only debugging). Skip for simple queries.",
   isInstalled: () => isMcpInstalled(MCP_NAME),
   install: async (ctx) => {
@@ -783,6 +788,7 @@ var context7 = {
   name: "context7",
   description: "upstash/context7 \u2014 up-to-date docs from any library (HTTP MCP, optional API key)",
   type: "mcp",
+  required: true,
   whenToUse: "for any library / SDK / framework / API / Claude Code docs lookup. Use instead of web search.",
   isInstalled: () => isMcpInstalled(MCP_NAME2),
   configPrompts: async ({ t: t2 }) => {
@@ -913,7 +919,15 @@ var ORDER = [
   "sequential-thinking",
   "pua"
 ];
-var DOCS_RE = /\b(api|sdk|library|libraries|framework|docs?|documentation|version|upgrade|dependency|dependencies|claude code|plugin|mcp|hook|hooks|skill|skills|agent|agents|react|vue|spring|spring boot|spring cloud|next\.?js|vite|webpack|npm|node)\b|最新|文档|依赖|框架|插件|官方|联网|搜索/i;
+var CORE_REQUIRED = /* @__PURE__ */ new Set([
+  "context7",
+  "claude-mem",
+  "frontend-design",
+  "chrome-devtools-mcp",
+  "sequential-thinking",
+  "pua"
+]);
+var EXTERNAL_DOCS_RE = /\b(api|sdk|library|libraries|framework|version|upgrade|dependency|dependencies|official docs?|latest docs?|claude code|plugin|mcp|hook|hooks|skill|skills|agent|agents|react|vue|spring|spring boot|spring cloud|next\.?js|vite|webpack|npm|node)\b|最新|依赖|框架|插件|官方|联网|搜索|文档.*(最新|官方|API|SDK|框架|插件|依赖)/i;
 var MEMORY_RE = /\b(previous|before|again|remember|memory|history|similar|repeated|regression|already solved|same bug|past decision)\b|之前|上次|记得|历史|做过|又|重复|老问题/i;
 var UI_RE = /\b(ui|ux|frontend|front-end|browser|chrome|dom|css|html|layout|component|page|form|modal|responsive|visual|render|react|vue|vite|next\.?js|screenshot|interaction)\b|前端|页面|浏览器|样式|交互|组件|布局|视觉|截图/i;
 var BROWSER_VERIFY_RE = /\b(browser|chrome|dom|css|network|console|performance|render|screenshot|e2e|playwright|visual regression|interaction)\b|浏览器|控制台|网络|性能|渲染|截图|端到端/i;
@@ -928,11 +942,14 @@ function hasAny(values, candidates) {
   const set = new Set((values ?? []).map((v) => v.toLowerCase()));
   return candidates.some((candidate) => set.has(candidate.toLowerCase()));
 }
-function capabilityAllowed(id, available) {
-  return available === null || available.has(id);
+function capabilityAvailability(id, available) {
+  if (CORE_REQUIRED.has(id)) return "core-required";
+  if (available === null) return "check-if-installed";
+  return available.has(id) ? "known-available" : null;
 }
 function pushRecommendation(out, available, id, phase, reason, instruction) {
-  if (!capabilityAllowed(id, available)) return;
+  const availability = capabilityAvailability(id, available);
+  if (availability === null) return;
   if (out.some((rec) => rec.id === id)) return;
   const cap = CAPABILITIES[id];
   out.push({
@@ -941,6 +958,7 @@ function pushRecommendation(out, available, id, phase, reason, instruction) {
     type: cap.type,
     invocation: cap.invocation,
     phase,
+    availability,
     reason,
     instruction
   });
@@ -960,16 +978,17 @@ function recommendToolCapabilities(input) {
   if (missingRoots > 0) {
     return recs;
   }
-  const localLowRisk = LOW_RISK_LOCAL_RE.test(goal) && route === "direct-change";
+  const externalDocsRelevant = EXTERNAL_DOCS_RE.test(goal);
+  const localLowRisk = LOW_RISK_LOCAL_RE.test(goal) && route === "direct-change" && !externalDocsRelevant;
   if (localLowRisk) {
     return recs;
   }
   const hasFrontend = UI_RE.test(goal) || hasAny(topologyKinds, ["frontend-app"]) || hasAny(topologyFrameworks, ["react", "vue", "next.js", "vite"]);
   const browserRuntime = BROWSER_VERIFY_RE.test(goal) || hasFrontend;
-  const complex = COMPLEX_RE.test(goal) || risk === "high" || risk === "critical" || route === "full-spec" || route === "epic-split";
+  const complex = COMPLEX_RE.test(goal) && route !== "direct-change" || risk === "high" || risk === "critical" || route === "full-spec" || route === "epic-split";
   const stuck = STUCK_RE.test(goal);
   const parallel = PARALLEL_RE.test(goal) || route === "epic-split";
-  if (DOCS_RE.test(goal)) {
+  if (externalDocsRelevant) {
     pushRecommendation(
       recs,
       available,
@@ -1045,25 +1064,25 @@ function renderInstalledCapabilityRules(availableCapabilities) {
 }
 function renderCapabilityDecisionTree(availableCapabilities) {
   const available = new Set(availableCapabilities);
-  const lines = [
-    "1. Can the edit be finished safely from local code in 1-2 steps? -> Do it directly."
+  const rules = [
+    "Can the edit be finished safely from local code in 1-2 steps? -> Do it directly."
   ];
   if (available.has("context7")) {
-    lines.push("2. Does correctness depend on external docs, SDKs, APIs, or Claude Code behavior? -> use the Context7 MCP before editing.");
+    rules.push("Does correctness depend on external docs, SDKs, APIs, or Claude Code behavior? -> use the Context7 MCP before editing.");
   }
   if (available.has("claude-mem")) {
-    lines.push("3. Might similar work, a prior decision, or a repeated failure exist? -> Start with `/claude-mem:mem-search`.");
+    rules.push("Might similar work, a prior decision, or a repeated failure exist? -> Start with `/claude-mem:mem-search`.");
   }
   if (available.has("frontend-design") || available.has("chrome-devtools-mcp")) {
-    lines.push("4. Is visible frontend behavior in scope? -> Use frontend-design for UI decisions and Chrome DevTools MCP for runtime proof when installed.");
+    rules.push("Is visible frontend behavior in scope? -> Use frontend-design for UI decisions and Chrome DevTools MCP for runtime proof when installed.");
   }
   if (available.has("sequential-thinking")) {
-    lines.push("5. Is the work high-risk, architectural, or assumption-heavy? -> Use sequential-thinking after reading the relevant code.");
+    rules.push("Is the work high-risk, architectural, or assumption-heavy? -> Use sequential-thinking after reading the relevant code.");
   }
   if (available.has("pua")) {
-    lines.push("6. Are there multiple failed attempts or truly independent parallel slices? -> Use `/pua:pua-loop` for recovery or `/pua:p9` for bounded parallel planning.");
+    rules.push("Are there multiple failed attempts or truly independent parallel slices? -> Use `/pua:pua-loop` for recovery or `/pua:p9` for bounded parallel planning.");
   }
-  return lines;
+  return rules.map((rule, idx) => `${idx + 1}. ${rule}`);
 }
 function parseList(value) {
   if (!value) return [];
@@ -1133,7 +1152,7 @@ function buildSkipRules(ids) {
 function buildDecisionTree(ids) {
   const out = renderCapabilityDecisionTree([...ids]);
   if (ids.has("curdx-flow")) {
-    out.push("7. Is the request ambiguous, cross-cutting, phase-based, or multi-root? -> Run /curdx-flow:start.");
+    out.push(`${out.length + 1}. Is the request ambiguous, cross-cutting, phase-based, or multi-root? -> Run /curdx-flow:start.`);
   }
   return out;
 }
@@ -1851,7 +1870,7 @@ var analyzeCmd = defineCommand({
     const limit = typeof limitRaw === "string" && limitRaw.length > 0 ? Number(limitRaw) : void 0;
     const topRaw = args.top;
     const top = typeof topRaw === "string" && topRaw.length > 0 ? Number(topRaw) : void 0;
-    const { runAnalyze } = await import("./analyze-FX2PCSL6.mjs");
+    const { runAnalyze } = await import("./analyze-I4TXK4S7.mjs");
     await runAnalyze({
       out: typeof args.out === "string" ? args.out : void 0,
       json: Boolean(args.json),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@curdx/flow",
-  "version": "7.1.19",
+  "version": "7.1.21",
   "description": "Interactive installer for Claude Code plugins and MCP servers",
   "type": "module",
   "bin": "./dist/index.mjs",
@@ -21,6 +21,7 @@
     "test:analyze": "vitest run tests/analyze",
     "test:runner": "vitest run tests/runner",
     "test:claudecc": "npm run build:hooks && node scripts/claudecc-smoke.mjs",
+    "test:claudecc:e2e": "npm run build && npm run build:hooks && node scripts/claudecc-e2e-flow.mjs",
     "start": "node ./dist/index.mjs",
     "typecheck": "tsc --noEmit",
     "verify": "npm run typecheck && npm run check-versions && npm run check:hooks-fresh && npm run build && npm run check:bundle && npm run test:hooks && npm run test:analyze && npm run test:runner && node scripts/check-verification-blocks.mjs",