@vibecodeqa/cli 0.38.1 → 0.39.0

package/dist/runners/architecture.d.ts

@@ -23,5 +23,5 @@ export interface ArchGraph {
     godModules: string[];
     orphans: string[];
 }
-export declare function runArchitecture(cwd: string, workspace?: WorkspaceInfo): CheckResult;
+export declare function runArchitecture(cwd: string, workspace?: WorkspaceInfo): Promise<CheckResult>;
 export { generateArchSVG, generateDSM, generateLayerDiagram, generatePackageDiagram, generateSequenceDiagram } from "./diagrams.js";

package/dist/runners/architecture.js

@@ -9,11 +9,13 @@
  *   6. Layer violations (optional: detect cross-layer imports)
  *   7. SVG architecture diagram
  */
+import { existsSync } from "node:fs";
 import { basename, dirname, extname } from "node:path";
+import { cruise } from "dependency-cruiser";
 import { getProductionFiles } from "../fs-utils.js";
 import { gradeFromScore } from "../types.js";
 import { generateContainerDiagram } from "./diagrams.js";
-export function runArchitecture(cwd, workspace) {
+export async function runArchitecture(cwd, workspace) {
     const start = Date.now();
     const issues = [];
     const files = getProductionFiles(cwd);
@@ -27,7 +29,12 @@ export function runArchitecture(cwd, workspace) {
             duration: Date.now() - start,
         };
     }
-    const graph = buildGraph(files);
+    // Prefer dependency-cruiser (real module resolution incl. tsconfig paths) and
+    // fall back to the built-in resolver when it can't cover the project (e.g. SFC-
+    // heavy or monorepos where cross-package imports use bare specifiers).
+    const cruised = await buildGraphViaCruise(cwd, files);
+    const graph = cruised ?? buildGraph(files);
+    const tool = cruised ? "dependency-cruiser" : "built-in";
     // ── Circular dependencies ──
     const cycles = findCycles(graph.nodes);
     for (const cycle of cycles.slice(0, 5)) {
@@ -142,12 +149,77 @@ export function runArchitecture(cwd, workspace) {
             graph: graphData,
             containerSvg: generateContainerDiagram(cwd),
             assessment,
+            tool,
         },
         issues,
         duration: Date.now() - start,
     };
 }
 // ── Graph building ──
+/** Build the import graph with dependency-cruiser (real module resolution,
+ *  tsconfig path aliases, transitive TS deps). Returns null — so the caller
+ *  falls back to the built-in resolver — if it errors or covers too little of
+ *  the project (e.g. .vue/.svelte SFCs it can't resolve, or monorepo bare imports). */
+async function buildGraphViaCruise(cwd, files) {
+    // .vue/.svelte single-file components need our SFC-aware resolver; dependency-cruiser
+    // doesn't resolve them without a bundler plugin, so hand those projects to the built-in.
+    if (files.some((f) => f.ext === ".vue" || f.ext === ".svelte"))
+        return null;
+    const fileSet = new Set(files.map((f) => f.path));
+    const roots = [...new Set(files.map((f) => f.path.split("/")[0]).filter(Boolean))];
+    if (roots.length === 0)
+        return null;
+    const prevCwd = process.cwd();
+    try {
+        process.chdir(cwd); // dependency-cruiser resolves relative to process.cwd()
+        const options = {
+            doNotFollow: { path: "node_modules" },
+            exclude: { path: "(\\.test\\.|\\.spec\\.|/node_modules/|/dist/|/\\.vibe-check/)" },
+            tsPreCompilationDeps: true,
+        };
+        if (existsSync("tsconfig.json"))
+            options.tsConfig = { fileName: "tsconfig.json" };
+        const result = await cruise(roots, options);
+        const output = result.output;
+        if (typeof output === "string")
+            return null;
+        const modules = output.modules ?? [];
+        const nodes = new Map();
+        for (const m of modules) {
+            if (fileSet.has(m.source)) {
+                nodes.set(m.source, { path: m.source, imports: [], importedBy: [], dir: dirname(m.source), exports: 0 });
+            }
+        }
+        for (const m of modules) {
+            const node = nodes.get(m.source);
+            if (!node)
+                continue;
+            const seen = new Set();
+            for (const d of m.dependencies ?? []) {
+                if (d.coreModule || !d.resolved || d.resolved === m.source)
+                    continue;
+                if (!fileSet.has(d.resolved) || seen.has(d.resolved))
+                    continue;
+                seen.add(d.resolved);
+                node.imports.push(d.resolved);
+            }
+        }
+        for (const [path, node] of nodes) {
+            for (const imp of node.imports)
+                nodes.get(imp)?.importedBy.push(path);
+        }
+        // Too little coverage → the built-in resolver is a better bet for this project.
+        if (nodes.size < files.length * 0.5)
+            return null;
+        return { nodes };
+    }
+    catch {
+        return null;
+    }
+    finally {
+        process.chdir(prevCwd);
+    }
+}
 function buildGraph(files) {
     const filePaths = new Set(files.map((f) => f.path));
     const nodes = new Map();

package/dist/runners/duplication.d.ts

@@ -1,3 +1,24 @@
-/** Code duplication — delegates to jscpd when available, falls back to built-in line-hash. */
+/** Code duplication.
+ *
+ *  - If the project depends on jscpd, shell out to it (opt-in, richest output).
+ *  - Otherwise run jscpd's detection engine (@jscpd/core — battle-tested Rabin-Karp
+ *    matching + maximal-clone extension) over our own lightweight tokenizer. We
+ *    supply the tokens (zero heavy language-grammar dependency) and let the mature
+ *    engine do the matching/merging/statistics.
+ *
+ *  Catches Type-1/2 clones (exact, modulo formatting), like jscpd's default. */
 import type { CheckResult } from "../types.js";
-export declare function runDuplication(cwd: string): CheckResult;
+export declare function runDuplication(cwd: string): Promise<CheckResult>;
+interface Token {
+    text: string;
+    line: number;
+    endLine: number;
+    start: number;
+    end: number;
+}
+/** Lex source into tokens, skipping whitespace and comments. */
+export declare function tokenize(content: string): Token[];
+/** Blank out import / re-export statements (preserving line numbers) before
+ *  tokenizing. Identical import headers aren't refactorable duplication. */
+export declare function stripImports(content: string): string;
+export {};

package/dist/runners/duplication.js

@@ -1,12 +1,27 @@
-/** Code duplication — delegates to jscpd when available, falls back to built-in line-hash. */
+/** Code duplication.
+ *
+ *  - If the project depends on jscpd, shell out to it (opt-in, richest output).
+ *  - Otherwise run jscpd's detection engine (@jscpd/core — battle-tested Rabin-Karp
+ *    matching + maximal-clone extension) over our own lightweight tokenizer. We
+ *    supply the tokens (zero heavy language-grammar dependency) and let the mature
+ *    engine do the matching/merging/statistics.
+ *
+ *  Catches Type-1/2 clones (exact, modulo formatting), like jscpd's default. */
+import { createHash } from "node:crypto";
 import { readFileSync, rmdirSync, unlinkSync } from "node:fs";
 import { join } from "node:path";
+import { Detector, getDefaultOptions, MemoryStore, Statistic, } from "@jscpd/core";
 import { getProductionFiles, readDeps } from "../fs-utils.js";
 import { gradeFromScore } from "../types.js";
 import { run } from "./exec.js";
-const MIN_LINES = 6; // minimum duplicate block size
-const MIN_TOKENS = 50; // minimum token count for a duplicate
-export function runDuplication(cwd) {
+// jscpd-parity thresholds: ≥ MIN_TOKENS consecutive matching tokens AND ≥ MIN_LINES lines.
+const MIN_TOKENS = 50;
+const MIN_LINES = 6;
+const MAX_ISSUES = 20;
+// One synthetic "format" for every file so clones are detected across all sources
+// (jscpd only compares within a format; we want cross-file/cross-extension matches).
+const FORMAT = "src";
+export async function runDuplication(cwd) {
     const start = Date.now();
     // Try jscpd if it's an explicit project dependency (opt-in, not auto-npx)
     const deps = readDeps(cwd);
@@ -15,108 +30,244 @@ export function runDuplication(cwd) {
         jscpdResult.duration = Date.now() - start;
         return jscpdResult;
     }
-    // Fallback: built-in line-hash detection
-    const issues = [];
-    const sourceFiles = getProductionFiles(cwd);
-    if (sourceFiles.length < 2) {
+    const files = getProductionFiles(cwd);
+    if (files.length < 2) {
         return {
             name: "duplication",
             score: 100,
             grade: "A",
-            details: { filesScanned: sourceFiles.length, duplicates: 0, tool: "built-in" },
+            details: { filesScanned: files.length, duplicates: 0, tool: "built-in" },
             issues: [],
             duration: Date.now() - start,
         };
     }
-    // Simple line-based duplicate detection
-    // Build a map of normalized line hashes → locations
-    const lineMap = new Map();
-    let totalSourceLines = 0;
-    for (const sf of sourceFiles) {
-        const lines = sf.content.split("\n");
-        totalSourceLines += lines.length;
-        for (let i = 0; i <= lines.length - MIN_LINES; i++) {
-            const block = lines
-                .slice(i, i + MIN_LINES)
-                .map((l) => l.trim())
-                .filter((l) => l.length > 0 &&
-                !l.startsWith("//") &&
-                !l.startsWith("*") &&
-                !l.startsWith("import ") &&
-                !l.startsWith("export {") &&
-                l !== "{" &&
-                l !== "}" &&
-                l !== "");
-            if (block.length < MIN_LINES - 2)
-                continue; // too many empty/trivial lines
-            const key = block.join("\n");
-            if (key.length < MIN_TOKENS)
-                continue;
-            const locs = lineMap.get(key) || [];
-            locs.push({ file: sf.path, line: i + 1 });
-            lineMap.set(key, locs);
-        }
-    }
-    // Find blocks that appear in 2+ locations
-    const duplicates = [];
-    const seen = new Set();
-    for (const [key, locs] of lineMap) {
-        if (locs.length < 2)
-            continue;
-        // Deduplicate: same file, adjacent lines are the same block
-        const unique = locs.filter((l, i) => i === 0 || l.file !== locs[i - 1].file || l.line > locs[i - 1].line + MIN_LINES);
-        if (unique.length < 2)
-            continue;
-        // Only report each pair once
-        for (let i = 0; i < unique.length - 1; i++) {
-            const a = unique[i];
-            const b = unique[i + 1];
-            const pairKey = `${a.file}:${a.line}-${b.file}:${b.line}`;
-            if (seen.has(pairKey))
-                continue;
-            seen.add(pairKey);
-            duplicates.push({ fileA: a.file, lineA: a.line, fileB: b.file, lineB: b.line, lines: MIN_LINES, content: key });
-        }
-    }
-    for (const d of duplicates.slice(0, 20)) {
-        // Show first 3 lines of the duplicated content, truncate at word boundary
-        const lines = d.content.split("\n").slice(0, 3);
-        const preview = lines.join(" \u2502 "); // use │ separator
-        const maxLen = 120;
-        const truncated = preview.length > maxLen ? `${preview.slice(0, preview.lastIndexOf(" ", maxLen) || maxLen)}...` : preview;
-        // First line of content is the best search string
-        const searchSnippet = d.content.split("\n")[0];
-        issues.push({
-            severity: "warning",
-            message: `Duplicate (${d.lines} lines): ${truncated}`,
-            file: `${d.fileA}:${d.lineA} ↔ ${d.fileB}:${d.lineB}`,
-            rule: "duplicate-code",
-            snippet: searchSnippet,
-        });
-    }
-    const dupPct = totalSourceLines > 0 ? Math.round((duplicates.length * MIN_LINES * 100) / totalSourceLines) : 0;
-    // Score based on duplication percentage (not absolute count)
-    const score = Math.max(0, Math.min(100, 100 - dupPct * 5));
+    const { clones, percentage, totalLines } = await detectWithCore(files);
+    const score = scoreFromPct(percentage);
+    // Rank biggest clones first; stable tie-break for deterministic output.
+    clones.sort((a, b) => b.lines - a.lines || a.fileA.localeCompare(b.fileA) || a.lineA - b.lineA);
+    const issues = clones.slice(0, MAX_ISSUES).map((c) => ({
+        severity: "warning",
+        message: `Duplicate (${c.lines} lines): ${c.snippet.slice(0, 100)}`,
+        file: `${c.fileA}:${c.lineA} ↔ ${c.fileB}:${c.lineB}`,
+        rule: "duplicate-code",
+        snippet: c.snippet,
+    }));
     return {
         name: "duplication",
         score,
         grade: gradeFromScore(score),
         details: {
-            filesScanned: sourceFiles.length,
-            totalSourceLines,
-            duplicateBlocks: duplicates.length,
-            duplicationPct: `${dupPct}%`,
+            filesScanned: files.length,
+            totalSourceLines: totalLines,
+            duplicateBlocks: clones.length,
+            duplicationPct: `${percentage}%`,
             tool: "built-in",
         },
         issues,
         duration: Date.now() - start,
     };
 }
+/** Run @jscpd/core over our tokens. Files are fed sequentially into one Detector
+ *  sharing a store, so later files match earlier ones (cross-file clones). */
+async function detectWithCore(files) {
+    const options = { ...getDefaultOptions(), minTokens: MIN_TOKENS, minLines: MIN_LINES, maxLines: 100_000 };
+    const detector = new Detector(new VcqaTokenizer(), new MemoryStore(), undefined, options);
+    const statistic = new Statistic();
+    const handlers = statistic.subscribe();
+    for (const event of Object.keys(handlers)) {
+        const handler = handlers[event];
+        if (handler)
+            detector.on(event, handler);
+    }
+    const clones = [];
+    for (const f of files) {
+        const found = await detector.detect(f.path, f.content, FORMAT);
+        for (const c of found) {
+            const a = c.duplicationA; // the current file
+            const b = c.duplicationB; // the earlier match
+            const lines = Math.max(1, a.end.line - a.start.line + 1);
+            const snippet = (f.content.split("\n")[a.start.line - 1] ?? "").trim();
+            clones.push({ fileA: a.sourceId, lineA: a.start.line, fileB: b.sourceId, lineB: b.start.line, lines, snippet });
+        }
+    }
+    const stat = statistic.getStatistic();
+    const percentage = Math.round((stat.total.percentage ?? 0) * 10) / 10;
+    return { clones, percentage, totalLines: stat.total.lines ?? 0 };
+}
+const isWord = (ch) => (ch >= "a" && ch <= "z") || (ch >= "A" && ch <= "Z") || (ch >= "0" && ch <= "9") || ch === "_" || ch === "$";
+/** Consume a string/template literal, returning its full text. One token, so the
+ *  detector never matches *inside* a string; text retained so distinct strings differ. */
+function scanString(content, cur) {
+    const quote = content[cur.i];
+    const start = cur.i;
+    const n = content.length;
+    cur.i++;
+    while (cur.i < n && content[cur.i] !== quote) {
+        if (content[cur.i] === "\\") {
+            if (content[cur.i + 1] === "\n")
+                cur.line++;
+            cur.i += 2;
+            continue;
+        }
+        if (content[cur.i] === "\n")
+            cur.line++;
+        cur.i++;
+    }
+    const text = content.slice(start, Math.min(cur.i + 1, n));
+    cur.i++; // skip closing quote
+    return text;
+}
+/** Consume a block comment, advancing past the closing star-slash. */
+function scanBlockComment(content, cur) {
+    const n = content.length;
+    cur.i += 2;
+    while (cur.i < n && !(content[cur.i] === "*" && content[cur.i + 1] === "/")) {
+        if (content[cur.i] === "\n")
+            cur.line++;
+        cur.i++;
+    }
+    cur.i += 2;
+}
+/** Lex source into tokens, skipping whitespace and comments. */
+export function tokenize(content) {
+    const tokens = [];
+    const n = content.length;
+    const cur = { i: 0, line: 1 };
+    while (cur.i < n) {
+        const c = content[cur.i];
+        if (c === "\n") {
+            cur.line++;
+            cur.i++;
+            continue;
+        }
+        if (c === " " || c === "\t" || c === "\r") {
+            cur.i++;
+            continue;
+        }
+        if (c === "/" && content[cur.i + 1] === "/") {
+            while (cur.i < n && content[cur.i] !== "\n")
+                cur.i++;
+            continue;
+        }
+        if (c === "/" && content[cur.i + 1] === "*") {
+            scanBlockComment(content, cur);
+            continue;
+        }
+        const startLine = cur.line;
+        const startPos = cur.i;
+        if (c === '"' || c === "'" || c === "`") {
+            const text = scanString(content, cur);
+            tokens.push({ text, line: startLine, endLine: cur.line, start: startPos, end: cur.i });
+            continue;
+        }
+        if (isWord(c)) {
+            let j = cur.i + 1;
+            while (j < n && isWord(content[j]))
+                j++;
+            tokens.push({ text: content.slice(cur.i, j), line: startLine, endLine: startLine, start: cur.i, end: j });
+            cur.i = j;
+            continue;
+        }
+        tokens.push({ text: c, line: startLine, endLine: startLine, start: startPos, end: cur.i + 1 }); // punctuation
+        cur.i++;
+    }
+    return tokens;
+}
+/** Blank out import / re-export statements (preserving line numbers) before
+ *  tokenizing. Identical import headers aren't refactorable duplication. */
+export function stripImports(content) {
+    const lines = content.split("\n");
+    let cont = false; // inside a multi-line import
+    for (let i = 0; i < lines.length; i++) {
+        const t = lines[i].trim();
+        if (cont) {
+            lines[i] = "";
+            if (/from\s*['"][^'"]*['"]\s*;?\s*$/.test(t) || /;\s*$/.test(t))
+                cont = false;
+            continue;
+        }
+        if (/^import\b/.test(t)) {
+            const done = /from\s*['"][^'"]*['"]\s*;?\s*$/.test(t) || /^import\s*['"][^'"]*['"]\s*;?\s*$/.test(t);
+            lines[i] = "";
+            if (!done)
+                cont = true;
+            continue;
+        }
+        if (/^export\s*(\*|\{[^}]*\})\s*from\s*['"]/.test(t))
+            lines[i] = "";
+    }
+    return lines.join("\n");
+}
+function toIToken(t, format) {
+    return {
+        type: "code",
+        value: t.text,
+        length: t.text.length,
+        format,
+        range: [t.start, t.end],
+        loc: { start: { line: t.line, column: 0, position: t.start }, end: { line: t.endLine, column: 0, position: t.end } },
+    };
+}
+/** Bridges our tokens into the @jscpd/core engine: hashes rolling MIN_TOKENS-token
+ *  windows into map frames the Rabin-Karp detector matches against. */
+class VcqaTokenizer {
+    generateMaps(id, data, format, options) {
+        const w = options.minTokens ?? MIN_TOKENS;
+        const toks = tokenize(stripImports(data)).map((t) => toIToken(t, format));
+        const frames = [];
+        for (let i = 0; i + w <= toks.length; i++) {
+            const id2 = createHash("md5").update(toks.slice(i, i + w).map((t) => t.value).join("")).digest("hex");
+            frames.push({ id: id2, sourceId: id, start: toks[i], end: toks[i + w - 1] });
+        }
+        return [new VcqaTokensMap(id, format, data, toks.length, frames)];
+    }
+}
+class VcqaTokensMap {
+    id;
+    format;
+    data;
+    tokensCount;
+    frames;
+    p = 0;
+    constructor(id, format, data, tokensCount, frames) {
+        this.id = id;
+        this.format = format;
+        this.data = data;
+        this.tokensCount = tokensCount;
+        this.frames = frames;
+    }
+    getFormat() { return this.format; }
+    getId() { return this.id; }
+    getLinesCount() { return this.data.split("\n").length; }
+    getTokensCount() { return this.tokensCount; }
+    next() {
+        if (this.p < this.frames.length)
+            return { value: this.frames[this.p++], done: false };
+        return { value: false, done: true };
+    }
+}
+/** Shared scoring band (used by both built-in and jscpd paths).
+ *  Industry benchmarks: <5% great, 5-10% good, 10-20% acceptable, >20% needs work. */
+function scoreFromPct(dupPct) {
+    if (dupPct <= 5)
+        return 100;
+    if (dupPct <= 10)
+        return 90;
+    if (dupPct <= 20)
+        return 75;
+    if (dupPct <= 30)
+        return 60;
+    if (dupPct <= 40)
+        return 45;
+    if (dupPct <= 50)
+        return 30;
+    return Math.max(10, Math.round(30 - (dupPct - 50)));
+}
 function tryJscpd(cwd) {
     // jscpd writes JSON to a file, not stdout. Use a temp output dir.
     const tmpDir = join(cwd, ".vibe-check", "jscpd-tmp");
     const ignores = "node_modules/**,dist/**,build/**,.vibe-check/**,coverage/**,.next/**,.nuxt/**,**/*.json,**/*.lock,**/*.yaml,**/*.md";
-    run(`npx jscpd . --min-lines 6 --min-tokens 50 --reporters json --output "${tmpDir}" --ignore "${ignores}" --silent 2>/dev/null || true`, cwd, 30_000);
+    run(`npx jscpd . --min-lines ${MIN_LINES} --min-tokens ${MIN_TOKENS} --reporters json --output "${tmpDir}" --ignore "${ignores}" --silent 2>/dev/null || true`, cwd, 30_000);
     const reportPath = join(tmpDir, "jscpd-report.json");
     let rawData;
     try {
@@ -139,33 +290,19 @@ function tryJscpd(cwd) {
             return null;
         const issues = [];
         const clones = data.duplicates || [];
-        for (const d of clones.slice(0, 20)) {
+        for (const d of clones.slice(0, MAX_ISSUES)) {
             const fileA = d.firstFile?.name || "?";
             const fileB = d.secondFile?.name || "?";
             const lines = d.lines || 0;
             issues.push({
                 severity: "warning",
                 message: `Duplicate (${lines} lines)`,
-                file: `${fileA}:${d.firstFile?.start} \u2194 ${fileB}:${d.secondFile?.start}`,
+                file: `${fileA}:${d.firstFile?.start} ↔ ${fileB}:${d.secondFile?.start}`,
                 rule: "duplicate-code",
             });
         }
         const dupPct = Math.round((data.statistics.total?.percentage || 0) * 100) / 100;
-        // jscpd token-level detection is more aggressive than line-hash
-        // Industry benchmarks: <10% good, 10-20% acceptable, 20-40% needs work, >40% poor
-        const score = dupPct <= 5
-            ? 100
-            : dupPct <= 10
-                ? 90
-                : dupPct <= 20
-                    ? 75
-                    : dupPct <= 30
-                        ? 60
-                        : dupPct <= 40
-                            ? 45
-                            : dupPct <= 50
-                                ? 30
-                                : Math.max(10, Math.round(30 - (dupPct - 50)));
+        const score = scoreFromPct(dupPct);
         return {
             name: "duplication",
             score,

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@vibecodeqa/cli",
-	"version": "0.38.1",
+	"version": "0.39.0",
 	"description": "Code health scanner for the AI coding era. 25 checks, zero config, full report.",
 	"type": "module",
 	"bin": {
@@ -53,6 +53,8 @@
 		"vitest": "^4.1.6"
 	},
 	"dependencies": {
+		"@jscpd/core": "^4.2.4",
+		"dependency-cruiser": "^17.4.3",
 		"ink": "^5.2.1",
 		"react": "^18.3.1"
 	}