ai-spec-dev 0.1.0 → 0.17.0

package/core/reviewer.ts

@@ -0,0 +1,259 @@
+import chalk from "chalk";
+import { execSync } from "child_process";
+import * as path from "path";
+import * as fs from "fs-extra";
+import { AIProvider } from "./spec-generator";
+import {
+  reviewSystemPrompt,
+  reviewArchitectureSystemPrompt,
+  reviewImplementationSystemPrompt,
+} from "../prompts/codegen.prompt";
+
+// ─── Review History ────────────────────────────────────────────────────────────
+
+interface ReviewHistoryEntry {
+  date: string;
+  specFile: string;
+  score: number;
+  topIssues: string[];
+}
+
+const REVIEW_HISTORY_FILE = ".ai-spec-reviews.json";
+
+async function loadReviewHistory(projectRoot: string): Promise<ReviewHistoryEntry[]> {
+  const historyPath = path.join(projectRoot, REVIEW_HISTORY_FILE);
+  try {
+    if (await fs.pathExists(historyPath)) {
+      return await fs.readJson(historyPath);
+    }
+  } catch {
+    // ignore
+  }
+  return [];
+}
+
+async function appendReviewHistory(
+  projectRoot: string,
+  entry: ReviewHistoryEntry
+): Promise<void> {
+  const historyPath = path.join(projectRoot, REVIEW_HISTORY_FILE);
+  const existing = await loadReviewHistory(projectRoot);
+  // Keep the last 20 entries
+  const updated = [...existing, entry].slice(-20);
+  try {
+    await fs.writeJson(historyPath, updated, { spaces: 2 });
+  } catch {
+    // ignore — history is non-critical
+  }
+}
+
+/** Extract numeric score from a review result string (looks for "Score: X/10") */
+function extractScore(reviewText: string): number {
+  const match = reviewText.match(/Score:\s*(\d+(?:\.\d+)?)\s*\/\s*10/i);
+  return match ? parseFloat(match[1]) : 0;
+}
+
+/** Extract top issue lines from a review result (lines starting with - or · under ⚠️ section) */
+function extractTopIssues(reviewText: string): string[] {
+  const issuesSection = reviewText.match(/##.*?问题.*?\n([\s\S]*?)(?=##|$)/i)?.[1] ?? "";
+  return issuesSection
+    .split("\n")
+    .filter((l) => /^[-·•*]/.test(l.trim()))
+    .map((l) => l.replace(/^[-·•*]\s*/, "").trim())
+    .filter(Boolean)
+    .slice(0, 3);
+}
+
+/** Format recent review history as a context string for Pass 2 */
+function buildHistoryContext(history: ReviewHistoryEntry[]): string {
+  if (history.length === 0) return "";
+  const recent = history.slice(-5);
+  const lines = ["\n=== 历史审查问题 (Past Review Issues — check if any recur) ==="];
+  for (const entry of recent) {
+    lines.push(`\n[${entry.date}] ${path.basename(entry.specFile)} — Score: ${entry.score}/10`);
+    entry.topIssues.forEach((issue) => lines.push(`  · ${issue}`));
+  }
+  return lines.join("\n") + "\n";
+}
+
+// ─── CodeReviewer ─────────────────────────────────────────────────────────────
+
+export class CodeReviewer {
+  constructor(
+    private provider: AIProvider,
+    private projectRoot: string = process.cwd()
+  ) {}
+
+  private getGitDiff(): string {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    const silent: any = { encoding: "utf-8", stdio: "pipe" };
+    try {
+      execSync("git rev-parse --is-inside-work-tree", silent);
+    } catch {
+      return "";
+    }
+    try {
+      let diff: string = execSync("git diff --cached", silent) as string;
+      if (!diff.trim()) diff = execSync("git diff HEAD", silent) as string;
+      if (!diff.trim()) diff = execSync("git diff", silent) as string;
+      return diff;
+    } catch {
+      return "";
+    }
+  }
+
+  private getDiffStats(diff: string): { files: number; added: number; removed: number } {
+    const lines = diff.split("\n");
+    return {
+      files: lines.filter((l) => l.startsWith("diff --git")).length,
+      added: lines.filter((l) => l.startsWith("+") && !l.startsWith("+++")).length,
+      removed: lines.filter((l) => l.startsWith("-") && !l.startsWith("---")).length,
+    };
+  }
+
+  /**
+   * Two-pass review:
+   *   Pass 1 — architecture (spec compliance, layer separation, auth)
+   *   Pass 2 — implementation details (validation, error handling, edge cases)
+   *             + historical issue recurrence check
+   *
+   * Falls back to single-pass if the two-pass flag is not set.
+   */
+  private async runTwoPassReview(
+    specContent: string,
+    codeContext: string,
+    specFile?: string
+  ): Promise<string> {
+    console.log(chalk.gray("  Pass 1/2: Architecture review..."));
+
+    // ── Pass 1: Architecture ──────────────────────────────────────────────────
+    const archPrompt = `Review the architecture of this change.
+
+=== Feature Spec ===
+${specContent || "(No spec — review for general code quality)"}
+
+=== Code ===
+${codeContext}`;
+
+    const archReview = await this.provider.generate(archPrompt, reviewArchitectureSystemPrompt);
+    console.log(chalk.gray("  Pass 2/2: Implementation review..."));
+
+    // ── Pass 2: Implementation + History ─────────────────────────────────────
+    const history = await loadReviewHistory(this.projectRoot);
+    const historyContext = buildHistoryContext(history);
+
+    const implPrompt = `Review the implementation details of this change.
+
+=== Feature Spec ===
+${specContent || "(No spec — review for general code quality)"}
+
+=== Code ===
+${codeContext}
+
+=== Architecture Review (Pass 1 — do NOT repeat these findings) ===
+${archReview}
+${historyContext}`;
+
+    const implReview = await this.provider.generate(implPrompt, reviewImplementationSystemPrompt);
+
+    // ── Combine ───────────────────────────────────────────────────────────────
+    const combined = `${archReview}\n\n${"─".repeat(52)}\n\n${implReview}`;
+
+    // ── Persist history ───────────────────────────────────────────────────────
+    const score = extractScore(implReview) || extractScore(archReview);
+    const topIssues = extractTopIssues(implReview);
+    if (score > 0 && specFile) {
+      await appendReviewHistory(this.projectRoot, {
+        date: new Date().toISOString().slice(0, 10),
+        specFile: path.relative(this.projectRoot, specFile),
+        score,
+        topIssues,
+      });
+    }
+
+    return combined;
+  }
+
+  async reviewCode(specContent: string, specFile?: string): Promise<string> {
+    console.log(chalk.cyan("\n─── Automated Code Review ───────────────────────"));
+
+    const diff = this.getGitDiff();
+    if (!diff.trim()) {
+      console.log(
+        chalk.yellow("  No git diff found. Stage or commit changes first, then run review.")
+      );
+      console.log(chalk.gray("  Tip: run `git add .` then `ai-spec review` to review your work."));
+      return "No changes";
+    }
+
+    const { files, added, removed } = this.getDiffStats(diff);
+    console.log(
+      chalk.gray(`  Diff: ${files} file(s), ${chalk.green("+" + added)} ${chalk.red("-" + removed)}`)
+    );
+    console.log(
+      chalk.blue(`  Reviewing with ${this.provider.providerName}/${this.provider.modelName}...`)
+    );
+
+    const codeContext = diff.slice(0, 10000);
+    const reviewResult = await this.runTwoPassReview(specContent, codeContext, specFile);
+
+    console.log(chalk.cyan("\n─── Review Result ───────────────────────────────"));
+    console.log(reviewResult);
+    console.log(chalk.cyan("─────────────────────────────────────────────────\n"));
+
+    return reviewResult;
+  }
+
+  /**
+   * Review directly from generated file contents (for api mode where git diff is empty).
+   */
+  async reviewFiles(
+    specContent: string,
+    filePaths: string[],
+    workingDir: string,
+    specFile?: string
+  ): Promise<string> {
+    console.log(chalk.cyan("\n─── Automated Code Review (file-based) ─────────"));
+    console.log(chalk.gray(`  Reviewing ${filePaths.length} generated file(s)...`));
+    console.log(
+      chalk.blue(`  Reviewing with ${this.provider.providerName}/${this.provider.modelName}...`)
+    );
+
+    let filesSection = "";
+    for (const filePath of filePaths) {
+      const fullPath = path.join(workingDir, filePath);
+      try {
+        const content = await fs.readFile(fullPath, "utf-8");
+        filesSection += `\n\n=== ${filePath} ===\n${content.slice(0, 3000)}`;
+        if (content.length > 3000) filesSection += `\n... (truncated, ${content.length} chars total)`;
+      } catch {
+        filesSection += `\n\n=== ${filePath} ===\n(file not found)`;
+      }
+    }
+
+    const reviewResult = await this.runTwoPassReview(specContent, filesSection, specFile);
+
+    console.log(chalk.cyan("\n─── Review Result ───────────────────────────────"));
+    console.log(reviewResult);
+    console.log(chalk.cyan("─────────────────────────────────────────────────\n"));
+
+    return reviewResult;
+  }
+
+  /** Print score trend from history (last N reviews) */
+  async printScoreTrend(limit = 5): Promise<void> {
+    const history = await loadReviewHistory(this.projectRoot);
+    if (history.length === 0) {
+      console.log(chalk.gray("  No review history yet."));
+      return;
+    }
+    const recent = history.slice(-limit);
+    console.log(chalk.cyan("\n─── Review Score Trend ──────────────────────────"));
+    for (const entry of recent) {
+      const bar = "█".repeat(entry.score) + "░".repeat(10 - entry.score);
+      const color = entry.score >= 8 ? chalk.green : entry.score >= 6 ? chalk.yellow : chalk.red;
+      console.log(`  ${entry.date}  [${color(bar)}] ${color(entry.score + "/10")}  ${path.basename(entry.specFile)}`);
+    }
+    console.log(chalk.cyan("─────────────────────────────────────────────────"));
+  }
+}

package/core/spec-assessor.ts

@@ -0,0 +1,99 @@
+import chalk from "chalk";
+import { AIProvider } from "./spec-generator";
+import { specAssessSystemPrompt } from "../prompts/spec-assess.prompt";
+
+// ─── Types ─────────────────────────────────────────────────────────────────────
+
+export interface SpecAssessment {
+  coverageScore: number;
+  clarityScore: number;
+  constitutionScore: number;
+  overallScore: number;
+  issues: string[];
+  suggestions: string[];
+  dslExtractable: boolean;
+}
+
+// ─── Helpers ───────────────────────────────────────────────────────────────────
+
+function scoreBar(score: number): string {
+  const filled = Math.round(score);
+  const bar = chalk.green("█".repeat(filled)) + chalk.gray("░".repeat(10 - filled));
+  const color = score >= 8 ? chalk.green : score >= 6 ? chalk.yellow : chalk.red;
+  return `[${bar}] ${color(score + "/10")}`;
+}
+
+function parseAssessment(raw: string): SpecAssessment | null {
+  // Strip markdown fences if present
+  const stripped = raw.replace(/```(?:json)?\n?/g, "").replace(/```\s*$/g, "").trim();
+  try {
+    const parsed = JSON.parse(stripped);
+    if (
+      typeof parsed.coverageScore === "number" &&
+      typeof parsed.clarityScore === "number" &&
+      typeof parsed.overallScore === "number"
+    ) {
+      return parsed as SpecAssessment;
+    }
+  } catch {
+    // fall through
+  }
+  return null;
+}
+
+// ─── Main export ───────────────────────────────────────────────────────────────
+
+/**
+ * Run a pre-Approval-Gate quality check on the spec.
+ * Advisory only — does not block the flow regardless of scores.
+ * Returns null if the AI call fails (graceful degradation).
+ */
+export async function assessSpec(
+  provider: AIProvider,
+  spec: string,
+  constitution?: string
+): Promise<SpecAssessment | null> {
+  const prompt = `Assess the following feature specification.
+${constitution ? `\n=== Project Constitution (check consistency against this) ===\n${constitution}\n` : ""}
+=== Feature Spec ===
+${spec}`;
+
+  try {
+    const raw = await provider.generate(prompt, specAssessSystemPrompt);
+    return parseAssessment(raw);
+  } catch {
+    return null;
+  }
+}
+
+/**
+ * Print the spec assessment panel to stdout.
+ */
+export function printSpecAssessment(assessment: SpecAssessment): void {
+  console.log(chalk.blue("\n─── Spec Quality Assessment ─────────────────────"));
+  console.log(`  Coverage    ${scoreBar(assessment.coverageScore)}  error handling, edge cases, auth`);
+  console.log(`  Clarity     ${scoreBar(assessment.clarityScore)}  API contracts, response shapes`);
+  console.log(`  Constitution${scoreBar(assessment.constitutionScore)}  naming, error codes, conventions`);
+  console.log(chalk.bold(`  Overall     ${scoreBar(assessment.overallScore)}`));
+
+  if (!assessment.dslExtractable) {
+    console.log(
+      chalk.yellow(
+        "\n  ⚠  DSL extraction may be unreliable — clarityScore < 6 or no structured API section."
+      )
+    );
+    console.log(chalk.gray("     Consider adding explicit request/response shapes before proceeding."));
+  }
+
+  if (assessment.issues.length > 0) {
+    console.log(chalk.yellow(`\n  Issues found (${assessment.issues.length}):`));
+    assessment.issues.forEach((issue) => console.log(chalk.yellow(`  · ${issue}`)));
+  }
+
+  if (assessment.suggestions.length > 0) {
+    console.log(chalk.cyan("\n  Suggestions:"));
+    assessment.suggestions.forEach((s) => console.log(chalk.cyan(`  💡 ${s}`)));
+  }
+
+  console.log(chalk.blue("─────────────────────────────────────────────────"));
+}