edsger 0.81.0 → 0.82.0

package/dist/index.js

@@ -1,13 +1,13 @@
 #!/usr/bin/env node
 
 // src/index.ts
-import { readFileSync as readFileSync3 } from "fs";
+import { readFileSync as readFileSync4 } from "fs";
 import { resolve as resolve2 } from "path";
 import { Command, Option } from "commander";
-import pc7 from "picocolors";
+import pc9 from "picocolors";
 
 // src/commands/benchmark.ts
-import { mkdirSync, writeFileSync } from "fs";
+import { appendFileSync, mkdirSync, writeFileSync } from "fs";
 import { basename, join as join3 } from "path";
 import pc3 from "picocolors";
 
@@ -490,28 +490,55 @@ function summarizeTool(name = "tool", input = {}) {
   }
 }
 
+// src/core/git.ts
+import { execFileSync as execFileSync2 } from "child_process";
+function tryGit(args, cwd) {
+  try {
+    return execFileSync2("git", args, {
+      cwd,
+      encoding: "utf8",
+      stdio: ["ignore", "pipe", "ignore"]
+    }).trim();
+  } catch {
+    return null;
+  }
+}
+function gitInfo(cwd) {
+  const commit = tryGit(["rev-parse", "--short", "HEAD"], cwd);
+  const branch = tryGit(["rev-parse", "--abbrev-ref", "HEAD"], cwd);
+  const status = tryGit(["status", "--porcelain"], cwd);
+  return {
+    commit,
+    branch: branch && branch !== "HEAD" ? branch : null,
+    dirty: Boolean(status && status.length > 0)
+  };
+}
+
 // src/core/report.ts
 function scoreBenchmark(standard, result) {
   const scale = standard.scoreScale || 4;
   const totalWeight = standard.categories.reduce((s, c) => s + c.weight, 0) || 1;
-  const scoreLookup = /* @__PURE__ */ new Map();
-  for (const cat of result.categories) {
-    for (const cr of cat.criteria) scoreLookup.set(`${cat.id}/${cr.id}`, cr.score);
-  }
-  const justifyLookup = /* @__PURE__ */ new Map();
+  const tail = (id) => id.slice(id.lastIndexOf("/") + 1);
+  const byKey = /* @__PURE__ */ new Map();
+  const byCrit = /* @__PURE__ */ new Map();
   for (const cat of result.categories) {
     for (const cr of cat.criteria) {
-      justifyLookup.set(`${cat.id}/${cr.id}`, {
+      const meta = {
+        score: cr.score,
         justification: cr.justification,
         evidence: cr.evidence ?? []
-      });
+      };
+      byKey.set(`${tail(cat.id)}/${tail(cr.id)}`, meta);
+      byCrit.set(tail(cr.id), meta);
     }
   }
+  let matched = 0;
   const categories = standard.categories.map((cat) => {
     const critWeightTotal = cat.criteria.reduce((s, c) => s + (c.weight ?? 1), 0) || 1;
     const scored = cat.criteria.map((cr) => {
-      const raw = clamp(scoreLookup.get(`${cat.id}/${cr.id}`) ?? 0, 0, scale);
-      const meta = justifyLookup.get(`${cat.id}/${cr.id}`);
+      const meta = byKey.get(`${tail(cat.id)}/${tail(cr.id)}`) ?? byCrit.get(tail(cr.id));
+      if (meta) matched += 1;
+      const raw = clamp(meta?.score ?? 0, 0, scale);
       return {
         id: cr.id,
         title: cr.title,
@@ -534,6 +561,14 @@ function scoreBenchmark(standard, result) {
       criteria: scored
     };
   });
+  const totalCriteria = standard.categories.reduce((s, c) => s + c.criteria.length, 0);
+  if (result.categories.length > 0 && matched === 0) {
+    log.warn(
+      "The agent returned scores but none matched the rubric's criterion ids \u2014 reporting 0. This is a scoring bug, not a real result."
+    );
+  } else if (matched < totalCriteria) {
+    log.warn(`Only ${matched}/${totalCriteria} criteria were scored; the rest default to 0.`);
+  }
   const overall = round(
     categories.reduce((s, c) => s + c.points, 0),
     1
@@ -704,6 +739,7 @@ async function runBenchmark(opts) {
   const scored = scoreBenchmark(standard, result);
   const generatedAt = (/* @__PURE__ */ new Date()).toISOString();
   const runId = generatedAt.replace(/[-:]/g, "").replace(/\.\d+Z$/, "Z");
+  const git = gitInfo(opts.cwd);
   const md = renderBenchmarkMarkdown(scored, {
     repo: basename(opts.cwd),
     languages: resolved.languages,
@@ -717,6 +753,7 @@ async function runBenchmark(opts) {
     generatedAt,
     model: opts.model,
     repo: basename(opts.cwd),
+    git,
     languages: resolved.languages,
     standard: { name: standard.name, version: standard.version, layers: resolved.layers },
     overall: scored.overall,
@@ -728,6 +765,19 @@ async function runBenchmark(opts) {
     recommendations: scored.recommendations,
     costUsd
   };
+  const historyRecord = {
+    runId,
+    generatedAt,
+    commit: git.commit,
+    branch: git.branch,
+    dirty: git.dirty,
+    model: opts.model,
+    standardVersion: standard.version,
+    overall: scored.overall,
+    grade: scored.grade,
+    categories: Object.fromEntries(scored.categories.map((c) => [c.id, c.fraction])),
+    costUsd
+  };
   if (!opts.noWrite && !opts.dryRun) {
     const baseDir = join3(opts.cwd, opts.out);
     const mdOut = md + "\n";
@@ -738,6 +788,7 @@ async function runBenchmark(opts) {
     writeFileSync(join3(runDir, "report.json"), jsonOut, "utf8");
     writeFileSync(join3(baseDir, "report.md"), mdOut, "utf8");
     writeFileSync(join3(baseDir, "report.json"), jsonOut, "utf8");
+    appendFileSync(join3(baseDir, "history.jsonl"), JSON.stringify(historyRecord) + "\n", "utf8");
     log.success(
       `Report written to ${pc3.bold(join3(opts.out, "report.md"))} ${pc3.dim(`(run archived at ${join3(opts.out, "runs", runId)})`)}`
     );
@@ -765,7 +816,7 @@ function buildPrompt(standard, cwd) {
   const rubric = standard.categories.map((cat) => {
     const crits = cat.criteria.map((cr) => {
       const anchors = cr.rubric ? Object.entries(cr.rubric).map(([k, v]) => `      - ${k}: ${v}`).join("\n") : "";
-      return `    - id: ${cat.id}/${cr.id}
+      return `    - id: ${cr.id}
       title: ${cr.title}
       guidance: ${cr.guidance.trim()}` + (anchors ? `
       anchors:
@@ -1210,49 +1261,212 @@ function buildPrompt3(ctx, feedback, instruction) {
   ].join("\n");
 }
 
+// src/commands/serve.ts
+import { createReadStream, existsSync as existsSync4, statSync as statSync2 } from "fs";
+import { createServer } from "http";
+import { extname, join as join6, normalize } from "path";
+import pc6 from "picocolors";
+
+// src/core/dashboard.ts
+import { existsSync as existsSync3, readFileSync as readFileSync2, readdirSync as readdirSync2 } from "fs";
+import { join as join5 } from "path";
+function assembleData(benchmarkDir) {
+  const latest = readJson(join5(benchmarkDir, "report.json"));
+  const history = readJsonl(join5(benchmarkDir, "history.jsonl"));
+  const runs = {};
+  const runsDir = join5(benchmarkDir, "runs");
+  if (existsSync3(runsDir)) {
+    for (const entry of readdirSync2(runsDir)) {
+      const report = readJson(join5(runsDir, entry, "report.json"));
+      if (report) runs[entry] = report;
+    }
+  }
+  const repo = latest && typeof latest === "object" && "repo" in latest ? latest.repo : void 0;
+  return { repo, latest, history, runs };
+}
+function hasData(data) {
+  return Boolean(data.latest) || data.history.length > 0;
+}
+function findViewerDir() {
+  const dir = join5(findBundledAssetsDir(), "viewer");
+  if (!existsSync3(join5(dir, "index.html"))) {
+    throw new Error(
+      "Dashboard viewer assets not found. Reinstall edsger, or run `npm run sync-viewer` in a dev checkout."
+    );
+  }
+  return dir;
+}
+function readViewerSingleTemplate() {
+  const file = join5(findBundledAssetsDir(), "viewer-single.html");
+  if (!existsSync3(file)) {
+    throw new Error(
+      "Single-file viewer template not found. Reinstall edsger, or run `npm run sync-viewer`."
+    );
+  }
+  return readFileSync2(file, "utf8");
+}
+function renderSingleFile(template, data) {
+  const json = JSON.stringify(data).replace(/<\//g, "<\\/");
+  const tag = `<script>window.__EDSGER_DATA__=${json}</script>`;
+  if (template.includes("</head>")) return template.replace("</head>", `${tag}</head>`);
+  return template.replace("<body>", `<body>${tag}`);
+}
+function readJson(path) {
+  try {
+    return JSON.parse(readFileSync2(path, "utf8"));
+  } catch {
+    return null;
+  }
+}
+function readJsonl(path) {
+  if (!existsSync3(path)) return [];
+  const out = [];
+  for (const line of readFileSync2(path, "utf8").split("\n")) {
+    const t = line.trim();
+    if (!t) continue;
+    try {
+      out.push(JSON.parse(t));
+    } catch {
+    }
+  }
+  return out;
+}
+
+// src/core/open.ts
+import { spawn } from "child_process";
+function openInBrowser(target) {
+  const platform = process.platform;
+  const cmd = platform === "darwin" ? "open" : platform === "win32" ? "cmd" : "xdg-open";
+  const args = platform === "win32" ? ["/c", "start", "", target] : [target];
+  try {
+    spawn(cmd, args, { stdio: "ignore", detached: true }).unref();
+  } catch {
+  }
+}
+
+// src/commands/serve.ts
+var MIME = {
+  ".html": "text/html; charset=utf-8",
+  ".js": "text/javascript; charset=utf-8",
+  ".css": "text/css; charset=utf-8",
+  ".json": "application/json; charset=utf-8",
+  ".svg": "image/svg+xml",
+  ".map": "application/json",
+  ".ico": "image/x-icon",
+  ".woff2": "font/woff2",
+  ".woff": "font/woff"
+};
+async function runServe(opts) {
+  const benchmarkDir = join6(opts.cwd, opts.dir);
+  const viewerDir = findViewerDir();
+  const server = createServer((req, res) => {
+    const url = (req.url ?? "/").split("?")[0] ?? "/";
+    if (url === "/edsger-data.json") {
+      res.writeHead(200, {
+        "content-type": "application/json; charset=utf-8",
+        "cache-control": "no-store"
+      });
+      res.end(JSON.stringify(assembleData(benchmarkDir)));
+      return;
+    }
+    const rel = url === "/" ? "index.html" : decodeURIComponent(url.replace(/^\/+/, ""));
+    const safe = normalize(rel).replace(/^(\.\.(?:[/\\]|$))+/, "");
+    let file = join6(viewerDir, safe);
+    if (!file.startsWith(viewerDir)) {
+      res.writeHead(403);
+      res.end("forbidden");
+      return;
+    }
+    if (!existsSync4(file) || statSync2(file).isDirectory()) file = join6(viewerDir, "index.html");
+    res.writeHead(200, { "content-type": MIME[extname(file)] ?? "application/octet-stream" });
+    createReadStream(file).pipe(res);
+  });
+  return new Promise((resolve3, reject) => {
+    server.on("error", reject);
+    server.listen(opts.port, opts.host, () => {
+      const addr = server.address();
+      const port = typeof addr === "object" && addr ? addr.port : opts.port;
+      const shownHost = opts.host === "0.0.0.0" ? "localhost" : opts.host;
+      const url = `http://${shownHost}:${port}/`;
+      if (!hasData(assembleData(benchmarkDir))) {
+        log.warn(`No benchmark data in ${opts.dir} yet \u2014 run \`edsger benchmark\` first.`);
+      }
+      log.success(`Dashboard at ${pc6.bold(pc6.cyan(url))}  ${pc6.dim("(Ctrl-C to stop)")}`);
+      if (opts.open) openInBrowser(url);
+    });
+    const shutdown = () => {
+      server.close(() => resolve3(0));
+    };
+    process.on("SIGINT", shutdown);
+    process.on("SIGTERM", shutdown);
+  });
+}
+
+// src/commands/report.ts
+import { mkdirSync as mkdirSync2, writeFileSync as writeFileSync3 } from "fs";
+import { dirname as dirname2, isAbsolute as isAbsolute2, join as join7 } from "path";
+import pc7 from "picocolors";
+async function runReport(opts) {
+  const benchmarkDir = join7(opts.cwd, opts.dir);
+  const data = assembleData(benchmarkDir);
+  if (!hasData(data)) {
+    log.error(`No benchmark data in ${opts.dir}. Run \`edsger benchmark\` first.`);
+    return 1;
+  }
+  const html = renderSingleFile(readViewerSingleTemplate(), data);
+  const out = isAbsolute2(opts.html) ? opts.html : join7(opts.cwd, opts.html);
+  mkdirSync2(dirname2(out), { recursive: true });
+  writeFileSync3(out, html, "utf8");
+  log.success(
+    `Wrote self-contained report to ${pc7.bold(opts.html)} ${pc7.dim(`(${Math.round(html.length / 1024)} kB, open in any browser)`)}`
+  );
+  if (opts.open) openInBrowser(out);
+  return 0;
+}
+
 // src/commands/skills.ts
 import {
   cpSync,
-  existsSync as existsSync3,
-  mkdirSync as mkdirSync2,
-  readFileSync as readFileSync2,
-  readdirSync as readdirSync2,
+  existsSync as existsSync5,
+  mkdirSync as mkdirSync3,
+  readFileSync as readFileSync3,
+  readdirSync as readdirSync3,
   rmSync as rmSync2,
-  statSync as statSync2,
-  writeFileSync as writeFileSync3
+  statSync as statSync3,
+  writeFileSync as writeFileSync4
 } from "fs";
 import { homedir as homedir2 } from "os";
-import { join as join5 } from "path";
-import pc6 from "picocolors";
+import { join as join8 } from "path";
+import pc8 from "picocolors";
 var MANAGED = ["benchmark", "pr-review", "pr-resolve"];
 function targetSkillsDir(scope) {
-  return scope.global ? join5(homedir2(), ".claude", "skills") : join5(scope.cwd, ".claude", "skills");
+  return scope.global ? join8(homedir2(), ".claude", "skills") : join8(scope.cwd, ".claude", "skills");
 }
 function sourceSkillsDir() {
-  return join5(findBundledAssetsDir(), "skills");
+  return join8(findBundledAssetsDir(), "skills");
 }
 async function runSkillsInstall(opts) {
   const assetsDir = findBundledAssetsDir();
   const src = sourceSkillsDir();
-  if (!existsSync3(src)) {
+  if (!existsSync5(src)) {
     log.error(`Bundled skills not found at ${src}. Reinstall edsger.`);
     return 1;
   }
   const dest = targetSkillsDir(opts);
-  mkdirSync2(dest, { recursive: true });
-  const available = readdirSync2(src).filter((n) => existsSync3(join5(src, n, "SKILL.md")));
+  mkdirSync3(dest, { recursive: true });
+  const available = readdirSync3(src).filter((n) => existsSync5(join8(src, n, "SKILL.md")));
   let installed = 0;
   for (const name of available) {
-    const outDir = join5(dest, name);
-    if (existsSync3(outDir) && !opts.force) {
+    const outDir = join8(dest, name);
+    if (existsSync5(outDir) && !opts.force) {
       log.warn(`skill "${name}" already exists \u2014 use --force to overwrite`);
       continue;
     }
     rmSync2(outDir, { recursive: true, force: true });
-    cpSync(join5(src, name), outDir, { recursive: true });
-    const skillFile = join5(outDir, "SKILL.md");
-    const rewritten = readFileSync2(skillFile, "utf8").split("${CLAUDE_PLUGIN_ROOT}").join(assetsDir);
-    writeFileSync3(skillFile, rewritten, "utf8");
+    cpSync(join8(src, name), outDir, { recursive: true });
+    const skillFile = join8(outDir, "SKILL.md");
+    const rewritten = readFileSync3(skillFile, "utf8").split("${CLAUDE_PLUGIN_ROOT}").join(assetsDir);
+    writeFileSync4(skillFile, rewritten, "utf8");
     installed++;
     log.success(`installed /${name}`);
   }
@@ -1260,7 +1474,7 @@ async function runSkillsInstall(opts) {
     log.info("Nothing installed.");
     return 0;
   }
-  log.heading(`Installed ${installed} skill(s) into ${pc6.bold(dest)}`);
+  log.heading(`Installed ${installed} skill(s) into ${pc8.bold(dest)}`);
   log.step(`scope: ${opts.global ? "user (all projects)" : "this project"}`);
   log.step(`standards: ${assetsDir}`);
   log.step("Open Claude Code and run /benchmark, /pr-review, or /pr-resolve.");
@@ -1270,8 +1484,8 @@ async function runSkillsUninstall(opts) {
   const dest = targetSkillsDir(opts);
   let removed = 0;
   for (const name of MANAGED) {
-    const dir = join5(dest, name);
-    if (existsSync3(dir)) {
+    const dir = join8(dest, name);
+    if (existsSync5(dir)) {
       rmSync2(dir, { recursive: true, force: true });
       removed++;
       log.success(`removed /${name}`);
@@ -1287,20 +1501,20 @@ async function runSkillsList(opts) {
   ]) {
     const dir = targetSkillsDir(scope);
     const present = MANAGED.filter((n) => {
-      const p = join5(dir, n, "SKILL.md");
-      return existsSync3(p) && statSync2(p).isFile();
+      const p = join8(dir, n, "SKILL.md");
+      return existsSync5(p) && statSync3(p).isFile();
     });
     log.heading(`${label}: ${dir}`);
-    if (present.length) for (const n of present) process.stderr.write(`  ${pc6.green("\u2713")} /${n}
+    if (present.length) for (const n of present) process.stderr.write(`  ${pc8.green("\u2713")} /${n}
 `);
-    else process.stderr.write(`  ${pc6.dim("(none installed)")}
+    else process.stderr.write(`  ${pc8.dim("(none installed)")}
 `);
   }
   return 0;
 }
 
 // src/index.ts
-var pkg = JSON.parse(readFileSync3(new URL("../package.json", import.meta.url), "utf8"));
+var pkg = JSON.parse(readFileSync4(new URL("../package.json", import.meta.url), "utf8"));
 var VERSION = pkg.version;
 var DEFAULT_MODEL = "claude-sonnet-4-6";
 function toGlobal(raw) {
@@ -1338,10 +1552,38 @@ program.name("edsger").description(
   "Agentic CLI to benchmark repositories and review/resolve pull requests\nagainst industrial-grade standards (from edsger-assets, fully overridable)."
 ).version(VERSION, "-V, --version");
 addGlobalOptions(
-  program.command("benchmark").description("Score a repository against a benchmark rubric and write a report").option("--out <dir>", "report output directory (relative to repo)", ".edsger/benchmark").option("--no-write", "do not write report files into the repo")
+  program.command("benchmark").description("Score a repository against a benchmark rubric and write a report").option("--out <dir>", "report output directory (relative to repo)", ".edsger/benchmark").option("--no-write", "do not write report files into the repo").option("--serve", "open the dashboard after benchmarking")
 ).action(async (opts) => {
+  await guard(async () => {
+    const g = toGlobal(opts);
+    const code = await runBenchmark({ ...g, out: opts.out, noWrite: opts.write === false });
+    if (code === 0 && opts.serve) {
+      return runServe({ cwd: g.cwd, dir: opts.out, port: 4577, host: "127.0.0.1", open: true });
+    }
+    return code;
+  });
+});
+program.command("serve").description("Serve the benchmark dashboard (current, history, trends) locally").option("-C, --cwd <dir>", "repository directory").option("--dir <dir>", "benchmark output directory (relative to repo)", ".edsger/benchmark").option("--port <n>", "port to listen on", "4577").option("--host <host>", "host to bind", "127.0.0.1").option("--no-open", "do not open the browser automatically").action(
+  async (opts) => {
+    await guard(
+      () => runServe({
+        cwd: resolve2(opts.cwd ?? process.cwd()),
+        dir: opts.dir,
+        port: Number.parseInt(opts.port, 10) || 4577,
+        host: opts.host,
+        open: opts.open !== false
+      })
+    );
+  }
+);
+program.command("report").description("Export a self-contained HTML dashboard (data inlined) for sharing").option("-C, --cwd <dir>", "repository directory").option("--dir <dir>", "benchmark output directory (relative to repo)", ".edsger/benchmark").option("--html <path>", "output HTML file path", "edsger-report.html").option("--open", "open the exported file after writing").action(async (opts) => {
   await guard(
-    () => runBenchmark({ ...toGlobal(opts), out: opts.out, noWrite: opts.write === false })
+    () => runReport({
+      cwd: resolve2(opts.cwd ?? process.cwd()),
+      dir: opts.dir,
+      html: opts.html,
+      open: Boolean(opts.open)
+    })
   );
 });
 addGlobalOptions(
@@ -1392,11 +1634,11 @@ program.addHelpText(
   "after",
   `
 Examples:
-  ${pc7.dim("$")} edsger benchmark
-  ${pc7.dim("$")} edsger benchmark -C ./my-repo --json
-  ${pc7.dim("$")} edsger pr-review 123 --post
-  ${pc7.dim("$")} edsger pr-resolve 123 --instruction "fix the failing test" --push
-  ${pc7.dim("$")} edsger skills install   ${pc7.dim("# enable /benchmark etc. in Claude Code")}
+  ${pc9.dim("$")} edsger benchmark
+  ${pc9.dim("$")} edsger benchmark -C ./my-repo --json
+  ${pc9.dim("$")} edsger pr-review 123 --post
+  ${pc9.dim("$")} edsger pr-resolve 123 --instruction "fix the failing test" --push
+  ${pc9.dim("$")} edsger skills install   ${pc9.dim("# enable /benchmark etc. in Claude Code")}
 
 Auth: the agent uses your Claude Code login (run \`claude\` once to sign in); run \`gh auth login\` for PR commands.
 `