@maycombhq/atticus 0.0.0-stage → 2.0.0-beta.11

package/dist/chunk-3WU7LVTL.js

@@ -0,0 +1,203 @@
+import { createRequire as __atticusCreateRequire } from "node:module";
+import { fileURLToPath as __atticusFileURLToPath } from "node:url";
+var require = __atticusCreateRequire(import.meta.url);
+var __atticusUrl = (p) => new URL("./" + p, import.meta.url).href;
+var __atticusPath = (p) => __atticusFileURLToPath(__atticusUrl(p));
+import {
+  buildSite,
+  catalogRecords,
+  openChanges
+} from "./chunk-HR4DVO6X.js";
+import {
+  doctorChecks
+} from "./chunk-WCLAOKCH.js";
+import {
+  readProbes,
+  whatRunsWhere
+} from "./chunk-6RGIM2ZH.js";
+import {
+  baseClonePath,
+  loadRepos
+} from "./chunk-XZJE6WIT.js";
+import {
+  CatalogBaseline,
+  DEFAULT_RUNTIME,
+  WORKSPACE_SCHEMA_VERSION,
+  catalogDirOf,
+  inputsDirOf,
+  loadOrgPack,
+  orgPackDirOf,
+  require_dist
+} from "./chunk-JMOLB347.js";
+import {
+  __toESM
+} from "./chunk-TV45EUGM.js";
+
+// packages/cli/dist/baseline-commands.js
+var import_yaml = __toESM(require_dist(), 1);
+import { readdir, readFile, stat, writeFile } from "node:fs/promises";
+import { join, resolve } from "node:path";
+var BASELINE_FILE = "baseline.yaml";
+async function readBaseline(root) {
+  try {
+    const r = CatalogBaseline.safeParse((0, import_yaml.parse)(await readFile(join(catalogDirOf(root), BASELINE_FILE), "utf8")));
+    return r.success ? r.data : void 0;
+  } catch {
+    return void 0;
+  }
+}
+async function countFiles(dir) {
+  try {
+    const all = await readdir(dir, { recursive: true, withFileTypes: true });
+    return all.filter((e) => e.isFile()).length;
+  } catch {
+    return 0;
+  }
+}
+async function setupSteps(root, o = {}) {
+  const clonedOf = async (repos2) => (await Promise.all(repos2.map((r) => stat(join(baseClonePath(root, r.id), ".git")).then(() => true, () => false)))).filter(Boolean).length;
+  const [doctor, op, repos, records, inputs, baseline, probes] = await Promise.all([
+    o.doctor ?? doctorChecks(root),
+    loadOrgPack(orgPackDirOf(root)),
+    loadRepos(root).catch(() => []),
+    catalogRecords(root),
+    countFiles(inputsDirOf(root)),
+    readBaseline(root),
+    readProbes(root).then((p) => p.results)
+  ]);
+  const failing = doctor.filter((c) => c.status === "fail");
+  const cloned = await clonedOf(repos);
+  const generated = records.filter((r) => r.evidence.some((e) => e.source !== "manual")).length;
+  const wikis = op.sources.wikis.length;
+  const steps = [
+    {
+      id: "machine",
+      label: "This machine has what Atticus needs",
+      done: failing.length === 0,
+      detail: failing.length ? `to fix: ${failing.map((c) => c.label).join(", ")}` : `${doctor.filter((c) => c.status === "warn").length} to look at`,
+      fix: failing.length ? `Run atticus doctor: ${failing.map((c) => c.fix ?? c.label).join(" ")}` : "Run atticus doctor for the details."
+    },
+    {
+      id: "org-pack",
+      label: "The org pack's files are valid",
+      done: op.problems.length === 0,
+      detail: op.problems.length ? `${op.problems.length} problem${op.problems.length === 1 ? "" : "s"}` : "no problems in its files",
+      fix: "Run atticus orgpack validate and fix each problem it names (each says the file and field)."
+    },
+    {
+      // A new metarepo's org pack is valid as atticus init wrote it, but nobody owns it yet.
+      id: "org-pack-owner",
+      label: "The org pack has an owner",
+      done: !!op.pack?.owner && op.pack.owner !== "unassigned",
+      detail: op.pack?.owner && op.pack.owner !== "unassigned" ? op.pack.owner : "unassigned",
+      fix: "Set owner in org-pack/pack.yaml to the person who looks after the org pack (normally the platform lead), and commit it through a pull request."
+    },
+    {
+      // ADR 0121 item 6: how AI runs here is a choice, not the default left by accident.
+      id: "ai",
+      label: "Choose how AI runs here",
+      // runtime.yaml's version alone (1 from an older init, ADR 0121 addendum) isn't a choice.
+      done: JSON.stringify({ ...op.runtime, schemaVersion: DEFAULT_RUNTIME.schemaVersion }) !== JSON.stringify(DEFAULT_RUNTIME),
+      detail: whatRunsWhere(op.runtime, { probes }),
+      fix: "Choose the assistants, the modes and the model provider on Set up \u203A AI in the app (or in org-pack/runtime.yaml), and commit it through a pull request. As atticus init wrote it, steps are handed to Copilot, which needs nothing more.",
+      optional: true
+    },
+    {
+      id: "repos",
+      label: "Repos are listed",
+      done: repos.length > 0,
+      detail: `${repos.length} repo${repos.length === 1 ? "" : "s"} in the manifests`,
+      fix: "Run atticus repos discover <folder> (or --org <org>) and add what it proposes to repos.manifest.yaml."
+    },
+    {
+      id: "clones",
+      label: "Repos are cloned here",
+      done: repos.length > 0 && cloned === repos.length,
+      detail: `${cloned} of ${repos.length} cloned`,
+      fix: "Run atticus workspace sync (or Sync repos on the Scans screen); it clones what's missing."
+    },
+    {
+      // ADR 0124: your assistant reads each repo and writes its facts; Atticus checks every one
+      // against its line and puts them in the catalog. The id stays: the app keys on it.
+      id: "scan",
+      label: "Every repo is read and verified",
+      done: generated > 0,
+      detail: `${records.length} record${records.length === 1 ? "" : "s"} in the catalog`,
+      fix: "Read each repo: /atticus-map <repo> in your assistant (or atticus map <repo>), then atticus verify <repo> --write. Verify checks every fact against the line it cites and runs no model."
+    },
+    {
+      id: "knowledge",
+      label: "Existing knowledge is added",
+      done: inputs > 0 || wikis > 0,
+      detail: `${inputs} file${inputs === 1 ? "" : "s"} in inputs/, ${wikis} wiki source${wikis === 1 ? "" : "s"}`,
+      fix: "Drop diagrams, partner specs and exports into inputs/ (atticus intake add <files>), and add wikis to sources.yaml (atticus sources add).",
+      optional: true
+    },
+    {
+      id: "baseline",
+      label: "The catalog is accepted",
+      done: !!baseline,
+      detail: baseline ? `accepted by ${baseline.acceptedBy} on ${baseline.acceptedAt.slice(0, 10)}` : "not yet",
+      fix: "Walk the ecosystem map with the architect, then accept it: atticus baseline (or Accept the catalog on Home), and commit catalog/baseline.yaml through a pull request."
+    }
+  ];
+  return { steps, baselined: !!baseline };
+}
+async function setupCommand(root, o, io) {
+  const s = await setupSteps(root);
+  const open = s.steps.filter((x) => !x.done && !x.optional);
+  if (o.json)
+    io.out(JSON.stringify(s, null, 2));
+  else {
+    for (const x of s.steps) {
+      io.out(`${x.done ? "[x]" : "[ ]"} ${x.label}: ${x.detail}${x.optional ? " (optional)" : ""}`);
+      if (!x.done && x.fix)
+        io.out(`    ${x.fix}`);
+    }
+    io.out(open.length ? `${open.length} step${open.length === 1 ? "" : "s"} to go.` : "Setup is done: the catalog is accepted.");
+  }
+  return open.length ? 1 : 0;
+}
+async function baselineCommand(root, o, io) {
+  if (!o.by) {
+    io.err("Usage: atticus baseline [--by <you>] [--note <text>] [--site <dir>]");
+    return 2;
+  }
+  const records = await catalogRecords(root, io);
+  if (records.length === 0) {
+    io.err("The catalog is empty: read the repos first (atticus map <repo>, then atticus verify <repo> --write), then walk the map and accept it.");
+    return 1;
+  }
+  const kinds = {};
+  for (const r of records)
+    kinds[r.kind] = (kinds[r.kind] ?? 0) + 1;
+  const systems = records.filter((r) => r.kind === "System").map((r) => r.id).sort();
+  const b = CatalogBaseline.parse({
+    schemaVersion: WORKSPACE_SCHEMA_VERSION,
+    acceptedBy: o.by,
+    acceptedAt: o.now ?? (/* @__PURE__ */ new Date()).toISOString().replace(/\.\d{3}Z$/, "Z"),
+    kinds: Object.fromEntries(Object.entries(kinds).sort(([a], [b2]) => a.localeCompare(b2))),
+    systems,
+    ...o.note ? { note: o.note } : {}
+  });
+  await writeFile(join(catalogDirOf(root), BASELINE_FILE), (0, import_yaml.stringify)(b), "utf8");
+  io.out(`Accepted the catalog as the baseline as ${o.by}: ${records.length} records in ${Object.keys(kinds).length} kinds, ${systems.length} systems. Wrote catalog/${BASELINE_FILE}; commit it with the catalog and open a pull request, whose merge makes it the baseline on main.`);
+  if (o.site) {
+    const dir = resolve(o.site);
+    const s = await buildSite(records, dir, {
+      title: "Catalog",
+      builtAt: b.acceptedAt.slice(0, 10),
+      changes: await openChanges(root)
+    });
+    io.out(`Wrote the read-only catalog site to ${dir} (${s.systems.length} systems).`);
+  }
+  return 0;
+}
+
+export {
+  BASELINE_FILE,
+  readBaseline,
+  setupSteps,
+  setupCommand,
+  baselineCommand
+};

package/dist/chunk-4AYZQ3SY.js

@@ -0,0 +1,285 @@
+import { createRequire as __atticusCreateRequire } from "node:module";
+import { fileURLToPath as __atticusFileURLToPath } from "node:url";
+var require = __atticusCreateRequire(import.meta.url);
+var __atticusUrl = (p) => new URL("./" + p, import.meta.url).href;
+var __atticusPath = (p) => __atticusFileURLToPath(__atticusUrl(p));
+import {
+  compareReports,
+  findRepoRoot,
+  listReports,
+  loadDatasets,
+  readQuality,
+  runEvals,
+  saveReport,
+  writeQuality
+} from "./chunk-NELU7LSL.js";
+import {
+  table
+} from "./chunk-VOGCSYOI.js";
+import {
+  readDocument
+} from "./chunk-5V5BJACL.js";
+import {
+  combine,
+  createRuntime,
+  formatOf,
+  screenAbsorbed,
+  screenDiagramXml,
+  screenOfficeXml,
+  screenText
+} from "./chunk-BKFGGJKA.js";
+import {
+  loadOrgPack,
+  orgPackDirOf,
+  readZip,
+  stateDirOf
+} from "./chunk-JMOLB347.js";
+
+// packages/cli/dist/eval-commands.js
+import { readdir, readFile, stat } from "node:fs/promises";
+import { isAbsolute, join, relative, sep } from "node:path";
+var MODES = ["mock", "replay", "live"];
+async function evalRun(root, o, io, deps = {}) {
+  const mode = o.mode ?? "mock";
+  if (!MODES.includes(mode)) {
+    io.err(`--mode is one of ${MODES.join(", ")}.`);
+    return 2;
+  }
+  if (o.runs !== void 0 && (!Number.isInteger(o.runs) || o.runs < 1)) {
+    io.err("--runs is a whole number of at least 1.");
+    return 2;
+  }
+  const env = deps.env ?? process.env;
+  let liveRuntime = deps.liveRuntime;
+  if (mode === "live" && !liveRuntime) {
+    if (env.CI) {
+      io.err("Live evals call a real model, so they never run in CI. Use --mode replay there.");
+      return 2;
+    }
+    const op = await loadOrgPack(orgPackDirOf(root));
+    try {
+      const rt = await createRuntime(op, {
+        mode: "model",
+        root,
+        ...deps.providers ? { providers: deps.providers } : {}
+      });
+      liveRuntime = () => rt;
+    } catch (e) {
+      io.err(`${e.message} Live evals need a model provider in runtime.yaml (adapters.model).`);
+      return 2;
+    }
+  }
+  const repoRoot = deps.repoRoot ?? findRepoRoot();
+  const ds = await loadDatasets({
+    ...repoRoot ? { repoRoot } : {},
+    orgPackDir: orgPackDirOf(root)
+  });
+  for (const p of ds.problems)
+    io.err(p);
+  for (const u of ds.unavailable)
+    io.err(`Left out: ${u}.`);
+  const agents = o.agent === void 0 ? void 0 : Array.isArray(o.agent) ? o.agent : [o.agent];
+  const known = new Set(ds.cases.map((c) => c.agent));
+  const unknown = (agents ?? []).filter((a) => !known.has(a));
+  if (unknown.length) {
+    io.err(`No eval cases for ${unknown.join(", ")}. Agents with cases: ${[...known].sort().join(", ")}.`);
+    return 2;
+  }
+  const previous = (await listReports(root)).filter((r) => r.mode === mode).at(-1);
+  const report = await runEvals({
+    cases: ds.cases,
+    mode,
+    ...o.runs ? { runs: o.runs } : {},
+    ...agents ? { agents } : {},
+    ...liveRuntime ? { liveRuntime } : {},
+    repoRoot: repoRoot ?? root,
+    ...deps.now ? { now: deps.now } : {}
+  });
+  const path = await saveReport(root, report);
+  await writeQuality(root, (deps.now ?? (() => /* @__PURE__ */ new Date()))());
+  const cmp = previous ? compareReports(previous, report) : void 0;
+  const failed = report.results.filter((r) => !r.pass);
+  if (o.json) {
+    io.out(JSON.stringify({ report, comparison: cmp ?? null, file: relative(root, path).split(sep).join("/") }, null, 2));
+    return failed.length || report.liveBar && !report.liveBar.pass ? 1 : 0;
+  }
+  io.out(`Eval run ${report.runId} (${mode}, ${report.runs} run${report.runs === 1 ? "" : "s"} per case):`);
+  for (const l of table(["AGENT", "CASES", "PASSED", "RUNS", "PASS RATE"], report.agents.map((a) => [
+    a.agent,
+    String(a.cases),
+    String(a.passed),
+    String(a.runs),
+    `${Math.round(a.passRate * 100)}%`
+  ])))
+    io.out(l);
+  for (const f of failed)
+    io.out(`FAIL ${f.agent}/${f.case} run ${f.run}: ${f.error ?? f.grades.filter((g) => !g.pass).map((g) => `${g.grader}${g.notes.length ? ` (${g.notes.slice(0, 3).join("; ")})` : ""}`).join(", ")}`);
+  for (const s of report.skipped)
+    io.out(`Skipped ${s.agent}/${s.case}: ${s.reason}.`);
+  if (report.liveBar)
+    io.out(`Live bar (${report.liveBar.min} of ${report.liveBar.runs} runs): ${report.liveBar.pass ? "met" : "not met"}${report.liveBar.pass ? "" : ` (${report.liveBar.cases.filter((c) => !c.ok).map((c) => `${c.agent}/${c.case} ${c.passed} of ${c.runs}`).join(", ")})`}.`);
+  if (cmp?.regressed.length)
+    io.out(`Regressed since ${previous.runId}: ${cmp.regressed.join(", ")}.`);
+  if (cmp?.fixed.length)
+    io.out(`Fixed since ${previous.runId}: ${cmp.fixed.join(", ")}.`);
+  io.out(`Saved ${relative(root, path).split(sep).join("/")}.`);
+  return failed.length || report.liveBar && !report.liveBar.pass ? 1 : 0;
+}
+async function evalReport(root, o, io) {
+  const q = await readQuality(root);
+  if (!q) {
+    io.out("No eval runs yet. Run atticus eval run (mock or replay), or --mode live with a configured model.");
+    return 0;
+  }
+  if (o.json) {
+    io.out(JSON.stringify(q, null, 2));
+    return q.agents.some((a) => a.regressed.length || a.liveBar?.pass === false) ? 1 : 0;
+  }
+  for (const l of table(["AGENT", "LAST RUN", "MODE", "CASES", "PASS RATE", "LIVE BAR", "REGRESSED"], q.agents.map((a) => [
+    a.agent,
+    a.lastRunAt,
+    a.mode,
+    String(a.cases),
+    `${Math.round(a.passRate * 100)}%`,
+    a.liveBar ? `${a.liveBar.pass ? "met" : "not met"} (${a.liveBar.min}/${a.liveBar.runs})` : "no live run",
+    a.regressed.join(", ") || "-"
+  ])))
+    io.out(l);
+  return q.agents.some((a) => a.regressed.length || a.liveBar?.pass === false) ? 1 : 0;
+}
+var TEXT = /\.(md|markdown|txt|text|html?|ya?ml|json|csv|java|kt|ts|tsx|js|jsx|mjs|py|go|cs|rb|sql|sh|properties|xml|feature|mmd|gherkin)$/i;
+var INSTRUCTION_FILE = /(^|\/)(\.github\/(prompts|agents|chatmodes|instructions)\/|\.claude\/|\.cursor\/|prompts\/)|\.(prompt|agent|chatmode|instructions)\.md$|(^|\/)(AGENTS|CLAUDE|copilot-instructions)\.md$/i;
+var ACTION = {
+  high: "quarantined: left out of every AI step until a person cleans it",
+  medium: "kept as data, suspicious lines marked",
+  low: "kept as data, invisible characters removed",
+  none: "kept as data"
+};
+async function screenFile(full, rel) {
+  const bytes = new Uint8Array(await readFile(full));
+  if (/\.(docx|pptx|xlsx)$/i.test(rel)) {
+    const parts = [];
+    try {
+      const zip = readZip(bytes);
+      for (const n of zip.names.filter((x) => /^(word\/document|word\/(header|footer)\d*|ppt\/slides\/slide\d+|ppt\/notesSlides\/notesSlide\d+)\.xml$/.test(x)))
+        parts.push(screenOfficeXml(zip.text(n) ?? ""));
+    } catch {
+    }
+    const doc = readDocument(rel, bytes);
+    parts.push(screenText(doc.text.sections.map((s) => `${s.heading}
+${s.text}`).join("\n"), {
+      format: "office"
+    }));
+    return combine(...parts);
+  }
+  if (/\.pdf$/i.test(rel)) {
+    const doc = readDocument(rel, bytes);
+    return screenText(doc.text.sections.map((s) => `${s.heading}
+${s.text}`).join("\n"), {
+      format: "pdf"
+    });
+  }
+  if (/\.drawio$/i.test(rel) || /\.xml$/i.test(rel) && new TextDecoder().decode(bytes.slice(0, 200)).includes("<mxfile"))
+    return screenDiagramXml(new TextDecoder().decode(bytes));
+  if (!TEXT.test(rel) || bytes.includes(0))
+    return void 0;
+  const text = new TextDecoder().decode(bytes);
+  if (INSTRUCTION_FILE.test(rel)) {
+    const f = screenAbsorbed([{ id: rel, path: rel, text }])[0];
+    return {
+      risk: f.risk,
+      score: 0,
+      signals: f.signals,
+      rules: [...new Set(f.signals.map((s) => s.rule))],
+      reason: f.reason
+    };
+  }
+  return screenText(text, { format: formatOf("document", rel) });
+}
+async function defencesScan(root, path, o, io) {
+  const targets = [];
+  if (path)
+    targets.push(isAbsolute(path) ? path : join(root, path));
+  else {
+    targets.push(join(stateDirOf(root), "inputs"));
+    try {
+      const op = await loadOrgPack(orgPackDirOf(root));
+      for (const w of op.sources.wikis)
+        if (w.type === "markdown-folder")
+          targets.push(join(orgPackDirOf(root), w.location));
+    } catch {
+    }
+  }
+  const files = [];
+  for (const t of targets) {
+    let info;
+    try {
+      info = await stat(t);
+    } catch {
+      if (path) {
+        io.err(`${path} doesn't exist.`);
+        return 2;
+      }
+      continue;
+    }
+    const walk = async (d) => {
+      for (const e of (await readdir(d, { withFileTypes: true })).sort((a, b) => a.name < b.name ? -1 : 1)) {
+        if (e.name === ".git" || e.name === "node_modules")
+          continue;
+        const full = join(d, e.name);
+        if (e.isDirectory())
+          await walk(full);
+        else
+          files.push({ full, rel: relative(root, full).split(sep).join("/") });
+      }
+    };
+    if (info.isDirectory())
+      await walk(t);
+    else
+      files.push({ full: t, rel: relative(root, t).split(sep).join("/") });
+  }
+  const out = [];
+  let skipped = 0;
+  for (const f of files) {
+    const s = await screenFile(f.full, f.rel).catch(() => void 0);
+    if (!s) {
+      skipped++;
+      continue;
+    }
+    out.push({
+      path: f.rel,
+      risk: s.risk,
+      action: ACTION[s.risk],
+      reason: s.reason,
+      rules: s.rules,
+      signals: s.signals.map((x) => ({
+        rule: x.rule,
+        line: x.line,
+        excerpt: x.excerpt,
+        ...x.hidden ? { hidden: true } : {},
+        ...x.encoded ? { encoded: true } : {},
+        ...x.obfuscated ? { obfuscated: true } : {}
+      }))
+    });
+  }
+  const risky = out.filter((f) => f.risk === "medium" || f.risk === "high");
+  if (o.json) {
+    io.out(JSON.stringify({ files: o.all ? out : out.filter((f) => f.risk !== "none"), scanned: out.length, skipped }, null, 2));
+    return risky.length ? 1 : 0;
+  }
+  const shown = o.all ? out : out.filter((f) => f.risk !== "none");
+  if (shown.length)
+    for (const l of table(["RISK", "FILE", "WHAT ATTICUS DOES", "WHY"], shown.map((f) => [f.risk, f.path, f.action, f.reason])))
+      io.out(l);
+  for (const f of risky)
+    for (const s of f.signals.slice(0, 5))
+      io.out(`  ${f.path}:${s.line} ${s.rule}${s.hidden ? " (hidden)" : s.encoded ? " (encoded)" : s.obfuscated ? " (disguised)" : ""}: ${s.excerpt}`);
+  io.out(`Scanned ${out.length} file${out.length === 1 ? "" : "s"}${skipped ? ` (${skipped} not text or documents)` : ""}: ${risky.filter((f) => f.risk === "high").length} high, ${risky.filter((f) => f.risk === "medium").length} medium.`);
+  return risky.length ? 1 : 0;
+}
+
+export {
+  evalRun,
+  evalReport,
+  defencesScan
+};