kairn-cli 2.2.10 → 2.3.0

package/dist/cli.js

@@ -1,10 +1,11 @@
 // src/cli.ts
 import { Command as Command12 } from "commander";
 import chalk15 from "chalk";
+import { createRequire } from "module";
 
 // src/commands/init.ts
 import { Command } from "commander";
-import { input, password, select } from "@inquirer/prompts";
+import { confirm, input, password, select } from "@inquirer/prompts";
 import chalk3 from "chalk";
 import Anthropic from "@anthropic-ai/sdk";
 import OpenAI from "openai";
@@ -62,6 +63,59 @@ async function saveConfig(config) {
   await fs.writeFile(CONFIG_PATH, JSON.stringify(config, null, 2), "utf-8");
 }
 
+// src/auth/keychain.ts
+import { exec } from "child_process";
+import { promisify } from "util";
+var execAsync = promisify(exec);
+var KEYCHAIN_SERVICE = "Claude Code-credentials";
+var TOKEN_EXPIRY_BUFFER_MS = 6e4;
+function parseKeychainCredentials(raw) {
+  let parsed;
+  try {
+    parsed = JSON.parse(raw);
+  } catch {
+    return null;
+  }
+  if (typeof parsed !== "object" || parsed === null) return null;
+  const obj = parsed;
+  const oauth = obj["claudeAiOauth"];
+  if (typeof oauth !== "object" || oauth === null) return null;
+  const oauthObj = oauth;
+  const accessToken = oauthObj["accessToken"];
+  const refreshToken = oauthObj["refreshToken"];
+  const expiresAt = oauthObj["expiresAt"];
+  const subscriptionType = oauthObj["subscriptionType"];
+  if (typeof accessToken !== "string" || !accessToken) return null;
+  if (typeof refreshToken !== "string") return null;
+  if (typeof expiresAt !== "number") return null;
+  return {
+    accessToken,
+    refreshToken,
+    expiresAt,
+    subscriptionType: typeof subscriptionType === "string" ? subscriptionType : "unknown"
+  };
+}
+function isTokenExpired(credentials) {
+  return Date.now() + TOKEN_EXPIRY_BUFFER_MS >= credentials.expiresAt;
+}
+async function readClaudeCodeCredentials(account) {
+  if (process.platform !== "darwin") return null;
+  try {
+    const acct = account ?? "";
+    const cmd = acct ? `security find-generic-password -s "${KEYCHAIN_SERVICE}" -a "${acct}" -w` : `security find-generic-password -s "${KEYCHAIN_SERVICE}" -w`;
+    const { stdout } = await execAsync(cmd, { timeout: 5e3 });
+    return parseKeychainCredentials(stdout.trim());
+  } catch {
+    return null;
+  }
+}
+async function getAccessToken(account) {
+  const creds = await readClaudeCodeCredentials(account);
+  if (!creds) return null;
+  if (isTokenExpired(creds)) return null;
+  return creds.accessToken;
+}
+
 // src/providers.ts
 var PROVIDER_CONFIGS = {
   anthropic: {
@@ -508,30 +562,49 @@ var initCommand = new Command("init").description("Set up Kairn with your API ke
       choices: PROVIDER_MODELS[provider]
     });
   }
-  const apiKey = await password({
-    message: `${providerDisplayName} API key${provider === "other" ? " (Enter to skip)" : ""}`,
-    mask: "*"
-  });
-  if (!apiKey && provider !== "other") {
-    console.log(ui.error("No API key provided. Aborting."));
-    process.exit(1);
+  let apiKey = "";
+  let authType = "api-key";
+  if (provider === "anthropic") {
+    const oauthToken = await getAccessToken();
+    if (oauthToken) {
+      const useOAuth = await confirm({
+        message: "Claude Code subscription detected. Use it instead of an API key? (experimental \u2014 may break)",
+        default: true
+      });
+      if (useOAuth) {
+        authType = "claude-code-oauth";
+        console.log(ui.warn("Using Claude Code OAuth token. This is undocumented and may break at any time."));
+        console.log(ui.success("OAuth token validated"));
+      }
+    }
   }
-  if (apiKey) {
-    console.log(chalk3.dim("\n  Verifying API key..."));
-    const valid = await verifyKey(provider, apiKey, baseURL, model);
-    if (!valid) {
-      console.log(ui.error("Invalid API key. Check your key and try again."));
+  if (authType === "api-key") {
+    apiKey = await password({
+      message: `${providerDisplayName} API key${provider === "other" ? " (Enter to skip)" : ""}`,
+      mask: "*"
+    });
+    if (!apiKey && provider !== "other") {
+      console.log(ui.error("No API key provided. Aborting."));
       process.exit(1);
     }
-    console.log(ui.success("API key verified"));
-  } else {
-    console.log(ui.warn("No API key \u2014 skipping verification"));
+    if (apiKey) {
+      console.log(chalk3.dim("\n  Verifying API key..."));
+      const valid = await verifyKey(provider, apiKey, baseURL, model);
+      if (!valid) {
+        console.log(ui.error("Invalid API key. Check your key and try again."));
+        process.exit(1);
+      }
+      console.log(ui.success("API key verified"));
+    } else {
+      console.log(ui.warn("No API key \u2014 skipping verification"));
+    }
   }
   const config = {
     provider,
-    api_key: apiKey || "",
+    api_key: apiKey,
     model,
     ...baseURL ? { base_url: baseURL } : {},
+    ...authType !== "api-key" ? { auth_type: authType } : {},
     default_runtime: "claude-code",
     created_at: (/* @__PURE__ */ new Date()).toISOString()
   };
@@ -555,7 +628,7 @@ var initCommand = new Command("init").description("Set up Kairn with your API ke
 
 // src/commands/describe.ts
 import { Command as Command2 } from "commander";
-import { input as input2, confirm, select as select2 } from "@inquirer/prompts";
+import { input as input2, confirm as confirm2, select as select2 } from "@inquirer/prompts";
 import chalk5 from "chalk";
 
 // src/compiler/compile.ts
@@ -1240,8 +1313,18 @@ async function callLLM(config, userMessage, options) {
   const { systemPrompt } = options;
   const jsonMode = options.jsonMode ?? false;
   const providerName = getProviderName(config.provider);
+  let apiKey = config.api_key;
+  if (config.auth_type === "claude-code-oauth") {
+    const oauthToken = await getAccessToken();
+    if (!oauthToken) {
+      throw new Error(
+        "Claude Code OAuth token unavailable or expired. Run `kairn init` to reconfigure, or launch Claude Code to refresh the token."
+      );
+    }
+    apiKey = oauthToken;
+  }
   if (config.provider === "anthropic") {
-    const client2 = new Anthropic2({ apiKey: config.api_key });
+    const client2 = new Anthropic2({ apiKey });
     const messages = [
       { role: "user", content: userMessage }
     ];
@@ -1262,7 +1345,7 @@ async function callLLM(config, userMessage, options) {
     }
   }
   const resolvedBaseURL = getBaseURL(config.provider, config.base_url);
-  const clientOptions = { apiKey: config.api_key };
+  const clientOptions = { apiKey };
   if (resolvedBaseURL) clientOptions.baseURL = resolvedBaseURL;
   const client = new OpenAI2(clientOptions);
   try {
@@ -2441,7 +2524,7 @@ Autonomy level: ${autonomyLevel} (${autonomyLabel(autonomyLevel)})`;
       console.log("");
     }
   }
-  const proceed = options.yes || await confirm({
+  const proceed = options.yes || await confirm2({
     message: "Generate environment in current directory?",
     default: true
   });
@@ -2656,7 +2739,7 @@ var updateRegistryCommand = new Command5("update-registry").description("Fetch t
 
 // src/commands/optimize.ts
 import { Command as Command6 } from "commander";
-import { confirm as confirm2 } from "@inquirer/prompts";
+import { confirm as confirm3 } from "@inquirer/prompts";
 import chalk9 from "chalk";
 import ora from "ora";
 import fs12 from "fs/promises";
@@ -2718,7 +2801,7 @@ function detectFramework(deps) {
   ];
   const detected = [];
   for (const [packages, name] of frameworks) {
-    if (packages.some((pkg) => deps.includes(pkg))) {
+    if (packages.some((pkg2) => deps.includes(pkg2))) {
       detected.push(name);
     }
   }
@@ -2742,11 +2825,11 @@ function extractEnvKeys(content) {
   return keys;
 }
 async function scanProject(dir) {
-  const pkg = await readJsonSafe(path11.join(dir, "package.json"));
-  const deps = pkg?.dependencies ? Object.keys(pkg.dependencies) : [];
-  const devDeps = pkg?.devDependencies ? Object.keys(pkg.devDependencies) : [];
+  const pkg2 = await readJsonSafe(path11.join(dir, "package.json"));
+  const deps = pkg2?.dependencies ? Object.keys(pkg2.dependencies) : [];
+  const devDeps = pkg2?.devDependencies ? Object.keys(pkg2.devDependencies) : [];
   const allDeps = [...deps, ...devDeps];
-  const scripts = pkg?.scripts || {};
+  const scripts = pkg2?.scripts || {};
   const rootFiles = await listDirSafe(dir);
   const keyFiles = rootFiles.filter(
     (f) => [
@@ -2808,8 +2891,8 @@ async function scanProject(dir) {
     existingSkills = await listDirSafe(path11.join(claudeDir, "skills"));
     existingAgents = (await listDirSafe(path11.join(claudeDir, "agents"))).filter((f) => f.endsWith(".md")).map((f) => f.replace(".md", ""));
   }
-  const name = pkg?.name || path11.basename(dir);
-  const description = pkg?.description || "";
+  const name = pkg2?.name || path11.basename(dir);
+  const description = pkg2?.description || "";
   return {
     name,
     description,
@@ -3028,7 +3111,7 @@ var optimizeCommand = new Command6("optimize").description("Scan an existing pro
     }
     if (!options.yes) {
       console.log("");
-      const proceed = await confirm2({
+      const proceed = await confirm3({
         message: "Generate optimized environment? This will overwrite existing .claude/ files.",
         default: false
       });
@@ -3040,7 +3123,7 @@ var optimizeCommand = new Command6("optimize").description("Scan an existing pro
   } else {
     console.log(chalk9.dim("\n  No existing .claude/ directory found \u2014 generating from scratch.\n"));
     if (!options.yes) {
-      const proceed = await confirm2({
+      const proceed = await confirm3({
         message: "Generate Claude Code environment for this project?",
         default: true
       });
@@ -3103,7 +3186,7 @@ var optimizeCommand = new Command6("optimize").description("Scan an existing pro
       }
     }
     console.log("");
-    const apply = await confirm2({
+    const apply = await confirm3({
       message: "Apply these changes?",
       default: true
     });
@@ -3702,7 +3785,7 @@ import ora2 from "ora";
 import fs24 from "fs/promises";
 import path24 from "path";
 import { parse as yamlParse2 } from "yaml";
-import { confirm as confirm3, select as select4 } from "@inquirer/prompts";
+import { confirm as confirm4, select as select4 } from "@inquirer/prompts";
 
 // src/evolve/init.ts
 import fs15 from "fs/promises";
@@ -3945,14 +4028,14 @@ async function buildProjectProfile(projectRoot) {
       path15.join(projectRoot, "package.json"),
       "utf-8"
     );
-    const pkg = JSON.parse(pkgStr);
+    const pkg2 = JSON.parse(pkgStr);
     profile.language = "typescript";
-    if (pkg.scripts && typeof pkg.scripts === "object") {
-      profile.scripts = pkg.scripts;
+    if (pkg2.scripts && typeof pkg2.scripts === "object") {
+      profile.scripts = pkg2.scripts;
     }
     const deps = {
-      ...pkg.dependencies ?? {},
-      ...pkg.devDependencies ?? {}
+      ...pkg2.dependencies ?? {},
+      ...pkg2.devDependencies ?? {}
     };
     if (deps.next) {
       profile.framework = "Next.js";
@@ -4051,8 +4134,8 @@ async function copyDir(src, dest) {
 }
 
 // src/evolve/runner.ts
-import { exec as exec2, spawn } from "child_process";
-import { promisify as promisify2 } from "util";
+import { exec as exec3, spawn } from "child_process";
+import { promisify as promisify3 } from "util";
 import fs18 from "fs/promises";
 import os3 from "os";
 import path18 from "path";
@@ -4177,11 +4260,11 @@ async function loadIterationLog(workspacePath, iteration) {
 }
 
 // src/evolve/exec.ts
-import { exec } from "child_process";
-import { promisify } from "util";
-var execAsync = promisify(exec);
+import { exec as exec2 } from "child_process";
+import { promisify as promisify2 } from "util";
+var execAsync2 = promisify2(exec2);
 async function execCommand(cmd, cwd, timeoutMs = 3e4) {
-  return execAsync(cmd, { cwd, timeout: timeoutMs });
+  return execAsync2(cmd, { cwd, timeout: timeoutMs });
 }
 
 // src/evolve/scorers.ts
@@ -4341,21 +4424,47 @@ async function rubricScorer(task, workspacePath, stdout, stderr, config) {
     breakdown
   };
 }
+function classifyFailure(score, stdout, stderr) {
+  if (score.pass) return score;
+  const combined = `${stdout}
+${stderr}`.toLowerCase();
+  const scoreValue = score.score ?? 0;
+  let failureCategory = "unknown";
+  let failureReason = "";
+  if (stderr.includes("[setup]") && stderr.includes("Error") || combined.includes("command not found") || combined.includes("no such file or directory")) {
+    failureCategory = "task";
+    failureReason = "Task setup failed or references missing resources";
+  } else if (combined.includes("token limit") || combined.includes("context length") || combined.includes("rate limit") || combined.includes("api error") || combined.includes("429") || combined.includes("overloaded")) {
+    failureCategory = "model";
+    failureReason = "Model API error, token limit, or rate limit";
+  } else if (combined.includes("build failed") && combined.includes("before") || combined.includes("merge conflict") || combined.includes("git dirty") || combined.includes("uncommitted changes")) {
+    failureCategory = "repo";
+    failureReason = "Pre-existing repo issues (build failure, dirty state)";
+  } else if (scoreValue >= 20 && scoreValue < 80) {
+    failureCategory = "harness";
+    failureReason = "Agent attempted the task but did not follow harness conventions";
+  }
+  return { ...score, failureCategory, failureReason };
+}
 async function scoreTask(task, workspacePath, stdout, stderr, config) {
+  let score;
   if (task.scoring === "pass-fail") {
-    return passFailScorer(task, workspacePath, stdout, stderr);
-  }
-  if (task.scoring === "llm-judge" && config) {
-    return llmJudgeScorer(task, workspacePath, stdout, stderr, config);
+    score = await passFailScorer(task, workspacePath, stdout, stderr);
+  } else if (task.scoring === "llm-judge" && config) {
+    score = await llmJudgeScorer(task, workspacePath, stdout, stderr, config);
+  } else if (task.scoring === "rubric" && config) {
+    score = await rubricScorer(task, workspacePath, stdout, stderr, config);
+  } else {
+    score = await passFailScorer(task, workspacePath, stdout, stderr);
   }
-  if (task.scoring === "rubric" && config) {
-    return rubricScorer(task, workspacePath, stdout, stderr, config);
+  if (!score.pass) {
+    score = classifyFailure(score, stdout, stderr);
   }
-  return passFailScorer(task, workspacePath, stdout, stderr);
+  return score;
 }
 
 // src/evolve/runner.ts
-var execAsync2 = promisify2(exec2);
+var execAsync3 = promisify3(exec3);
 var COPY_SKIP_DIRS = /* @__PURE__ */ new Set([".git", "node_modules", ".kairn-evolve", ".claude"]);
 async function deployMcpJson(harnessPath, workDir) {
   const src = path18.join(harnessPath, ".mcp.json");
@@ -4365,12 +4474,12 @@ async function deployMcpJson(harnessPath, workDir) {
 async function createIsolatedWorkspace(projectRoot, harnessPath) {
   const suffix = `${Date.now()}-${Math.random().toString(36).slice(2)}`;
   try {
-    await execAsync2("git rev-parse --is-inside-work-tree", {
+    await execAsync3("git rev-parse --is-inside-work-tree", {
       cwd: projectRoot,
       timeout: 5e3
     });
     const tmpDir2 = path18.join(os3.tmpdir(), `kairn-evolve-wt-${suffix}`);
-    await execAsync2(`git worktree add --detach "${tmpDir2}" HEAD`, {
+    await execAsync3(`git worktree add --detach "${tmpDir2}" HEAD`, {
       cwd: projectRoot,
       timeout: 3e4
     });
@@ -4409,14 +4518,14 @@ async function copyProjectDir(src, dest) {
 async function cleanupIsolatedWorkspace(workDir, isWorktree, projectRoot) {
   if (isWorktree) {
     try {
-      await execAsync2(`git worktree remove "${workDir}" --force`, {
+      await execAsync3(`git worktree remove "${workDir}" --force`, {
         cwd: projectRoot,
         timeout: 1e4
       });
     } catch {
       await fs18.rm(workDir, { recursive: true, force: true }).catch(() => {
       });
-      await execAsync2("git worktree prune", {
+      await execAsync3("git worktree prune", {
         cwd: projectRoot,
         timeout: 5e3
       }).catch(() => {
@@ -4437,7 +4546,7 @@ async function runTask(task, harnessPath, traceDir, iteration, projectRoot) {
     let setupStderr = "";
     if (task.setup.trim()) {
       try {
-        await execAsync2(task.setup, { cwd: workDir, timeout: 6e4 });
+        await execAsync3(task.setup, { cwd: workDir, timeout: 6e4 });
       } catch (err) {
         setupStderr = err instanceof Error ? err.message : String(err);
       }
@@ -5453,6 +5562,7 @@ function buildLeaderboard(iterations, tasks) {
   const taskIds = tasks.map((t) => t.id);
   return taskIds.map((taskId) => {
     const scores = {};
+    const variance = {};
     let bestScore = -1;
     let bestIteration = 0;
     for (const iter of iterations) {
@@ -5460,13 +5570,21 @@ function buildLeaderboard(iterations, tasks) {
       if (s) {
         const score = numericScore2(s);
         scores[iter.iteration] = score;
+        if (s.variance) {
+          variance[iter.iteration] = {
+            mean: s.variance.mean,
+            stddev: s.variance.stddev,
+            runs: s.variance.runs
+          };
+        }
         if (score > bestScore) {
           bestScore = score;
           bestIteration = iter.iteration;
         }
       }
     }
-    return { taskId, scores, bestIteration, bestScore };
+    const hasVariance = Object.keys(variance).length > 0;
+    return { taskId, scores, bestIteration, bestScore, ...hasVariance ? { variance } : {} };
   });
 }
 function iterationStatus(iter, bestIteration) {
@@ -5502,13 +5620,29 @@ async function generateMarkdownReport(workspacePath) {
   lines.push("");
   lines.push("## Iterations");
   lines.push("");
-  lines.push("| Iter | Score | Mutations | Status |");
-  lines.push("|------|-------|-----------|--------|");
+  const hasVariance = iterations.some(
+    (iter) => Object.values(iter.taskResults).some((s) => s.variance)
+  );
+  if (hasVariance) {
+    lines.push("| Iter | Score | Mutations | Status |");
+    lines.push("|------|-------|-----------|--------|");
+  } else {
+    lines.push("| Iter | Score | Mutations | Status |");
+    lines.push("|------|-------|-----------|--------|");
+  }
   for (const iter of iterations) {
     const mutations = iter.proposal?.mutations.length ?? 0;
     const mutStr = mutations > 0 ? mutations.toString() : "-";
     const status = iterationStatus(iter, bestIter.iteration);
-    lines.push(`| ${iter.iteration} | ${iter.score.toFixed(1)}% | ${mutStr} | ${status} |`);
+    let scoreStr = `${iter.score.toFixed(1)}%`;
+    if (hasVariance) {
+      const stddevs = Object.values(iter.taskResults).map((s) => s.variance?.stddev).filter((v) => v !== void 0);
+      if (stddevs.length > 0) {
+        const avgStddev = stddevs.reduce((a, b) => a + b, 0) / stddevs.length;
+        scoreStr = `${iter.score.toFixed(1)}% \xB1${avgStddev.toFixed(1)}`;
+      }
+    }
+    lines.push(`| ${iter.iteration} | ${scoreStr} | ${mutStr} | ${status} |`);
   }
   lines.push("");
   if (leaderboard.length > 0) {
@@ -5521,7 +5655,10 @@ async function generateMarkdownReport(workspacePath) {
     for (const entry of leaderboard) {
       const scoreCols = iterNums.map((n) => {
         const s = entry.scores[n];
-        return s !== void 0 ? `${s.toFixed(0)}%` : "-";
+        if (s === void 0) return "-";
+        const v = entry.variance?.[n];
+        if (v && v.runs > 1) return `${s.toFixed(0)}% \xB1${v.stddev.toFixed(1)}`;
+        return `${s.toFixed(0)}%`;
       });
       lines.push(`| ${entry.taskId} | ${scoreCols.join(" | ")} | ${entry.bestScore.toFixed(0)}% (iter ${entry.bestIteration}) |`);
     }
@@ -5571,12 +5708,17 @@ async function generateJsonReport(workspacePath) {
       bestIteration: bestIter.iteration,
       improvement
     },
-    iterations: iterations.map((iter) => ({
-      iteration: iter.iteration,
-      score: iter.score,
-      mutationCount: iter.proposal?.mutations.length ?? 0,
-      status: iterationStatus(iter, bestIter.iteration)
-    })),
+    iterations: iterations.map((iter) => {
+      const stddevs = Object.values(iter.taskResults).map((s) => s.variance?.stddev).filter((v) => v !== void 0);
+      const avgStddev = stddevs.length > 0 ? stddevs.reduce((a, b) => a + b, 0) / stddevs.length : void 0;
+      return {
+        iteration: iter.iteration,
+        score: iter.score,
+        ...avgStddev !== void 0 ? { stddev: avgStddev } : {},
+        mutationCount: iter.proposal?.mutations.length ?? 0,
+        status: iterationStatus(iter, bestIter.iteration)
+      };
+    }),
     leaderboard,
     counterfactuals
   };
@@ -5766,7 +5908,7 @@ evolveCommand.command("init").description("Initialize an evolution workspace wit
     let addMore = true;
     while (addMore) {
       try {
-        addMore = await confirm3({ message: "Add another eval task?", default: false });
+        addMore = await confirm4({ message: "Add another eval task?", default: false });
       } catch {
         addMore = false;
       }
@@ -6179,10 +6321,12 @@ async function countFiles(dir) {
 }
 
 // src/cli.ts
+var require2 = createRequire(import.meta.url);
+var pkg = require2("../package.json");
 var program = new Command12();
 program.name("kairn").description(
   "Compile natural language intent into optimized Claude Code environments"
-).version("1.9.0").option("--no-color", "Disable colored output");
+).version(pkg.version).option("--no-color", "Disable colored output");
 program.addCommand(initCommand);
 program.addCommand(describeCommand);
 program.addCommand(optimizeCommand);