@rulemetric/cli 0.7.2 → 0.7.4

package/dist/chunk-JULOLTZS.js

@@ -1,61 +0,0 @@
-import {
-  ensureWorktree
-} from "./chunk-Y4BJXXYV.js";
-import {
-  PermanentJobError
-} from "./chunk-DGHWRQXL.js";
-import {
-  apiPatch
-} from "./chunk-ZUJBM6HE.js";
-
-// src/lib/handlers/process-launch.ts
-async function resolveWorktreeForLaunch(config) {
-  if (!config.worktree?.create) {
-    return config;
-  }
-  const { branch, baseRef } = config.worktree;
-  const wt = await ensureWorktree({
-    repoRoot: config.projectPath,
-    branch,
-    baseRef
-  });
-  return {
-    ...config,
-    projectPath: wt.path,
-    gitBranch: wt.branch
-  };
-}
-async function processLaunch(payload, helpers) {
-  const { mirrorId, tool } = payload;
-  let { config } = payload;
-  const log = (msg) => helpers.logger.info(msg);
-  if (helpers.job.attempts > 1) {
-    throw new PermanentJobError(
-      `process_launch is not safe to retry (attempt ${helpers.job.attempts}). Re-launch manually.`
-    );
-  }
-  config = await resolveWorktreeForLaunch(config);
-  if (config.worktree?.create) {
-    log(`Worktree resolved: ${config.projectPath} (branch ${config.gitBranch})`);
-  }
-  log(`Launching ${tool}...`);
-  const { getLauncher } = await import("./dist-UNYP5PJR.js");
-  const launcher = getLauncher(tool);
-  const launcherResult = await launcher.launch(config);
-  log(`Launch completed: sessionId=${launcherResult.sessionId}`);
-  const result = {
-    ...launcherResult,
-    worktreePath: config.worktree?.create ? config.projectPath : void 0,
-    runId: config.runId
-  };
-  await apiPatch(`/api/launch-jobs/${mirrorId}/state`, { result }).catch((err) => {
-    log(`result persist failed: ${err instanceof Error ? err.message : String(err)}`);
-  });
-  return result;
-}
-
-export {
-  resolveWorktreeForLaunch,
-  processLaunch
-};
-//# sourceMappingURL=chunk-JULOLTZS.js.map

package/dist/chunk-JULOLTZS.js.map

@@ -1,7 +0,0 @@
-{
-  "version": 3,
-  "sources": ["../src/lib/handlers/process-launch.ts"],
-  "sourcesContent": ["import type { JobHelpers } from '../job-helpers.js';\nimport type { LaunchConfig as LauncherConfig } from '@rulemetric/session';\nimport { apiPatch } from '../api-client.js';\nimport { PermanentJobError } from '../permanent-job-error.js';\nimport { ensureWorktree } from '../worktree.js';\n\n/**\n * Extends the canonical launcher config (`@rulemetric/session`) with the\n * worker-only fields the `POST /api/runs` fan-out stamps. Extending \u2014 rather\n * than re-declaring \u2014 keeps the shared launcher fields (`projectPath`,\n * `prompt`, `mode`, `resumeSessionId`, \u2026) in lock-step with the launcher so\n * the two can't drift.\n */\nexport interface LaunchConfig extends LauncherConfig {\n  gitBranch?: string;\n  /**\n   * Set by the `POST /api/runs` fan-out (Task 1.1). When `create` is true the\n   * worker provisions an isolated git worktree+branch before launching, so the\n   * session runs inside its own checkout of the repo.\n   */\n  worktree?: {\n    create: boolean;\n    branch: string;\n    baseRef: string;\n  };\n  /**\n   * The owning run's id, stamped by the fan-out. Carried through untouched and\n   * passed to the launcher, which injects it as RULEMETRIC_RUN_ID into the\n   * spawned session's env (Task 3.1). The session-start hook\n   * (packages/hooks/scripts/_ensure-session.sh) forwards it as `runId` to\n   * POST /api/sessions, which stamps sessions.run_id after validating ownership.\n   */\n  runId?: string;\n}\n\ninterface LaunchPayload {\n  mirrorId: string;\n  tool: string;\n  config: LaunchConfig;\n}\n\n/**\n * If the launch carries a worktree spec, provision the worktree and return a\n * NEW config pointed at it (`projectPath` = worktree path, `gitBranch` = the\n * branch). When `worktree.create` is falsy the config is returned unchanged.\n *\n * Extracted as a pure-ish helper so it's unit-testable against a real temp git\n * repo without mocking the `@rulemetric/session` launcher. `allowExistingBranch`\n * is left at its default (false): same-run idempotency is handled by\n * `ensureWorktree`'s pre-existence guard, and the `attempts > 1` guard in\n * `processLaunch` prevents a retry from re-running this at all.\n */\nexport async function resolveWorktreeForLaunch(config: LaunchConfig): Promise<LaunchConfig> {\n  if (!config.worktree?.create) {\n    return config;\n  }\n  const { branch, baseRef } = config.worktree;\n  const wt = await ensureWorktree({\n    repoRoot: config.projectPath,\n    branch,\n    baseRef,\n  });\n  // New object \u2014 never mutate the caller's config in place.\n  return {\n    ...config,\n    projectPath: wt.path,\n    gitBranch: wt.branch,\n  };\n}\n\n/**\n * Graphile-shaped handler for the `process_launch` task.\n *\n * Lifted from the legacy `eval-worker.ts:processLaunchJob`, with the queue\n * plumbing stripped out:\n *   - no claim \u2014 Graphile already claimed the job before invoking us\n *   - no /complete POST \u2014 we return the launcher result (or throw)\n *   - no surrounding try/catch \u2014 errors propagate to Graphile's retry path\n *\n * The dynamic import of `@rulemetric/session` is intentional: it avoids\n * loading the launcher (and its tool-specific subprocess wrappers) unless a\n * launch job actually fires.\n */\nexport async function processLaunch(\n  payload: LaunchPayload,\n  helpers: JobHelpers,\n): Promise<Record<string, unknown>> {\n  const { mirrorId, tool } = payload;\n  let { config } = payload;\n  const log = (msg: string) => helpers.logger.info(msg);\n\n  // Fail-fast on retry. The launcher spawns subprocesses and registers a\n  // session \u2014 re-running on Graphile retry would create a second session\n  // with no link back to the first. Until the launcher path is properly\n  // idempotent (resume from sessionId persisted to the mirror row),\n  // surface the failure to the user instead of silently double-launching.\n  // See docs/audits/2026-05-07-worker-audit.md \"Handlers not idempotent\".\n  if (helpers.job.attempts > 1) {\n    throw new PermanentJobError(\n      `process_launch is not safe to retry (attempt ${helpers.job.attempts}). Re-launch manually.`,\n    );\n  }\n\n  // Provision an isolated worktree (idempotent) BEFORE the launcher spawns \u2014\n  // the launcher uses config.projectPath as the session cwd, so swapping it to\n  // the worktree path is what makes the session run inside the worktree.\n  config = await resolveWorktreeForLaunch(config);\n  if (config.worktree?.create) {\n    log(`Worktree resolved: ${config.projectPath} (branch ${config.gitBranch})`);\n  }\n\n  log(`Launching ${tool}...`);\n  const { getLauncher } = await import('@rulemetric/session');\n  const launcher = getLauncher(tool);\n  // config.runId rides through here untouched; the launcher injects it as\n  // RULEMETRIC_RUN_ID into the spawned session's env (Task 3.1).\n  const launcherResult = await launcher.launch(config);\n  log(`Launch completed: sessionId=${launcherResult.sessionId}`);\n\n  // Persist the RESOLVED worktree path + runId onto the launch result so the\n  // API's reconcile fallback (POST /api/runs/:id/link-sessions, Task 3.2) can\n  // match a run's launch jobs to its captured sessions on `projectPath`. The\n  // env\u2192hook path (Task 3.1) is the primary attribution channel; this is the\n  // defense-in-depth join key for sessions that channel misses (non-Claude-Code\n  // tools whose launchers don't inject RULEMETRIC_RUN_ID, proxy-created\n  // sessions, tools without our hooks). `config.projectPath` here is already\n  // the worktree path \u2014 `resolveWorktreeForLaunch` swapped it above \u2014 so the\n  // captured `sessions.projectPath` (the session's cwd) equals it.\n  const result: Record<string, unknown> = {\n    ...launcherResult,\n    worktreePath: config.worktree?.create ? config.projectPath : undefined,\n    runId: config.runId,\n  };\n\n  // Graphile discards task return values, so persist the launcher result\n  // (sessionId, etc.) onto the mirror row. Best-effort.\n  await apiPatch(`/api/launch-jobs/${mirrorId}/state`, { result }).catch((err: unknown) => {\n    log(`result persist failed: ${err instanceof Error ? err.message : String(err)}`);\n  });\n\n  return result;\n}\n"],
-  "mappings": ";;;;;;;;;;;AAoDA,eAAsB,yBAAyB,QAA6C;AAC1F,MAAI,CAAC,OAAO,UAAU,QAAQ;AAC5B,WAAO;AAAA,EACT;AACA,QAAM,EAAE,QAAQ,QAAQ,IAAI,OAAO;AACnC,QAAM,KAAK,MAAM,eAAe;AAAA,IAC9B,UAAU,OAAO;AAAA,IACjB;AAAA,IACA;AAAA,EACF,CAAC;AAED,SAAO;AAAA,IACL,GAAG;AAAA,IACH,aAAa,GAAG;AAAA,IAChB,WAAW,GAAG;AAAA,EAChB;AACF;AAeA,eAAsB,cACpB,SACA,SACkC;AAClC,QAAM,EAAE,UAAU,KAAK,IAAI;AAC3B,MAAI,EAAE,OAAO,IAAI;AACjB,QAAM,MAAM,CAAC,QAAgB,QAAQ,OAAO,KAAK,GAAG;AAQpD,MAAI,QAAQ,IAAI,WAAW,GAAG;AAC5B,UAAM,IAAI;AAAA,MACR,gDAAgD,QAAQ,IAAI,QAAQ;AAAA,IACtE;AAAA,EACF;AAKA,WAAS,MAAM,yBAAyB,MAAM;AAC9C,MAAI,OAAO,UAAU,QAAQ;AAC3B,QAAI,sBAAsB,OAAO,WAAW,YAAY,OAAO,SAAS,GAAG;AAAA,EAC7E;AAEA,MAAI,aAAa,IAAI,KAAK;AAC1B,QAAM,EAAE,YAAY,IAAI,MAAM,OAAO,oBAAqB;AAC1D,QAAM,WAAW,YAAY,IAAI;AAGjC,QAAM,iBAAiB,MAAM,SAAS,OAAO,MAAM;AACnD,MAAI,+BAA+B,eAAe,SAAS,EAAE;AAW7D,QAAM,SAAkC;AAAA,IACtC,GAAG;AAAA,IACH,cAAc,OAAO,UAAU,SAAS,OAAO,cAAc;AAAA,IAC7D,OAAO,OAAO;AAAA,EAChB;AAIA,QAAM,SAAS,oBAAoB,QAAQ,UAAU,EAAE,OAAO,CAAC,EAAE,MAAM,CAAC,QAAiB;AACvF,QAAI,0BAA0B,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG,CAAC,EAAE;AAAA,EAClF,CAAC;AAED,SAAO;AACT;",
-  "names": []
-}

package/dist/chunk-KJKBDQ4D.js

@@ -1,220 +0,0 @@
-import {
-  analyzeEvalResults
-} from "./chunk-3TIMQ3O6.js";
-import {
-  gradeEvalOutput
-} from "./chunk-YNIH3KMU.js";
-import {
-  executeEval
-} from "./chunk-QN34TGD3.js";
-import {
-  apiGet,
-  apiPost
-} from "./chunk-ZUJBM6HE.js";
-
-// src/lib/eval-conversation-handler.ts
-async function processConversationJob(jobConfig, evalTargetId, logger) {
-  const { conversationId, messageId, messageContent } = jobConfig;
-  const target = await apiGet(`/api/eval-targets/${evalTargetId}`);
-  const evalCases = await apiGet(`/api/eval-targets/${evalTargetId}/evals`);
-  const messages = await apiGet(
-    `/api/eval-conversations/${conversationId}/messages`
-  );
-  const msgIdx = messages.findIndex((m) => m.id === messageId);
-  if (msgIdx >= 0) {
-    const hasResponse = messages.slice(msgIdx + 1).some(
-      (m) => m.role === "assistant" && (m.type === "results" || m.type === "proposal" || m.type === "generated" || m.type === "text")
-    );
-    if (hasResponse) {
-      logger(`Message ${messageId} already processed \u2014 skipping (idempotency guard)`);
-      return;
-    }
-  }
-  const postMessage = async (content, type, data) => {
-    await apiPost(`/api/eval-conversations/${conversationId}/messages/assistant`, {
-      content,
-      type,
-      data
-    });
-  };
-  const decision = await decideAction(target, evalCases, messages, messageContent, logger);
-  logger(`Action: ${decision.action}`);
-  switch (decision.action) {
-    case "run_evals":
-      await handleRunEvals(target, evalCases, conversationId, postMessage, logger);
-      break;
-    case "generate_cases":
-      await handleGenerateCases(target, evalTargetId, conversationId, postMessage, logger);
-      break;
-    case "analyze_results":
-    case "propose_changes":
-      await handleAnalyzeAndPropose(target, evalCases, messages, messageContent, conversationId, postMessage, logger);
-      break;
-    case "respond":
-      await postMessage(decision.response, "text");
-      break;
-  }
-}
-async function decideAction(target, evalCases, messages, latestMessage, logger) {
-  const recentMessages = messages.slice(-10).map(
-    (m) => `${m.role}: ${m.content.slice(0, 200)}`
-  ).join("\n");
-  const prompt = `You are an eval assistant. Decide what action to take based on the user's message.
-
-## Current State
-- Eval target: "${target.name}" v${target.version} (${target.type})
-- Eval cases: ${evalCases.length} defined${evalCases.length > 0 ? ` (${evalCases.map((e) => e.name).join(", ")})` : ""}
-- Content length: ${target.content.length} chars
-
-## Recent Conversation
-${recentMessages}
-
-## User's Latest Message
-${latestMessage}
-
-## Available Actions
-- run_evals: Run all eval cases against current version (with + without target). Choose this when the user wants to test, evaluate, or see how the instruction performs.
-- generate_cases: Auto-generate eval test cases from the instruction content. Choose this when there are no eval cases (${evalCases.length} currently) or user asks to create/generate tests.
-- analyze_results: Analyze recent eval failures and propose improvements based on user feedback. Choose this when user gives qualitative feedback like "this is wrong", "missing X", etc.
-- propose_changes: Propose specific content edits based on conversation. Choose this when user asks for improvements or changes.
-- respond: Just answer a question without running any evals. Choose this for informational questions.
-
-Respond with ONLY valid JSON:
-{"action": "run_evals", "response": "What you want to say to the user", "params": {}}`;
-  const result = await executeEval({ prompt, timeout: 3e4 });
-  try {
-    let jsonStr = result.output.trim();
-    const jsonMatch = jsonStr.match(/\{[\s\S]*\}/);
-    if (jsonMatch) jsonStr = jsonMatch[0];
-    return JSON.parse(jsonStr);
-  } catch {
-    logger("Failed to parse action decision, defaulting to respond");
-    const lower = latestMessage.toLowerCase();
-    if (lower.includes("run") || lower.includes("test") || lower.includes("eval")) {
-      return { action: "run_evals", response: "Running evals now..." };
-    }
-    if (lower.includes("generate") || lower.includes("create") || evalCases.length === 0 && lower.includes("start")) {
-      return { action: "generate_cases", response: "Generating test cases..." };
-    }
-    return { action: "respond", response: "I can help you run evals, generate test cases, or analyze results. What would you like to do?" };
-  }
-}
-async function handleRunEvals(target, evalCases, _conversationId, postMessage, logger) {
-  if (evalCases.length === 0) {
-    await postMessage(
-      "No eval cases exist yet. Would you like me to generate some from the instruction content?",
-      "text"
-    );
-    return;
-  }
-  await postMessage(`Running ${evalCases.length} evals against "${target.name}" v${target.version}...`, "status");
-  const results = [];
-  for (const eval_ of evalCases) {
-    logger(`Running: ${eval_.name} (with_target)`);
-    const execResult = await executeEval({
-      prompt: eval_.prompt,
-      targetContent: target.content,
-      timeout: 12e4
-    });
-    let passed = false;
-    let evidence = "";
-    if (!execResult.error && eval_.expectations.length > 0) {
-      const grading = await gradeEvalOutput({
-        prompt: eval_.prompt,
-        output: execResult.output,
-        expectations: eval_.expectations,
-        timeout: 12e4
-      });
-      const summary = grading?.summary;
-      passed = summary?.pass_rate >= 1;
-      evidence = grading?.expectations?.map((e) => `${e.passed ? "PASS" : "FAIL"}: ${e.text} \u2014 ${e.evidence}`).join("\n") ?? "";
-    }
-    results.push({
-      evalName: eval_.name,
-      passed,
-      output: execResult.error ?? execResult.output.slice(0, 2e3),
-      expectations: eval_.expectations,
-      evidence,
-      configuration: "with_target"
-    });
-  }
-  const passedCount = results.filter((r) => r.passed).length;
-  const total = results.length;
-  await postMessage(
-    `${passedCount} of ${total} evals passed (${Math.round(passedCount / total * 100)}%)`,
-    "results",
-    {
-      summary: { passed: passedCount, failed: total - passedCount, total, passRate: passedCount / total },
-      results: results.map((r) => ({
-        evalName: r.evalName,
-        passed: r.passed,
-        output: r.output,
-        expectations: r.expectations,
-        evidence: r.evidence
-      }))
-    }
-  );
-}
-async function handleGenerateCases(target, _evalTargetId, _conversationId, postMessage, logger) {
-  await postMessage("Generating eval cases from instruction content...", "status");
-  logger("Calling eval-from-context endpoint");
-  const result = await apiPost("/api/eval-from-context", {
-    contentHash: "",
-    name: target.name,
-    type: target.type,
-    content: target.content,
-    config: { skipBaseline: false, skipGrading: false }
-  });
-  const cases = result.evals ?? [];
-  await postMessage(
-    `Generated ${cases.length} eval cases. Review them below:`,
-    "generated",
-    {
-      cases: cases.map((c) => ({
-        id: c.id,
-        name: c.name,
-        prompt: c.prompt,
-        expectations: c.expectations
-      }))
-    }
-  );
-}
-async function handleAnalyzeAndPropose(target, evalCases, messages, userFeedback, _conversationId, postMessage, logger) {
-  await postMessage("Analyzing results and preparing improvements...", "status");
-  const lastResults = [...messages].reverse().find((m) => m.type === "results");
-  const evalResults = lastResults?.data?.results?.map((r) => ({
-    evalName: r.evalName,
-    prompt: "",
-    expectations: r.expectations ?? [],
-    grading: {
-      summary: { pass_rate: r.passed ? 1 : 0 },
-      expectations: r.expectations?.map((e) => ({
-        text: e,
-        passed: r.passed,
-        evidence: r.evidence ?? ""
-      }))
-    }
-  })) ?? [];
-  logger(`Analyzing ${evalResults.length} eval results with user feedback`);
-  const analysis = await analyzeEvalResults({
-    content: target.content,
-    evalResults,
-    model: void 0,
-    timeout: 12e4,
-    userFeedback
-  });
-  await postMessage(
-    analysis.overallAssessment,
-    "proposal",
-    {
-      suggestions: analysis.suggestions,
-      proposedContent: analysis.proposedContent,
-      fromVersion: target.version
-    }
-  );
-}
-
-export {
-  processConversationJob
-};
-//# sourceMappingURL=chunk-KJKBDQ4D.js.map

package/dist/chunk-KJKBDQ4D.js.map

@@ -1,7 +0,0 @@
-{
-  "version": 3,
-  "sources": ["../src/lib/eval-conversation-handler.ts"],
-  "sourcesContent": ["import { apiGet, apiPost } from './api-client.js';\nimport { executeEval } from './eval-executor.js';\nimport { gradeEvalOutput } from './eval-grader.js';\nimport { analyzeEvalResults } from './eval-analyzer.js';\n\ninterface ConversationMessage {\n  id: string;\n  role: 'user' | 'assistant';\n  content: string;\n  type?: string;\n  data?: Record<string, unknown>;\n  createdAt: string;\n}\n\ninterface EvalTarget {\n  id: string;\n  name: string;\n  type: string;\n  content: string;\n  description: string;\n  version: number;\n}\n\ninterface EvalCase {\n  id: string;\n  name: string;\n  prompt: string;\n  expectedOutput: string;\n  expectations: string[];\n}\n\ninterface ConversationJobConfig {\n  type: 'conversation';\n  conversationId: string;\n  messageId: string;\n  messageContent: string;\n}\n\ninterface ActionDecision {\n  action: 'run_evals' | 'generate_cases' | 'analyze_results' | 'propose_changes' | 'respond';\n  response: string;\n  params?: Record<string, unknown>;\n}\n\nexport async function processConversationJob(\n  jobConfig: ConversationJobConfig,\n  evalTargetId: string,\n  logger: (msg: string) => void,\n): Promise<void> {\n  const { conversationId, messageId, messageContent } = jobConfig;\n\n  // Fetch context\n  const target = await apiGet<EvalTarget>(`/api/eval-targets/${evalTargetId}`);\n  const evalCases = await apiGet<EvalCase[]>(`/api/eval-targets/${evalTargetId}/evals`);\n  const messages = await apiGet<ConversationMessage[]>(\n    `/api/eval-conversations/${conversationId}/messages`,\n  );\n\n  // Idempotency: check if this message was already processed (has assistant responses after it)\n  const msgIdx = messages.findIndex((m) => m.id === messageId);\n  if (msgIdx >= 0) {\n    const hasResponse = messages.slice(msgIdx + 1).some(\n      (m) => m.role === 'assistant' && (m.type === 'results' || m.type === 'proposal' || m.type === 'generated' || m.type === 'text'),\n    );\n    if (hasResponse) {\n      logger(`Message ${messageId} already processed \u2014 skipping (idempotency guard)`);\n      return;\n    }\n  }\n\n  // Helper to post assistant messages\n  const postMessage = async (content: string, type: string, data?: Record<string, unknown>) => {\n    await apiPost(`/api/eval-conversations/${conversationId}/messages/assistant`, {\n      content,\n      type,\n      data,\n    });\n  };\n\n  // Decide what action to take\n  const decision = await decideAction(target, evalCases, messages, messageContent, logger);\n\n  logger(`Action: ${decision.action}`);\n\n  switch (decision.action) {\n    case 'run_evals':\n      await handleRunEvals(target, evalCases, conversationId, postMessage, logger);\n      break;\n\n    case 'generate_cases':\n      await handleGenerateCases(target, evalTargetId, conversationId, postMessage, logger);\n      break;\n\n    case 'analyze_results':\n    case 'propose_changes':\n      await handleAnalyzeAndPropose(target, evalCases, messages, messageContent, conversationId, postMessage, logger);\n      break;\n\n    case 'respond':\n      await postMessage(decision.response, 'text');\n      break;\n  }\n}\n\nasync function decideAction(\n  target: EvalTarget,\n  evalCases: EvalCase[],\n  messages: ConversationMessage[],\n  latestMessage: string,\n  logger: (msg: string) => void,\n): Promise<ActionDecision> {\n  // Build conversation summary for context\n  const recentMessages = messages.slice(-10).map((m) =>\n    `${m.role}: ${m.content.slice(0, 200)}`\n  ).join('\\n');\n\n  const prompt = `You are an eval assistant. Decide what action to take based on the user's message.\n\n## Current State\n- Eval target: \"${target.name}\" v${target.version} (${target.type})\n- Eval cases: ${evalCases.length} defined${evalCases.length > 0 ? ` (${evalCases.map((e) => e.name).join(', ')})` : ''}\n- Content length: ${target.content.length} chars\n\n## Recent Conversation\n${recentMessages}\n\n## User's Latest Message\n${latestMessage}\n\n## Available Actions\n- run_evals: Run all eval cases against current version (with + without target). Choose this when the user wants to test, evaluate, or see how the instruction performs.\n- generate_cases: Auto-generate eval test cases from the instruction content. Choose this when there are no eval cases (${evalCases.length} currently) or user asks to create/generate tests.\n- analyze_results: Analyze recent eval failures and propose improvements based on user feedback. Choose this when user gives qualitative feedback like \"this is wrong\", \"missing X\", etc.\n- propose_changes: Propose specific content edits based on conversation. Choose this when user asks for improvements or changes.\n- respond: Just answer a question without running any evals. Choose this for informational questions.\n\nRespond with ONLY valid JSON:\n{\"action\": \"run_evals\", \"response\": \"What you want to say to the user\", \"params\": {}}`;\n\n  const result = await executeEval({ prompt, timeout: 30_000 });\n\n  try {\n    let jsonStr = result.output.trim();\n    const jsonMatch = jsonStr.match(/\\{[\\s\\S]*\\}/);\n    if (jsonMatch) jsonStr = jsonMatch[0];\n    return JSON.parse(jsonStr) as ActionDecision;\n  } catch {\n    logger('Failed to parse action decision, defaulting to respond');\n    // Fallback: simple heuristics\n    const lower = latestMessage.toLowerCase();\n    if (lower.includes('run') || lower.includes('test') || lower.includes('eval')) {\n      return { action: 'run_evals', response: 'Running evals now...' };\n    }\n    if (lower.includes('generate') || lower.includes('create') || (evalCases.length === 0 && lower.includes('start'))) {\n      return { action: 'generate_cases', response: 'Generating test cases...' };\n    }\n    return { action: 'respond', response: 'I can help you run evals, generate test cases, or analyze results. What would you like to do?' };\n  }\n}\n\nasync function handleRunEvals(\n  target: EvalTarget,\n  evalCases: EvalCase[],\n  _conversationId: string,\n  postMessage: (content: string, type: string, data?: Record<string, unknown>) => Promise<void>,\n  logger: (msg: string) => void,\n): Promise<void> {\n  if (evalCases.length === 0) {\n    await postMessage(\n      'No eval cases exist yet. Would you like me to generate some from the instruction content?',\n      'text',\n    );\n    return;\n  }\n\n  await postMessage(`Running ${evalCases.length} evals against \"${target.name}\" v${target.version}...`, 'status');\n\n  const results: Array<{\n    evalName: string;\n    passed: boolean;\n    output: string;\n    expectations: string[];\n    evidence: string;\n    configuration: string;\n  }> = [];\n\n  // Run with_target\n  for (const eval_ of evalCases) {\n    logger(`Running: ${eval_.name} (with_target)`);\n    const execResult = await executeEval({\n      prompt: eval_.prompt,\n      targetContent: target.content,\n      timeout: 120_000,\n    });\n\n    let passed = false;\n    let evidence = '';\n\n    if (!execResult.error && eval_.expectations.length > 0) {\n      const grading = await gradeEvalOutput({\n        prompt: eval_.prompt,\n        output: execResult.output,\n        expectations: eval_.expectations,\n        timeout: 120_000,\n      });\n      const summary = (grading as any)?.summary;\n      passed = summary?.pass_rate >= 1.0;\n      evidence = (grading as any)?.expectations\n        ?.map((e: any) => `${e.passed ? 'PASS' : 'FAIL'}: ${e.text} \u2014 ${e.evidence}`)\n        .join('\\n') ?? '';\n    }\n\n    results.push({\n      evalName: eval_.name,\n      passed,\n      output: execResult.error ?? execResult.output.slice(0, 2000),\n      expectations: eval_.expectations,\n      evidence,\n      configuration: 'with_target',\n    });\n  }\n\n  const passedCount = results.filter((r) => r.passed).length;\n  const total = results.length;\n\n  await postMessage(\n    `${passedCount} of ${total} evals passed (${Math.round((passedCount / total) * 100)}%)`,\n    'results',\n    {\n      summary: { passed: passedCount, failed: total - passedCount, total, passRate: passedCount / total },\n      results: results.map((r) => ({\n        evalName: r.evalName,\n        passed: r.passed,\n        output: r.output,\n        expectations: r.expectations,\n        evidence: r.evidence,\n      })),\n    },\n  );\n}\n\nasync function handleGenerateCases(\n  target: EvalTarget,\n  _evalTargetId: string,\n  _conversationId: string,\n  postMessage: (content: string, type: string, data?: Record<string, unknown>) => Promise<void>,\n  logger: (msg: string) => void,\n): Promise<void> {\n  await postMessage('Generating eval cases from instruction content...', 'status');\n\n  logger('Calling eval-from-context endpoint');\n\n  // Use eval-from-context endpoint\n  const result = await apiPost<{\n    evalTarget: { id: string };\n    evals: Array<{ id: string; name: string; prompt: string; expectations: string[] }>;\n    job: { id: string } | null;\n  }>('/api/eval-from-context', {\n    contentHash: '',\n    name: target.name,\n    type: target.type,\n    content: target.content,\n    config: { skipBaseline: false, skipGrading: false },\n  });\n\n  const cases = result.evals ?? [];\n\n  await postMessage(\n    `Generated ${cases.length} eval cases. Review them below:`,\n    'generated',\n    {\n      cases: cases.map((c) => ({\n        id: c.id,\n        name: c.name,\n        prompt: c.prompt,\n        expectations: c.expectations,\n      })),\n    },\n  );\n}\n\nasync function handleAnalyzeAndPropose(\n  target: EvalTarget,\n  evalCases: EvalCase[],\n  messages: ConversationMessage[],\n  userFeedback: string,\n  _conversationId: string,\n  postMessage: (content: string, type: string, data?: Record<string, unknown>) => Promise<void>,\n  logger: (msg: string) => void,\n): Promise<void> {\n  await postMessage('Analyzing results and preparing improvements...', 'status');\n\n  // Find the most recent results message for context\n  const lastResults = [...messages]\n    .reverse()\n    .find((m) => m.type === 'results');\n\n  const evalResults = (lastResults?.data as any)?.results?.map((r: any) => ({\n    evalName: r.evalName,\n    prompt: '',\n    expectations: r.expectations ?? [],\n    grading: {\n      summary: { pass_rate: r.passed ? 1 : 0 },\n      expectations: r.expectations?.map((e: string) => ({\n        text: e,\n        passed: r.passed,\n        evidence: r.evidence ?? '',\n      })),\n    },\n  })) ?? [];\n\n  logger(`Analyzing ${evalResults.length} eval results with user feedback`);\n\n  // Include user feedback in the analysis\n  const analysis = await analyzeEvalResults({\n    content: target.content,\n    evalResults,\n    model: undefined,\n    timeout: 120_000,\n    userFeedback,\n  });\n\n  await postMessage(\n    analysis.overallAssessment,\n    'proposal',\n    {\n      suggestions: analysis.suggestions,\n      proposedContent: analysis.proposedContent,\n      fromVersion: target.version,\n    },\n  );\n}\n"],
-  "mappings": ";;;;;;;;;;;;;;;AA4CA,eAAsB,uBACpB,WACA,cACA,QACe;AACf,QAAM,EAAE,gBAAgB,WAAW,eAAe,IAAI;AAGtD,QAAM,SAAS,MAAM,OAAmB,qBAAqB,YAAY,EAAE;AAC3E,QAAM,YAAY,MAAM,OAAmB,qBAAqB,YAAY,QAAQ;AACpF,QAAM,WAAW,MAAM;AAAA,IACrB,2BAA2B,cAAc;AAAA,EAC3C;AAGA,QAAM,SAAS,SAAS,UAAU,CAAC,MAAM,EAAE,OAAO,SAAS;AAC3D,MAAI,UAAU,GAAG;AACf,UAAM,cAAc,SAAS,MAAM,SAAS,CAAC,EAAE;AAAA,MAC7C,CAAC,MAAM,EAAE,SAAS,gBAAgB,EAAE,SAAS,aAAa,EAAE,SAAS,cAAc,EAAE,SAAS,eAAe,EAAE,SAAS;AAAA,IAC1H;AACA,QAAI,aAAa;AACf,aAAO,WAAW,SAAS,wDAAmD;AAC9E;AAAA,IACF;AAAA,EACF;AAGA,QAAM,cAAc,OAAO,SAAiB,MAAc,SAAmC;AAC3F,UAAM,QAAQ,2BAA2B,cAAc,uBAAuB;AAAA,MAC5E;AAAA,MACA;AAAA,MACA;AAAA,IACF,CAAC;AAAA,EACH;AAGA,QAAM,WAAW,MAAM,aAAa,QAAQ,WAAW,UAAU,gBAAgB,MAAM;AAEvF,SAAO,WAAW,SAAS,MAAM,EAAE;AAEnC,UAAQ,SAAS,QAAQ;AAAA,IACvB,KAAK;AACH,YAAM,eAAe,QAAQ,WAAW,gBAAgB,aAAa,MAAM;AAC3E;AAAA,IAEF,KAAK;AACH,YAAM,oBAAoB,QAAQ,cAAc,gBAAgB,aAAa,MAAM;AACnF;AAAA,IAEF,KAAK;AAAA,IACL,KAAK;AACH,YAAM,wBAAwB,QAAQ,WAAW,UAAU,gBAAgB,gBAAgB,aAAa,MAAM;AAC9G;AAAA,IAEF,KAAK;AACH,YAAM,YAAY,SAAS,UAAU,MAAM;AAC3C;AAAA,EACJ;AACF;AAEA,eAAe,aACb,QACA,WACA,UACA,eACA,QACyB;AAEzB,QAAM,iBAAiB,SAAS,MAAM,GAAG,EAAE;AAAA,IAAI,CAAC,MAC9C,GAAG,EAAE,IAAI,KAAK,EAAE,QAAQ,MAAM,GAAG,GAAG,CAAC;AAAA,EACvC,EAAE,KAAK,IAAI;AAEX,QAAM,SAAS;AAAA;AAAA;AAAA,kBAGC,OAAO,IAAI,MAAM,OAAO,OAAO,KAAK,OAAO,IAAI;AAAA,gBACjD,UAAU,MAAM,WAAW,UAAU,SAAS,IAAI,KAAK,UAAU,IAAI,CAAC,MAAM,EAAE,IAAI,EAAE,KAAK,IAAI,CAAC,MAAM,EAAE;AAAA,oBAClG,OAAO,QAAQ,MAAM;AAAA;AAAA;AAAA,EAGvC,cAAc;AAAA;AAAA;AAAA,EAGd,aAAa;AAAA;AAAA;AAAA;AAAA,0HAI2G,UAAU,MAAM;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAQxI,QAAM,SAAS,MAAM,YAAY,EAAE,QAAQ,SAAS,IAAO,CAAC;AAE5D,MAAI;AACF,QAAI,UAAU,OAAO,OAAO,KAAK;AACjC,UAAM,YAAY,QAAQ,MAAM,aAAa;AAC7C,QAAI,UAAW,WAAU,UAAU,CAAC;AACpC,WAAO,KAAK,MAAM,OAAO;AAAA,EAC3B,QAAQ;AACN,WAAO,wDAAwD;AAE/D,UAAM,QAAQ,cAAc,YAAY;AACxC,QAAI,MAAM,SAAS,KAAK,KAAK,MAAM,SAAS,MAAM,KAAK,MAAM,SAAS,MAAM,GAAG;AAC7E,aAAO,EAAE,QAAQ,aAAa,UAAU,uBAAuB;AAAA,IACjE;AACA,QAAI,MAAM,SAAS,UAAU,KAAK,MAAM,SAAS,QAAQ,KAAM,UAAU,WAAW,KAAK,MAAM,SAAS,OAAO,GAAI;AACjH,aAAO,EAAE,QAAQ,kBAAkB,UAAU,2BAA2B;AAAA,IAC1E;AACA,WAAO,EAAE,QAAQ,WAAW,UAAU,gGAAgG;AAAA,EACxI;AACF;AAEA,eAAe,eACb,QACA,WACA,iBACA,aACA,QACe;AACf,MAAI,UAAU,WAAW,GAAG;AAC1B,UAAM;AAAA,MACJ;AAAA,MACA;AAAA,IACF;AACA;AAAA,EACF;AAEA,QAAM,YAAY,WAAW,UAAU,MAAM,mBAAmB,OAAO,IAAI,MAAM,OAAO,OAAO,OAAO,QAAQ;AAE9G,QAAM,UAOD,CAAC;AAGN,aAAW,SAAS,WAAW;AAC7B,WAAO,YAAY,MAAM,IAAI,gBAAgB;AAC7C,UAAM,aAAa,MAAM,YAAY;AAAA,MACnC,QAAQ,MAAM;AAAA,MACd,eAAe,OAAO;AAAA,MACtB,SAAS;AAAA,IACX,CAAC;AAED,QAAI,SAAS;AACb,QAAI,WAAW;AAEf,QAAI,CAAC,WAAW,SAAS,MAAM,aAAa,SAAS,GAAG;AACtD,YAAM,UAAU,MAAM,gBAAgB;AAAA,QACpC,QAAQ,MAAM;AAAA,QACd,QAAQ,WAAW;AAAA,QACnB,cAAc,MAAM;AAAA,QACpB,SAAS;AAAA,MACX,CAAC;AACD,YAAM,UAAW,SAAiB;AAClC,eAAS,SAAS,aAAa;AAC/B,iBAAY,SAAiB,cACzB,IAAI,CAAC,MAAW,GAAG,EAAE,SAAS,SAAS,MAAM,KAAK,EAAE,IAAI,WAAM,EAAE,QAAQ,EAAE,EAC3E,KAAK,IAAI,KAAK;AAAA,IACnB;AAEA,YAAQ,KAAK;AAAA,MACX,UAAU,MAAM;AAAA,MAChB;AAAA,MACA,QAAQ,WAAW,SAAS,WAAW,OAAO,MAAM,GAAG,GAAI;AAAA,MAC3D,cAAc,MAAM;AAAA,MACpB;AAAA,MACA,eAAe;AAAA,IACjB,CAAC;AAAA,EACH;AAEA,QAAM,cAAc,QAAQ,OAAO,CAAC,MAAM,EAAE,MAAM,EAAE;AACpD,QAAM,QAAQ,QAAQ;AAEtB,QAAM;AAAA,IACJ,GAAG,WAAW,OAAO,KAAK,kBAAkB,KAAK,MAAO,cAAc,QAAS,GAAG,CAAC;AAAA,IACnF;AAAA,IACA;AAAA,MACE,SAAS,EAAE,QAAQ,aAAa,QAAQ,QAAQ,aAAa,OAAO,UAAU,cAAc,MAAM;AAAA,MAClG,SAAS,QAAQ,IAAI,CAAC,OAAO;AAAA,QAC3B,UAAU,EAAE;AAAA,QACZ,QAAQ,EAAE;AAAA,QACV,QAAQ,EAAE;AAAA,QACV,cAAc,EAAE;AAAA,QAChB,UAAU,EAAE;AAAA,MACd,EAAE;AAAA,IACJ;AAAA,EACF;AACF;AAEA,eAAe,oBACb,QACA,eACA,iBACA,aACA,QACe;AACf,QAAM,YAAY,qDAAqD,QAAQ;AAE/E,SAAO,oCAAoC;AAG3C,QAAM,SAAS,MAAM,QAIlB,0BAA0B;AAAA,IAC3B,aAAa;AAAA,IACb,MAAM,OAAO;AAAA,IACb,MAAM,OAAO;AAAA,IACb,SAAS,OAAO;AAAA,IAChB,QAAQ,EAAE,cAAc,OAAO,aAAa,MAAM;AAAA,EACpD,CAAC;AAED,QAAM,QAAQ,OAAO,SAAS,CAAC;AAE/B,QAAM;AAAA,IACJ,aAAa,MAAM,MAAM;AAAA,IACzB;AAAA,IACA;AAAA,MACE,OAAO,MAAM,IAAI,CAAC,OAAO;AAAA,QACvB,IAAI,EAAE;AAAA,QACN,MAAM,EAAE;AAAA,QACR,QAAQ,EAAE;AAAA,QACV,cAAc,EAAE;AAAA,MAClB,EAAE;AAAA,IACJ;AAAA,EACF;AACF;AAEA,eAAe,wBACb,QACA,WACA,UACA,cACA,iBACA,aACA,QACe;AACf,QAAM,YAAY,mDAAmD,QAAQ;AAG7E,QAAM,cAAc,CAAC,GAAG,QAAQ,EAC7B,QAAQ,EACR,KAAK,CAAC,MAAM,EAAE,SAAS,SAAS;AAEnC,QAAM,cAAe,aAAa,MAAc,SAAS,IAAI,CAAC,OAAY;AAAA,IACxE,UAAU,EAAE;AAAA,IACZ,QAAQ;AAAA,IACR,cAAc,EAAE,gBAAgB,CAAC;AAAA,IACjC,SAAS;AAAA,MACP,SAAS,EAAE,WAAW,EAAE,SAAS,IAAI,EAAE;AAAA,MACvC,cAAc,EAAE,cAAc,IAAI,CAAC,OAAe;AAAA,QAChD,MAAM;AAAA,QACN,QAAQ,EAAE;AAAA,QACV,UAAU,EAAE,YAAY;AAAA,MAC1B,EAAE;AAAA,IACJ;AAAA,EACF,EAAE,KAAK,CAAC;AAER,SAAO,aAAa,YAAY,MAAM,kCAAkC;AAGxE,QAAM,WAAW,MAAM,mBAAmB;AAAA,IACxC,SAAS,OAAO;AAAA,IAChB;AAAA,IACA,OAAO;AAAA,IACP,SAAS;AAAA,IACT;AAAA,EACF,CAAC;AAED,QAAM;AAAA,IACJ,SAAS;AAAA,IACT;AAAA,IACA;AAAA,MACE,aAAa,SAAS;AAAA,MACtB,iBAAiB,SAAS;AAAA,MAC1B,aAAa,OAAO;AAAA,IACtB;AAAA,EACF;AACF;",
-  "names": []
-}

package/dist/chunk-KRBQLMOP.js

@@ -1,41 +0,0 @@
-// src/lib/tmux.ts
-import { spawn } from "node:child_process";
-function tmuxSpawnEnv() {
-  const extraDirs = ["/opt/homebrew/bin", "/usr/local/bin"];
-  const parts = (process.env.PATH ?? "").split(":").filter(Boolean);
-  const missing = extraDirs.filter((d) => !parts.includes(d));
-  if (missing.length === 0) return process.env;
-  return { ...process.env, PATH: [...parts, ...missing].join(":") };
-}
-function runTmux(args) {
-  return new Promise((resolve) => {
-    const proc = spawn("tmux", args, {
-      stdio: ["ignore", "ignore", "pipe"],
-      env: tmuxSpawnEnv()
-    });
-    let stderr = "";
-    proc.stderr.on("data", (d) => {
-      stderr += d.toString();
-    });
-    proc.on("close", (code) => resolve({ code, stderr }));
-    proc.on("error", (err) => resolve({ code: null, stderr: err.message }));
-  });
-}
-function buildTmuxHasSessionArgs(target) {
-  return ["has-session", "-t", target];
-}
-function buildTmuxSendLiteralArgs(target, message) {
-  return ["send-keys", "-t", target, "-l", "--", message];
-}
-function buildTmuxSendEnterArgs(target) {
-  return ["send-keys", "-t", target, "Enter"];
-}
-
-export {
-  tmuxSpawnEnv,
-  runTmux,
-  buildTmuxHasSessionArgs,
-  buildTmuxSendLiteralArgs,
-  buildTmuxSendEnterArgs
-};
-//# sourceMappingURL=chunk-KRBQLMOP.js.map

package/dist/chunk-KRBQLMOP.js.map

@@ -1,7 +0,0 @@
-{
-  "version": 3,
-  "sources": ["../src/lib/tmux.ts"],
-  "sourcesContent": ["import { spawn } from 'node:child_process';\n\nexport interface TmuxRunResult {\n  code: number | null;\n  stderr: string;\n}\n\n// Homebrew installs tmux to /opt/homebrew/bin (Apple Silicon) or\n// /usr/local/bin (Intel). The launchd worker's minimal PATH omits\n// /opt/homebrew/bin, so a bare spawn('tmux') fails with ENOENT under launchd\n// \u2014 which silently flips every terminal target to `missing` and breaks live\n// send. Augment PATH at the spawn boundary. Mirrors the same fix in\n// packages/session terminal tmux (verified 2026-06-09).\nexport function tmuxSpawnEnv(): NodeJS.ProcessEnv {\n  const extraDirs = ['/opt/homebrew/bin', '/usr/local/bin'];\n  const parts = (process.env.PATH ?? '').split(':').filter(Boolean);\n  const missing = extraDirs.filter((d) => !parts.includes(d));\n  if (missing.length === 0) return process.env;\n  return { ...process.env, PATH: [...parts, ...missing].join(':') };\n}\n\nexport function runTmux(args: string[]): Promise<TmuxRunResult> {\n  return new Promise((resolve) => {\n    const proc = spawn('tmux', args, {\n      stdio: ['ignore', 'ignore', 'pipe'],\n      env: tmuxSpawnEnv(),\n    });\n    let stderr = '';\n    proc.stderr.on('data', (d: Buffer) => {\n      stderr += d.toString();\n    });\n    proc.on('close', (code) => resolve({ code, stderr }));\n    proc.on('error', (err) => resolve({ code: null, stderr: err.message }));\n  });\n}\n\nexport function buildTmuxHasSessionArgs(target: string): string[] {\n  return ['has-session', '-t', target];\n}\n\nexport function buildTmuxSendLiteralArgs(target: string, message: string): string[] {\n  return ['send-keys', '-t', target, '-l', '--', message];\n}\n\nexport function buildTmuxSendEnterArgs(target: string): string[] {\n  return ['send-keys', '-t', target, 'Enter'];\n}\n"],
-  "mappings": ";AAAA,SAAS,aAAa;AAaf,SAAS,eAAkC;AAChD,QAAM,YAAY,CAAC,qBAAqB,gBAAgB;AACxD,QAAM,SAAS,QAAQ,IAAI,QAAQ,IAAI,MAAM,GAAG,EAAE,OAAO,OAAO;AAChE,QAAM,UAAU,UAAU,OAAO,CAAC,MAAM,CAAC,MAAM,SAAS,CAAC,CAAC;AAC1D,MAAI,QAAQ,WAAW,EAAG,QAAO,QAAQ;AACzC,SAAO,EAAE,GAAG,QAAQ,KAAK,MAAM,CAAC,GAAG,OAAO,GAAG,OAAO,EAAE,KAAK,GAAG,EAAE;AAClE;AAEO,SAAS,QAAQ,MAAwC;AAC9D,SAAO,IAAI,QAAQ,CAAC,YAAY;AAC9B,UAAM,OAAO,MAAM,QAAQ,MAAM;AAAA,MAC/B,OAAO,CAAC,UAAU,UAAU,MAAM;AAAA,MAClC,KAAK,aAAa;AAAA,IACpB,CAAC;AACD,QAAI,SAAS;AACb,SAAK,OAAO,GAAG,QAAQ,CAAC,MAAc;AACpC,gBAAU,EAAE,SAAS;AAAA,IACvB,CAAC;AACD,SAAK,GAAG,SAAS,CAAC,SAAS,QAAQ,EAAE,MAAM,OAAO,CAAC,CAAC;AACpD,SAAK,GAAG,SAAS,CAAC,QAAQ,QAAQ,EAAE,MAAM,MAAM,QAAQ,IAAI,QAAQ,CAAC,CAAC;AAAA,EACxE,CAAC;AACH;AAEO,SAAS,wBAAwB,QAA0B;AAChE,SAAO,CAAC,eAAe,MAAM,MAAM;AACrC;AAEO,SAAS,yBAAyB,QAAgB,SAA2B;AAClF,SAAO,CAAC,aAAa,MAAM,QAAQ,MAAM,MAAM,OAAO;AACxD;AAEO,SAAS,uBAAuB,QAA0B;AAC/D,SAAO,CAAC,aAAa,MAAM,QAAQ,OAAO;AAC5C;",
-  "names": []
-}