@pome-sh/cli 0.42.4 → 0.43.1

package/dist/runDemo-DRDMCG25.js

@@ -1,408 +0,0 @@
-import { newGroupId } from './chunk-JNXZBK3O.js';
-import { DemoCapacityError, capacityLabel, parseCapacityMarker, capacityKindFrom } from './chunk-ZX4WNSZ5.js';
-import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-WERG4AUH.js';
-import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, isNarrated, outcomeOf, criterionPhrase, narratorReadingLines } from './chunk-STMGQ7GU.js';
-import './chunk-NW7HGA2K.js';
-import './chunk-PASFBRK4.js';
-import './chunk-3FZY376K.js';
-import { bootTwin } from './chunk-5NPGY73F.js';
-import './chunk-RPPF2KDQ.js';
-import { getAvailablePort } from './chunk-XDU6TD4O.js';
-import { HostedQuotaError, HostedOrchError } from './chunk-NNXVR46L.js';
-import './chunk-CBFKZZBR.js';
-import './chunk-NBOQN5VX.js';
-import './chunk-YBWG5JK2.js';
-import './chunk-HRAD7MRX.js';
-import './chunk-6KJC4BTO.js';
-import './chunk-SG6ZTIMT.js';
-import './chunk-2K6BJ3PI.js';
-import './chunk-FBSA5L36.js';
-import { readFile } from 'node:fs/promises';
-import { join } from 'node:path';
-import { z } from 'zod';
-
-var demoSessionSchema = z.object({
-  session_id: z.string().min(1),
-  demo_token: z.string().min(1),
-  expires_at: z.string()
-});
-async function mintDemoSessions(options) {
-  const sessions = [];
-  for (let i = 0; i < options.count; i += 1) {
-    sessions.push(await mintOne(options));
-  }
-  return sessions;
-}
-async function mintOne(options) {
-  const doFetch = options.fetchImpl ?? fetch;
-  const ctrl = new AbortController();
-  const timer = setTimeout(() => ctrl.abort(), options.timeoutMs ?? 3e4);
-  let res;
-  try {
-    res = await doFetch(`${options.apiBase}/v1/demo/sessions`, {
-      method: "POST",
-      headers: { "content-type": "application/json" },
-      body: JSON.stringify({
-        task_name: options.taskName,
-        task_hash: "",
-        group_id: options.groupId
-      }),
-      signal: ctrl.signal
-    });
-  } catch (err) {
-    throw new HostedOrchError(
-      err instanceof Error ? err.message : "network error"
-    );
-  } finally {
-    clearTimeout(timer);
-  }
-  const text = await res.text();
-  let json = {};
-  try {
-    json = text.length ? JSON.parse(text) : {};
-  } catch {
-    throw new HostedOrchError(
-      `POST /v1/demo/sessions returned non-JSON (status ${res.status})`
-    );
-  }
-  if (!res.ok) {
-    const envelope = json.error;
-    let kind = capacityKindFrom(res.status, envelope?.details?.kind);
-    if (kind === "unknown_capacity") {
-      kind = res.status === 429 ? "demo_ip_mint_cap" : "demo_mint_quota";
-    }
-    if (kind) {
-      throw new DemoCapacityError(kind, envelope?.message ?? capacityLabel(kind));
-    }
-    throw new HostedOrchError(
-      envelope?.message ?? `POST /v1/demo/sessions \u2192 HTTP ${res.status}`,
-      void 0,
-      res.status
-    );
-  }
-  const parsed = demoSessionSchema.safeParse(json);
-  if (!parsed.success) {
-    throw new HostedOrchError(
-      `POST /v1/demo/sessions returned an unexpected shape: ${parsed.error.message}`
-    );
-  }
-  return parsed.data;
-}
-
-// src/demo/render.ts
-function reassuranceBox() {
-  const lines = [
-    "pome demo \xB7 running locally",
-    "No signup. No API keys.",
-    "The GitHub twin runs on your machine; model calls go through pome's",
-    "anonymous demo gateway; the trace is evaluated in pome cloud.",
-    "Your repo and your data are never touched."
-  ];
-  const width = Math.max(...lines.map((line) => line.length));
-  const top = `\u250C${"\u2500".repeat(width + 2)}\u2510`;
-  const bottom = `\u2514${"\u2500".repeat(width + 2)}\u2518`;
-  return [top, ...lines.map((line) => `\u2502 ${line.padEnd(width)} \u2502`), bottom];
-}
-function twinReadyLine(seconds) {
-  return `spinning up github twin \u2026 ready (${seconds.toFixed(1)}s)`;
-}
-function trialsHeaderLine(count, taskName) {
-  return `running ${count} isolated trials of ${taskName} \u2026`;
-}
-function evaluatingLine(index) {
-  return `trial ${index}  \u2026  evaluating in the cloud \u2026`;
-}
-function trialLine(index, verdict) {
-  switch (verdict.kind) {
-    case "passed":
-      return `trial ${index}  \u2713  passed   ${formatSeconds(verdict.seconds)}  ${checksPhrase(verdict.checks)}`;
-    case "failed":
-      return `trial ${index}  \u2717  failed   ${formatSeconds(verdict.seconds)}  ${checksPhrase(verdict.checks)}  ${verdict.note}`;
-    case "errored":
-      return `trial ${index}  \u26A0  errored         ${verdict.reason} \u2014 excluded`;
-  }
-}
-function checksPhrase(checks) {
-  return `${checks.passed} of ${checks.total} checks`;
-}
-function formatSeconds(seconds) {
-  return `${seconds.toFixed(1)}s`;
-}
-function summaryLines(input) {
-  const passed = input.verdicts.filter((v) => v.kind === "passed").length;
-  const failed = input.verdicts.filter((v) => v.kind === "failed").length;
-  const errored = input.verdicts.filter(
-    (v) => v.kind === "errored"
-  );
-  const denominator = passed + failed;
-  const lines = ["\u2500\u2500\u2500\u2500\u2500"];
-  let fraction;
-  if (denominator === 0) {
-    fraction = "no trials were evaluated";
-  } else {
-    fraction = `${passed} of ${denominator} passed`;
-  }
-  if (errored.length > 0) {
-    const reason = errored[0].reason;
-    const noun = errored.length === 1 ? "trial" : "trials";
-    fraction += ` \xB7 ${errored.length} ${noun} errored on ${reason}, excluded from the fraction`;
-  }
-  lines.push(fraction);
-  lines.push(...narratorReadingLines(input.narrated ?? []));
-  if (input.failingCriterionPhrase && typeof input.failingCriterionCount === "number" && input.failingCriterionCount > 0 && denominator > 0) {
-    lines.push(
-      `${input.failingCriterionPhrase} in ${input.failingCriterionCount} of ${denominator} \u2014 start there`
-    );
-  }
-  if (denominator > 0) {
-    lines.push("see the full breakdown \u2014 read-only, still no account:");
-    lines.push(`\u2192 ${input.previewUrl}`);
-  }
-  return lines;
-}
-
-// src/demo/runDemo.ts
-var DEMO_FINALIZE_TIMEOUT_MS = 6e4;
-async function runDemo(options) {
-  const out = options.out ?? ((line) => console.error(line));
-  const trials = options.trials ?? 5;
-  const artifactsDir = options.artifactsDir ?? "runs";
-  const runTaskFn = options.runTaskFn ?? runTask;
-  const mintFn = options.mintFn ?? mintDemoSessions;
-  const trialClientFactory = options.trialClientFactory ?? ((session) => createHostedClient({
-    baseUrl: options.apiBase,
-    apiKey: session.demo_token,
-    authScheme: "bearer",
-    timeoutMs: DEMO_FINALIZE_TIMEOUT_MS,
-    finalizeTimeoutMs: DEMO_FINALIZE_TIMEOUT_MS
-  }));
-  const groupId = newGroupId();
-  const taskPath = demoTaskPath();
-  for (const line of reassuranceBox()) out(line);
-  out("");
-  let sessions;
-  try {
-    sessions = await mintFn({
-      apiBase: options.apiBase,
-      taskName: DEMO_TASK_NAME,
-      groupId,
-      count: trials
-    });
-  } catch (err) {
-    if (err instanceof DemoCapacityError) {
-      out(capacityLabel(err.kind));
-      return { exitCode: 4, groupId, verdicts: [] };
-    }
-    out(
-      `could not reach the pome demo service (${err instanceof Error ? err.message : String(err)})`
-    );
-    return { exitCode: 2, groupId, verdicts: [] };
-  }
-  if (options.skipTwinWarmup !== true) {
-    const scenario = await parseTaskFile(taskPath);
-    const warmupStart = Date.now();
-    const port = await getAvailablePort();
-    const harness = await bootTwin({
-      twin: "github",
-      seedState: scenario.seedState,
-      runId: "demo-warmup",
-      twinBaseUrl: `http://127.0.0.1:${port}`
-    });
-    await harness.close();
-    out(twinReadyLine((Date.now() - warmupStart) / 1e3));
-  }
-  out(trialsHeaderLine(trials, DEMO_TASK_NAME));
-  const agentCommand = options.agentCommand ?? defaultDemoAgentCommand(process.execPath, process.argv[1]);
-  const verdicts = [];
-  const collectedFailures = [];
-  const collectedReadings = [];
-  let capacityAbort = null;
-  for (let i = 0; i < trials; i += 1) {
-    const session = sessions[i];
-    const trialNumber = i + 1;
-    const client = trialClientFactory(session);
-    const progress = { finalized: false };
-    let verdict;
-    let abortCode = null;
-    try {
-      verdict = await runOneTrial({
-        trialNumber,
-        session,
-        taskPath,
-        agentCommand,
-        artifactsDir,
-        apiBase: options.apiBase,
-        runTaskFn,
-        client,
-        captureServerCommand: options.captureServerCommand,
-        out,
-        collectedFailures,
-        collectedReadings,
-        progress
-      });
-    } catch (err) {
-      if (err instanceof DemoCapacityError) {
-        abortCode = err.kind;
-        capacityAbort = capacityLabel(err.kind);
-        verdict = { kind: "errored", reason: "demo at capacity" };
-      } else if (isJudgeCapQuota(err)) {
-        abortCode = "daily_judge_cap";
-        capacityAbort = capacityLabel("daily_judge_cap");
-        verdict = { kind: "errored", reason: "demo at capacity" };
-      } else {
-        verdict = {
-          kind: "errored",
-          reason: shortReason(err instanceof Error ? err.message : String(err))
-        };
-      }
-      if (!progress.finalized) {
-        await abandonQuietly(client, session.session_id, abortCode ?? "trial_crashed");
-      }
-    }
-    verdicts.push(verdict);
-    out(trialLine(trialNumber, verdict));
-    if (capacityAbort) {
-      for (let j = i + 1; j < trials; j += 1) {
-        const orphan = sessions[j];
-        await abandonQuietly(
-          trialClientFactory(orphan),
-          orphan.session_id,
-          abortCode ?? "unknown_capacity"
-        );
-      }
-      break;
-    }
-  }
-  if (capacityAbort) out(capacityAbort);
-  const failing = mostCommonFailedCriterion(collectedFailures);
-  out("");
-  for (const line of summaryLines({
-    verdicts,
-    failingCriterionPhrase: failing?.phrase,
-    failingCriterionCount: failing?.count,
-    narrated: collectedReadings,
-    previewUrl: `${options.dashboardBase.replace(/\/$/, "")}/demo/${groupId}`
-  })) {
-    out(line);
-  }
-  const evaluated = verdicts.filter((v) => v.kind !== "errored").length;
-  const exitCode = evaluated > 0 ? 0 : capacityAbort ? 4 : 1;
-  return { exitCode, groupId, verdicts };
-}
-async function runOneTrial(input) {
-  const startedAt = Date.now();
-  const result = await input.runTaskFn({
-    taskPath: input.taskPath,
-    agentCommand: input.agentCommand,
-    artifactsDir: input.artifactsDir,
-    captureServerCommand: input.captureServerCommand,
-    extraAgentEnv: {
-      POME_DEMO_LLM_URL: `${input.apiBase}/v1/demo/sessions/${input.session.session_id}/llm`,
-      POME_DEMO_TOKEN: input.session.demo_token,
-      POME_DEMO_TASK_NAME: DEMO_TASK_NAME,
-      POME_DEMO_REPO: DEMO_REPO
-    },
-    egressExtraHosts: [new URL(input.apiBase).hostname]
-  });
-  const agentSeconds = (Date.now() - startedAt) / 1e3;
-  if (result.exitCode !== 0) {
-    const capacityKind = parseCapacityMarker(result.agent.stderr);
-    if (capacityKind) {
-      throw new DemoCapacityError(capacityKind, capacityLabel(capacityKind));
-    }
-    if (result.agent.timedOut) {
-      await abandonQuietly(input.client, input.session.session_id, "agent_timeout");
-      return { kind: "errored", reason: "trial timed out" };
-    }
-    await abandonQuietly(input.client, input.session.session_id, "agent_exit_nonzero");
-    return {
-      kind: "errored",
-      reason: shortReason(lastLine(result.agent.stderr) ?? "agent failed")
-    };
-  }
-  input.out(evaluatingLine(input.trialNumber));
-  const runDir = result.artifacts.runDir;
-  const [eventsJsonl, stateInitialJson, stateFinalJson, metaJson] = await Promise.all([
-    readFile(join(runDir, "events.jsonl"), "utf8"),
-    readFile(join(runDir, "state_initial.json"), "utf8"),
-    readFile(join(runDir, "state_final.json"), "utf8"),
-    // D18.1 — best-effort: a demo trial's verdict must never hinge on the
-    // meta.json sidecar being readable.
-    readFile(join(runDir, "meta.json"), "utf8").catch(() => "{}")
-  ]);
-  const uploaded = await uploadRunBlobs(input.client, input.session.session_id, {
-    eventsJsonl,
-    stateInitialJson,
-    stateFinalJson,
-    signalsJsonl: "",
-    metaJson
-  });
-  const finalized = await input.client.finalize(input.session.session_id, {
-    stopReason: "completed",
-    exitCode: 0,
-    durationMs: Math.max(0, Math.round(agentSeconds * 1e3)),
-    agentModel: "demo-gateway",
-    agentSdk: null,
-    criteria: [],
-    taskName: DEMO_TASK_NAME,
-    taskHash: "",
-    taskPrompt: "",
-    expectedBehavior: "",
-    traceStorageKey: uploaded.eventsKey ?? void 0,
-    stateInitialStorageKey: uploaded.stateInitialKey ?? void 0,
-    stateFinalStorageKey: uploaded.stateFinalKey ?? void 0
-  });
-  input.progress.finalized = true;
-  const score = scoreFromFinalizeResponse(finalized);
-  const status = scoreStatus(score, 100);
-  const checks = { passed: score.passed, total: score.total_required };
-  if (status !== "incomplete") {
-    input.collectedReadings.push(...score.results.filter(isNarrated));
-  }
-  if (status === "pass") {
-    return { kind: "passed", seconds: agentSeconds, checks };
-  }
-  if (status === "fail") {
-    const failedResults = score.results.filter((r) => outcomeOf(r) === "failed");
-    for (const r of failedResults) input.collectedFailures.push(r.criterion.text);
-    const note = failedResults[0] ? criterionPhrase(failedResults[0].criterion.text) : "criterion not met";
-    return { kind: "failed", seconds: agentSeconds, note, checks };
-  }
-  return { kind: "errored", reason: "cloud could not evaluate the trace" };
-}
-async function abandonQuietly(client, sessionId, errorCode) {
-  try {
-    await client.abandonSession(sessionId, { errorCode });
-  } catch {
-  }
-}
-function isJudgeCapQuota(err) {
-  return err instanceof HostedQuotaError && capacityKindFrom(402, err.details?.kind) === "daily_judge_cap";
-}
-function mostCommonFailedCriterion(failures) {
-  if (failures.length === 0) return null;
-  const counts = /* @__PURE__ */ new Map();
-  for (const text2 of failures) {
-    counts.set(text2, (counts.get(text2) ?? 0) + 1);
-  }
-  const [text, count] = [...counts.entries()].sort((a, b) => b[1] - a[1])[0];
-  return { phrase: criterionPhrase(text), count };
-}
-function lastLine(text) {
-  const lines = text.split("\n").map((line) => line.trim()).filter((line) => line.length > 0);
-  return lines.length > 0 ? lines[lines.length - 1] : null;
-}
-function shortReason(reason) {
-  const flat = reason.replace(/\s+/g, " ").trim();
-  return flat.length > 72 ? `${flat.slice(0, 69)}\u2026` : flat;
-}
-function defaultDemoAgentCommand(execPath, scriptPath) {
-  if (!scriptPath || scriptPath.length === 0) {
-    throw new Error(
-      "Cannot determine the pome binary path for the demo agent (process.argv[1] is empty)."
-    );
-  }
-  return `"${execPath}" "${scriptPath}" demo-agent`;
-}
-
-export { defaultDemoAgentCommand, runDemo };

package/dist/twinHarness-ATYQTDA5.js

@@ -1,6 +0,0 @@
-export { UnsupportedTwinError, bootTwin } from './chunk-5NPGY73F.js';
-export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-RPPF2KDQ.js';
-import './chunk-HRAD7MRX.js';
-import './chunk-6KJC4BTO.js';
-import './chunk-SG6ZTIMT.js';
-import './chunk-FBSA5L36.js';