@moda-ai/cli 1.55.24 → 1.55.25

package/dist/{agent-b4tdzy9q.js → agent-t7dxmpdq.js}

@@ -1,8 +1,8 @@
 import {
   buildDoctorReport
-} from "./cli-vg88mg20.js";
+} from "./cli-mebf5vtr.js";
 import"./cli-0wr3483h.js";
-import"./cli-y2ty9vtv.js";
+import"./cli-m805msp6.js";
 import"./cli-mm48br00.js";
 import {
   HARNESS_REPORT_APPROVAL_PATH,

package/dist/{cli-rt26sbz7.js → cli-m0045npe.js}

@@ -70,6 +70,9 @@ function renderFixesBlock() {
     "- Propose the change in the code, prompt, tool definition, or skill that",
     "  causes the issue, and cite the Moda problem or trace ids it is based on",
     "  in the PR description.",
+    "- Test a prompt change on the Problem's failing traces before it ships (a paid",
+    "  replay; prompt changes only today): `moda prompts ab --problem=<problem_id>",
+    "  --baseline=<prompt_key> --candidate=<edited prompt file>`.",
     "- Prompts live in `.moda/prompts.yml` (version 2: `prompt_paths` globs plus",
     "  `sources:` mappings for prompts defined in code); after `moda prompts sync`,",
     "  pass `prompt: lock.prompts['<key>']` to createThinkTelemetry/createAI7Telemetry.",

package/dist/{cli-y2ty9vtv.js → cli-m805msp6.js}

@@ -17,6 +17,7 @@ import {
   createCommandContext
 } from "./cli-qvg26twp.js";
 import {
+  UUID_RE,
   assertFlags,
   assertPositionals
 } from "./cli-dn91fb56.js";
@@ -5561,6 +5562,9 @@ var DEFAULT_SEEDS_PER_CASE = 3;
 var DEFAULT_AUTO_GENERATE_CASES = 5;
 var MAX_AUTO_GENERATE_CASES = 100;
 var MAX_REPLAY_TRACES = 100;
+var DEFAULT_PROBLEM_MAX_CASES = 10;
+var PROBLEM_PAGE_LIMIT = 50;
+var PROBLEM_MAX_PAGES = 4;
 var PLAYOUT_CONFIRM_THRESHOLD = 200;
 var REPLAY_MODEL_ALLOWLIST = [
   "openai/gpt-5.6-luna",
@@ -5589,7 +5593,7 @@ async function runPromptAb(flags, profileOptions, context) {
   const baselineSource = flags.baseline || flags["baseline-file"] || flags["baseline-key"];
   const candidateSource = flags.candidate || flags["candidate-file"] || flags["candidate-key"];
   if (!baselineSource || !candidateSource) {
-    throw new Error("Usage: moda prompts ab --baseline=<file|key> --candidate=<file|key> [--auto-generate [--cases=1-100]|--set-id=ID|--traces=id1,id2[@msg_index] (max 100)] [--seeds=1-10] [--model=<slug>] [--yes] (legacy alias: --conversations=)");
+    throw new Error("Usage: moda prompts ab --baseline=<file|key> --candidate=<file|key> [--auto-generate [--cases=1-100]|--set-id=ID|--traces=id1,id2[@msg_index] (max 100)|--problem=<problem_id> [--max-cases=1-100]] [--seeds=1-10] [--model=<slug>] [--yes] (legacy alias: --conversations=)");
   }
   assertArmSourceShape(baselineSource, "baseline");
   assertArmSourceShape(candidateSource, "candidate");
@@ -5599,6 +5603,10 @@ async function runPromptAb(flags, profileOptions, context) {
   const pollIntervalMs = parsePositiveInt(flags["poll-interval"], DEFAULT_POLL_INTERVAL_MS, 120000, "--poll-interval");
   const assistantModel = assertAllowedModel(flags.model ?? flags["assistant-model"], flags, "--model");
   const plan = await planReplaySet(flags, tenantId, profileOptions);
+  const problemCases = plan.kind === "traces" ? plan.problem : undefined;
+  if (problemCases && context.outputMode === "human") {
+    process.stderr.write(describeProblemCases(problemCases));
+  }
   const plannedPlayouts = plan.cases === null ? null : plan.cases * seedsPerCase * 2;
   assertPlayoutBudget(plannedPlayouts, { seedsPerCase, cases: plan.cases }, context.yes, rerunWithYes("prompts ab", [], flags));
   if (context.outputMode === "human") {
@@ -5644,6 +5652,7 @@ async function runPromptAb(flags, profileOptions, context) {
     if (noWait) {
       context.output.writeData({
         replaySetId,
+        ...problemCases ? { problemCases } : {},
         ...caseSources ? { caseSources } : {},
         ...warnings.length ? { warnings } : {},
         runId: enqueue.runId,
@@ -5673,6 +5682,7 @@ async function runPromptAb(flags, profileOptions, context) {
     }
     const result = {
       replaySetId,
+      ...problemCases ? { problemCases } : {},
       ...caseSources ? { caseSources } : {},
       ...warnings.length ? { warnings } : {},
       plannedPlayouts,
@@ -5801,6 +5811,9 @@ function resolvePromptArmSpec(source, label, opts = {}) {
   };
 }
 async function planReplaySet(flags, tenantId, profileOptions) {
+  const problemPlan = await planProblemCases(flags);
+  if (problemPlan)
+    return problemPlan;
   const existingSetId = (flags["set-id"] || flags["replay-set-id"] || "").trim();
   if (existingSetId) {
     const snapshot = await fetchReplaySetCases(tenantId, existingSetId, profileOptions);
@@ -5821,6 +5834,151 @@ async function planReplaySet(flags, tenantId, profileOptions) {
     cases: parseCappedInt(flags.cases ?? flags["case-count"], DEFAULT_AUTO_GENERATE_CASES, MAX_AUTO_GENERATE_CASES, "--cases")
   };
 }
+var PROBLEM_CASE_CONFLICTS = [
+  ["traces", "--traces"],
+  ["conversations", "--conversations"],
+  ["conversation-ids", "--conversation-ids"],
+  ["set-id", "--set-id"],
+  ["replay-set-id", "--replay-set-id"],
+  ["cases", "--cases"],
+  ["case-count", "--case-count"]
+];
+async function planProblemCases(flags) {
+  const raw = flags.problem;
+  if (raw === undefined) {
+    if (flags["max-cases"] !== undefined) {
+      throw new CliInputError("--max-cases applies only with --problem, so nothing was sent.", "Use --cases=N with --auto-generate, or pass --problem=<problem_id>.");
+    }
+    return null;
+  }
+  const problemId = raw.trim();
+  const conflicts = PROBLEM_CASE_CONFLICTS.filter(([key]) => flags[key] !== undefined).map(([, name]) => name);
+  if (flags["auto-generate"] === "true")
+    conflicts.push("--auto-generate");
+  if (conflicts.length) {
+    throw new CliInputError(`--problem cannot be combined with ${conflicts.join(", ")}: each picks the replay cases, so nothing was sent.`, "Pass one of --problem=<problem_id> [--max-cases=N], --traces=<ids>, --set-id=<id>, or --auto-generate [--cases=N].");
+  }
+  if (!UUID_RE.test(problemId)) {
+    throw new CliInputError(`--problem needs a canonical problem UUID; got "${problemId}", so nothing was sent.`, "Run `moda problems` to find the problem_id.");
+  }
+  const maxCases = parsePositiveInt(flags["max-cases"], DEFAULT_PROBLEM_MAX_CASES, MAX_REPLAY_TRACES, "--max-cases");
+  const selection = await fetchProblemCases(problemId, maxCases);
+  if (!selection.traces.length) {
+    throw new CliInputError(`Problem ${problemId} has no affected traces to replay, so nothing was sent.`, `Check the id with \`moda problems get ${problemId}\`, or pass --traces=<ids>.`);
+  }
+  return {
+    kind: "traces",
+    traces: selection.traces.map((c) => c.msg_index === null ? c.conversation_id : `${c.conversation_id}@${c.msg_index}`),
+    cases: selection.traces.length,
+    problem: selection
+  };
+}
+async function readProblemPages(problemId, view, enough, drain = false) {
+  const rows = [];
+  let enoughAt = null;
+  let cursor = null;
+  let resolvedId = null;
+  let hasMore = false;
+  for (let page = 0;page < PROBLEM_MAX_PAGES; page++) {
+    const query = new URLSearchParams({ limit: String(PROBLEM_PAGE_LIMIT) });
+    if (cursor)
+      query.set("cursor", cursor);
+    const data = await callDataAPI(`/problems/${encodeURIComponent(problemId)}/${view}?${query.toString()}`);
+    if (typeof data?.problem_id === "string" && data.problem_id)
+      resolvedId = data.problem_id;
+    const pageRows = data?.[view];
+    if (Array.isArray(pageRows))
+      rows.push(...pageRows);
+    const next = data?.pagination?.next_cursor;
+    hasMore = data?.pagination?.has_more === true && typeof next === "string" && next !== "";
+    if (enoughAt === null && enough(rows))
+      enoughAt = rows.length;
+    if (!hasMore || enoughAt !== null && !drain)
+      break;
+    cursor = next;
+  }
+  return { rows, enoughAt, resolvedId, hasMore };
+}
+async function fetchProblemCases(problemId, maxCases) {
+  const anchored = (evidenceRows, traceRows2) => selectProblemCases(evidenceRows, traceRows2, MAX_REPLAY_TRACES).traces.filter((c) => c.msg_index !== null).length;
+  const evidence = await readProblemPages(problemId, "evidence", (rows) => anchored(rows, []) >= maxCases, true);
+  const choiceRows = evidence.enoughAt === null ? evidence.rows : evidence.rows.slice(0, evidence.enoughAt);
+  let traceRows = [];
+  let tracesHaveMore = false;
+  let resolvedId = evidence.resolvedId;
+  if (anchored(choiceRows, []) < maxCases) {
+    const traces2 = await readProblemPages(problemId, "conversations", (rows) => anchored(evidence.rows, rows) >= maxCases);
+    traceRows = traces2.rows;
+    tracesHaveMore = traces2.hasMore;
+    resolvedId ??= traces2.resolvedId;
+  }
+  const selected = selectProblemCases(choiceRows, traceRows, maxCases);
+  const all = selectProblemCases(evidence.rows, traceRows, Number.MAX_SAFE_INTEGER);
+  const earliestByTrace = new Map(all.traces.map((c) => [c.conversation_id, c]));
+  const traces = selected.traces.map((c) => {
+    const first = earliestByTrace.get(c.conversation_id);
+    return first && first.msg_index !== null && (c.msg_index === null || first.msg_index < c.msg_index) ? first : c;
+  });
+  const found = all.found;
+  const resolved = resolvedId ?? problemId;
+  return {
+    problemId: resolved,
+    ...resolved !== problemId ? { requestedProblemId: problemId } : {},
+    maxCases,
+    traces,
+    found,
+    moreAvailable: found > traces.length || evidence.hasMore || tracesHaveMore
+  };
+}
+function anchorIndex(value) {
+  return typeof value === "number" && Number.isSafeInteger(value) && value >= 0 ? value : null;
+}
+function selectProblemCases(evidenceRows, traceRows, maxCases) {
+  const byTrace = new Map;
+  const consider = (conversationId, msgIndex, source, family) => {
+    if (typeof conversationId !== "string" || !conversationId.trim())
+      return;
+    const id = conversationId.trim();
+    const existing = byTrace.get(id);
+    const signalFamily = typeof family === "string" && family ? family : undefined;
+    if (existing && (msgIndex === null || existing.msg_index !== null && existing.msg_index <= msgIndex))
+      return;
+    byTrace.set(id, {
+      conversation_id: id,
+      msg_index: msgIndex,
+      source,
+      ...signalFamily ? { signal_family: signalFamily } : {}
+    });
+  };
+  for (const row of evidenceRows) {
+    const r = row ?? {};
+    consider(r.conversation_id, anchorIndex(r.anchor_msg_index), "evidence", r.signal_family);
+  }
+  for (const row of traceRows) {
+    const r = row ?? {};
+    const anchor = r.latest_anchor ?? {};
+    consider(r.conversation_id, anchorIndex(anchor.anchor_msg_index), "trace", undefined);
+  }
+  const all = [...byTrace.values()];
+  const ordered = [...all.filter((c) => c.msg_index !== null), ...all.filter((c) => c.msg_index === null)];
+  return { traces: ordered.slice(0, maxCases), found: all.length };
+}
+function describeProblemCases(selection) {
+  const requested = selection.requestedProblemId ? ` (requested ${selection.requestedProblemId})` : "";
+  const lines = [
+    `Replaying ${selection.traces.length} trace(s) from Problem ${selection.problemId}${requested}, each from its evidence anchor:`
+  ];
+  for (const c of selection.traces) {
+    const where = c.msg_index === null ? " (no anchor: replays from the first message)" : ` @ msg ${c.msg_index}`;
+    lines.push(`  ${c.conversation_id}${where}${c.signal_family ? ` [${c.signal_family}]` : ""}`);
+  }
+  if (selection.moreAvailable) {
+    lines.push(`Capped at --max-cases=${selection.maxCases}; the Problem has more affected traces.`);
+  }
+  return `${lines.join(`
+`)}
+`;
+}
 async function fetchReplaySetCases(tenantId, setId, profileOptions) {
   try {
     const set = await callControlAPI(`/tenants/${encodeURIComponent(tenantId)}/replay-sets/${encodeURIComponent(setId)}`, { method: "GET" }, profileOptions);
@@ -5876,7 +6034,8 @@ async function ensureReplaySet(plan, flags, tenantId, profileOptions) {
   if (plan.kind === "existing")
     return { setId: plan.setId, warnings: [] };
   if (plan.kind === "traces") {
-    return createSetFromTraces(plan.traces, flags, tenantId, profileOptions);
+    const defaultName = plan.problem ? `Prompt A/B problem ${plan.problem.problemId.slice(0, 8)} (${plan.traces.length} trace(s))` : undefined;
+    return createSetFromTraces(plan.traces, flags, tenantId, profileOptions, defaultName);
   }
   const caseCount = plan.cases;
   const lookbackDays = parsePositiveInt(flags["lookback-days"], 30, 365, "--lookback-days");
@@ -5896,7 +6055,7 @@ function parseTraceRef(ref) {
     return { conversationId: ref.trim() };
   return { conversationId: match[1], startMsgIndex: Number(match[2]) };
 }
-async function createSetFromTraces(conversationIds, flags, tenantId, profileOptions) {
+async function createSetFromTraces(conversationIds, flags, tenantId, profileOptions, defaultName) {
   const refs = [...new Set(conversationIds)].map(parseTraceRef);
   const traces = [...new Set(refs.map((ref) => ref.conversationId))];
   const startByConversation = new Map;
@@ -5904,12 +6063,16 @@ async function createSetFromTraces(conversationIds, flags, tenantId, profileOpti
     if (ref.startMsgIndex !== undefined)
       startByConversation.set(ref.conversationId, ref.startMsgIndex);
   }
-  const name = (flags.name ?? flags["set-name"] ?? `Prompt A/B ${traces.length} trace(s)`).trim();
+  const name = (flags.name ?? flags["set-name"] ?? defaultName ?? `Prompt A/B ${traces.length} trace(s)`).trim();
   const base = `/tenants/${encodeURIComponent(tenantId)}/replay-sets`;
+  const traceAnchors = traces.filter((id) => startByConversation.has(id)).map((id) => ({ conversationId: id, msgIndex: startByConversation.get(id) }));
   let set = null;
   let unavailable = null;
   try {
-    set = await callControlAPI(`${base}/auto-generate`, { method: "POST", body: JSON.stringify({ name, conversationIds: traces }) }, profileOptions, { timeoutMs: 300000, retries: 0 });
+    set = await callControlAPI(`${base}/auto-generate`, {
+      method: "POST",
+      body: JSON.stringify({ name, conversationIds: traces, ...traceAnchors.length ? { traceAnchors } : {} })
+    }, profileOptions, { timeoutMs: 300000, retries: 0 });
     if (!set?.id)
       throw new Error("Auto-generate did not return a replay set id");
   } catch (error) {
@@ -5933,14 +6096,15 @@ async function createSetFromTraces(conversationIds, flags, tenantId, profileOpti
     const update = {};
     if (fallback.length && index !== undefined && existing.position !== index)
       update.position = index;
-    if (startMsgIndex !== undefined)
+    if (startMsgIndex !== undefined && existing.sourceStartMsgIndex !== startMsgIndex) {
       update.sourceStartMsgIndex = startMsgIndex;
+    }
     if (!Object.keys(update).length)
       continue;
     await callControlAPI(`${base}/${encodeURIComponent(set.id)}/cases/${encodeURIComponent(existing.id)}`, { method: "PUT", body: JSON.stringify(update) }, profileOptions);
   }
   for (const conversationId of fallback) {
-    const scenario = await loadScenarioFromConversation(conversationId);
+    const scenario = await loadScenarioFromConversation(conversationId, startByConversation.get(conversationId));
     await callControlAPI(`${base}/${encodeURIComponent(set.id)}/cases`, {
       method: "POST",
       body: JSON.stringify({
@@ -5960,19 +6124,40 @@ async function createSetFromTraces(conversationIds, flags, tenantId, profileOpti
   }
   return { setId: set.id, caseSources: { distilled, fallback }, warnings };
 }
-async function loadScenarioFromConversation(conversationId) {
+var ANCHOR_REQUEST_MAX_READS = 20;
+var CONTEXT_WINDOW = 10;
+async function loadScenarioFromConversation(conversationId, startMsgIndex) {
+  const anchored = startMsgIndex !== undefined && startMsgIndex > 0;
+  const userRequest = (messages, atOrBefore) => messages.filter((message) => message.role === "user" && String(message.content ?? "").trim() && (atOrBefore === undefined || typeof message.index === "number" && message.index <= atOrBefore));
+  const readContext = async (msgIndex, window2) => {
+    const data = await callDataAPI(`/conversations/${encodeURIComponent(conversationId)}/context?msg_index=${msgIndex}&window=${window2}`);
+    return data.context?.messages ?? data.messages ?? [];
+  };
   try {
-    const data = await callDataAPI(`/conversations/${encodeURIComponent(conversationId)}/context?msg_index=0&window=5`);
-    const messages = data.context?.messages ?? data.messages ?? [];
-    const firstUser = messages.find((message) => message.role === "user" && String(message.content ?? "").trim());
-    const opening = String(firstUser?.content ?? "").trim();
-    if (opening) {
-      return `The user says: "${opening}"
+    if (anchored) {
+      let center = startMsgIndex;
+      for (let read = 0;read < ANCHOR_REQUEST_MAX_READS; read++) {
+        const lead = userRequest(await readContext(center, CONTEXT_WINDOW), startMsgIndex).pop();
+        const request = String(lead?.content ?? "").trim();
+        if (request) {
+          return `Earlier in the conversation the user and the agent already talked; the replay keeps those turns as history. ` + `Now the user says: "${request}"
+
+Help them with this request using the available tools.`;
+        }
+        if (center - CONTEXT_WINDOW <= 0)
+          break;
+        center = Math.max(0, center - (2 * CONTEXT_WINDOW + 1));
+      }
+    } else {
+      const request = String(userRequest(await readContext(0, 5))[0]?.content ?? "").trim();
+      if (request) {
+        return `The user says: "${request}"
 
 Replay the conversation faithfully and help them accomplish their goal.`;
+      }
     }
   } catch {}
-  return `Replay conversation ${conversationId}: help the user accomplish their original goal using the available tools.`;
+  return anchored ? `Replay conversation ${conversationId} from message ${startMsgIndex}: help the user with their request at that point using the available tools.` : `Replay conversation ${conversationId}: help the user accomplish their original goal using the available tools.`;
 }
 async function fetchLatestRun(tenantId, replaySetId, runId, profileOptions) {
   return callControlAPI(`/tenants/${encodeURIComponent(tenantId)}/replay-sets/${encodeURIComponent(replaySetId)}/runs/latest?runId=${encodeURIComponent(runId)}`, { method: "GET" }, profileOptions, { timeoutMs: 60000, retries: 2 });
@@ -6505,7 +6690,9 @@ var PROMPT_WRITE_SUBCOMMAND_FLAGS = {
     "set-id",
     "set-name",
     "tenant-id",
-    "allow-any-model"
+    "allow-any-model",
+    "problem",
+    "max-cases"
   ],
   propose: [
     "gate",

package/dist/{cli-vg88mg20.js → cli-mebf5vtr.js}

@@ -4,7 +4,7 @@ import {
 } from "./cli-0wr3483h.js";
 import {
   inspectPromptManifest
-} from "./cli-y2ty9vtv.js";
+} from "./cli-m805msp6.js";
 import {
   readHarnessSyncState,
   renderDoctorHuman,

package/dist/cli.js

@@ -2522,31 +2522,31 @@ function leafManifestFields(leaf, commands) {
 import { fileURLToPath } from "url";
 import { realpathSync } from "fs";
 async function runPromptsCommand(...args) {
-  return (await import("./prompts-cvzt994j.js")).runPromptsCommand(...args);
+  return (await import("./prompts-3h5jcjrw.js")).runPromptsCommand(...args);
 }
 async function runSkillsCommand(...args) {
-  return (await import("./skills-tadh1jss.js")).runSkillsCommand(...args);
+  return (await import("./skills-vq3njc8a.js")).runSkillsCommand(...args);
 }
 async function runFixCommand(...args) {
-  return (await import("./fix-exvxzvtz.js")).runFixCommand(...args);
+  return (await import("./fix-0qdtdz5h.js")).runFixCommand(...args);
 }
 async function runFixesCommand(...args) {
-  return (await import("./fix-exvxzvtz.js")).runFixesCommand(...args);
+  return (await import("./fix-0qdtdz5h.js")).runFixesCommand(...args);
 }
 async function runHarnessCommand(...args) {
   return (await import("./harness-vzy7n7ea.js")).runHarnessCommand(...args);
 }
 async function runSyncCommand(...args) {
-  return (await import("./sync-g8wtecz1.js")).runSyncCommand(...args);
+  return (await import("./sync-hq9dpqpr.js")).runSyncCommand(...args);
 }
 async function runRegistryCommand(...args) {
   return (await import("./registry-s2vdsfw8.js")).runRegistryCommand(...args);
 }
 async function runDoctorCommand(...args) {
-  return (await import("./doctor-3r19ry92.js")).runDoctorCommand(...args);
+  return (await import("./doctor-q21mr5ya.js")).runDoctorCommand(...args);
 }
 async function runStatusCommand(...args) {
-  return (await import("./doctor-3r19ry92.js")).runStatusCommand(...args);
+  return (await import("./doctor-q21mr5ya.js")).runStatusCommand(...args);
 }
 async function runAuthCommand(...args) {
   return (await import("./profile-commands-6e2dgv72.js")).runAuthCommand(...args);
@@ -2558,7 +2558,7 @@ async function runProfilesCommand(...args) {
   return (await import("./profile-commands-6e2dgv72.js")).runProfilesCommand(...args);
 }
 async function runAgentCommand(...args) {
-  return (await import("./agent-b4tdzy9q.js")).runAgentCommand(...args);
+  return (await import("./agent-t7dxmpdq.js")).runAgentCommand(...args);
 }
 async function runProductionCommand(...args) {
   return (await import("./production-intelligence-c8e0xw3n.js")).runProductionCommand(...args);
@@ -5382,7 +5382,7 @@ var commandRegistry = createCommandRegistry([
     resetApiRequestCountBeforeRun: true,
     telemetry: "result-and-error",
     handler: async (context) => {
-      const { runInit } = await import("./index-e1pf1qz2.js");
+      const { runInit } = await import("./index-4b1ybz51.js");
       if (context.outputMode === "agent-stream") {
         context.output.writeEvent({
           event: "started",
@@ -6259,11 +6259,13 @@ var commandRegistry = createCommandRegistry([
       {
         name: "ab",
         description: "Run a judged prompt A/B replay experiment",
-        usage: "moda prompts ab --baseline=<path> --candidate=<path> [--traces=<ids>|--set-id=<id>|--auto-generate] [--yes]",
+        usage: "moda prompts ab --baseline=<path> --candidate=<path> [--traces=<ids>|--problem=<problem_id> [--max-cases=<n>]|--set-id=<id>|--auto-generate] [--yes]",
         flags: [
           { name: "--baseline=<path>", description: "Prompt file to treat as the control." },
           { name: "--candidate=<path>", description: "Prompt file to treat as the variant." },
           { name: "--traces=<ids>", description: "Comma-separated trace ids (conversation_id values) to replay, at most 100. Each case uses Moda's distilled scenario and success criteria for the trace; a trace that cannot be distilled falls back to its opening user message with generic criteria, and the run warns. Legacy alias: --conversations=<ids>." },
+          { name: "--problem=<problem_id>", description: "Replay the Problem's failing traces: one case per affected trace, starting at its evidence msg_index (like --traces=<id>@<n>), read from `moda problems get <id> --evidence`, then `--traces` if needed. Prints the traces it chose. Not combinable with --traces, --set-id, --auto-generate, or --cases." },
+          { name: "--max-cases=<n>", description: "With --problem: replay at most this many traces (default 10, max 100)." },
           { name: "--cases=<n>", description: "Cases to auto-generate (default 5, max 100)." },
           { name: "--seeds=<n>", description: "Repeats per case per arm (default 3, max 10)." },
           { name: "--model=<slug>", description: "Assistant model for both arms; must be in the replay allowlist unless --allow-any-model." },
@@ -6271,7 +6273,8 @@ var commandRegistry = createCommandRegistry([
           { name: "--promote", description: "Make this run's replay set the tenant's primary replay set. Off by default, so an ad-hoc A/B never replaces it." }
         ],
         examples: [
-          "moda prompts ab --baseline=prompts/agent.prompt.md --candidate=prompts/agent-v2.prompt.md --traces=conv_1,conv_2"
+          "moda prompts ab --baseline=prompts/agent.prompt.md --candidate=prompts/agent-v2.prompt.md --traces=conv_1,conv_2",
+          "moda prompts ab --baseline=prompts/agent.prompt.md --candidate=prompts/agent-v2.prompt.md --problem=<problem_id>"
         ]
       },
       {

package/dist/{doctor-3r19ry92.js → doctor-q21mr5ya.js}

@@ -3,9 +3,9 @@ import {
   promptsDiscoveredCheck,
   runDoctorCommand,
   runStatusCommand
-} from "./cli-vg88mg20.js";
+} from "./cli-mebf5vtr.js";
 import"./cli-0wr3483h.js";
-import"./cli-y2ty9vtv.js";
+import"./cli-m805msp6.js";
 import"./cli-mm48br00.js";
 import"./cli-7xc585hc.js";
 import"./cli-97fnztpy.js";

package/dist/{fix-exvxzvtz.js → fix-0qdtdz5h.js}

@@ -2,7 +2,7 @@ import {
   DEFAULT_POLL_INTERVAL_MS,
   DEFAULT_WAIT_TIMEOUT_MS,
   fetchLatestRun
-} from "./cli-y2ty9vtv.js";
+} from "./cli-m805msp6.js";
 import"./cli-mm48br00.js";
 import"./cli-7xc585hc.js";
 import"./cli-97fnztpy.js";

package/dist/{index-e1pf1qz2.js → index-4b1ybz51.js}

@@ -4,7 +4,7 @@ import {
 import"./cli-9hc5bg2r.js";
 import {
   buildDoctorReport
-} from "./cli-vg88mg20.js";
+} from "./cli-mebf5vtr.js";
 import {
   ConnectCodeError,
   connectWithCode,
@@ -15,7 +15,7 @@ import {
 import {
   initPrompts,
   runPromptSync
-} from "./cli-y2ty9vtv.js";
+} from "./cli-m805msp6.js";
 import"./cli-mm48br00.js";
 import {
   canonicalText,
@@ -47,7 +47,7 @@ import {
 import {
   agentsMdProblem,
   runSkillSync
-} from "./cli-rt26sbz7.js";
+} from "./cli-m0045npe.js";
 import {
   createCommandContext
 } from "./cli-qvg26twp.js";

package/dist/{prompts-cvzt994j.js → prompts-3h5jcjrw.js}

@@ -11,7 +11,7 @@ import {
   runPromptSync,
   runPromptsCommand,
   writePromptSyncProgress
-} from "./cli-y2ty9vtv.js";
+} from "./cli-m805msp6.js";
 import {
   promptSourceHashPayload
 } from "./cli-mm48br00.js";

package/dist/{skills-tadh1jss.js → skills-vq3njc8a.js}

@@ -3,7 +3,7 @@ import {
   pullSkills,
   runSkillSync,
   runSkillsCommand
-} from "./cli-rt26sbz7.js";
+} from "./cli-m0045npe.js";
 import"./cli-qvg26twp.js";
 import"./cli-dn91fb56.js";
 import"./cli-9btx12wp.js";

package/dist/{sync-g8wtecz1.js → sync-hq9dpqpr.js}

@@ -1,6 +1,6 @@
 import {
   runPromptsCommand
-} from "./cli-y2ty9vtv.js";
+} from "./cli-m805msp6.js";
 import"./cli-mm48br00.js";
 import {
   runHarnessCommand
@@ -8,7 +8,7 @@ import {
 import"./cli-97fnztpy.js";
 import {
   runSkillsCommand
-} from "./cli-rt26sbz7.js";
+} from "./cli-m0045npe.js";
 import"./cli-qvg26twp.js";
 import"./cli-dn91fb56.js";
 import"./cli-9btx12wp.js";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@moda-ai/cli",
-  "version": "1.55.24",
+  "version": "1.55.25",
   "description": "CLI for Moda: production monitoring and evals for AI agents.",
   "type": "module",
   "bin": {

package/skills/integration/index.json

@@ -1,7 +1,7 @@
 {
   "schema_version": "moda.skill_index.v1",
-  "bundled_at": "2026-10-15T13:10:26.478Z",
-  "cli_version": "1.55.24",
+  "bundled_at": "2026-10-15T20:30:38.730Z",
+  "cli_version": "1.55.25",
   "skills": [
     {
       "id": "integration-cloudflare-think",

package/skills/moda-cli/SKILL.md

@@ -82,7 +82,10 @@ on search, traces, and clusters. `moda <noun> <verb> --help` shows which.
 2. `moda problems get <problem_id> --evidence` and open two or three anchors with
    `moda traces get <id> --msg-index=<n>` before deciding on a cause.
 3. Make the change in code; cite the problem id and trace ids in the commit or PR.
-4. After it ships, and only if the user agrees:
+4. Test a prompt change on the Problem's failing traces (a paid replay, so ask
+   first; prompt changes only today):
+   `moda prompts ab --problem=<problem_id> --baseline=<prompt_key> --candidate=<edited file>`.
+5. After it ships, and only if the user agrees:
    `moda problems feedback <problem_id> --action=mark_fixed`.
 More: `references/production-intelligence.md`.
 

package/skills/moda-cli/references/production-intelligence.md

@@ -90,7 +90,8 @@ Rules that keep the fix grounded:
 - Cite the Moda problem id and the trace ids you relied on in the PR
   description, so a reviewer can open the same evidence.
 - To test a prompt change against real traffic before shipping it, A/B it with
-  `moda prompts ab --baseline=<path> --candidate=<path> --traces=<ids>` (see
-  section 7).
+  `moda prompts ab --baseline=<path> --candidate=<path> --problem=<problem_id>`
+  (replays the Problem's failing traces from their evidence anchors) or
+  `--traces=<ids>` (see section 7).
 - Mark the Problem fixed only after the change ships. If the issue is detected
   again, the Problem comes back; `moda problems reopen` reopens it by hand.

package/skills/moda-cli/references/prompts.md

@@ -72,7 +72,10 @@ read, they fail (exit 1) with the number and the exact command to re-run with `-
 Only add `--yes` when the user has agreed to that spend. Limits: `--cases`
 default 5, max 100; `--seeds` default 3, max 10; `--traces` at most 100
 (`--traces=conv_id@412` replays that trace from msg_index 412, with the
-earlier turns as history);
+earlier turns as history, and writes its case from the messages around 412);
+`--problem=<problem_id>` picks a Problem's affected traces, each anchored at
+its earliest evidence msg_index, up to
+`--max-cases` (default 10, max 100), and prints them (`problemCases`);
 `--model` / `--assistant-model` must be one of `openai/gpt-5.6-luna`,
 `openai/gpt-4o-mini`, `openai/gpt-4o`, `anthropic/claude-sonnet-4-5`,
 `anthropic/claude-haiku-4-5`, `google/gemini-2.5-flash`,
@@ -80,7 +83,8 @@ earlier turns as history);
 
 `prompts ab --traces=<ids>` builds one case per trace from Moda's distilled
 scenario and success criteria for that trace. A trace Moda cannot distill
-falls back to its opening user message with three generic criteria (which long
+falls back to its opening user message (for `id@N`, the user message leading
+into N) with three generic criteria (which long
 traces pass easily); the run warns and lists it under `caseSources.fallback`.
 `prompts ab` never makes its replay set the tenant's primary set unless you
 pass `--promote`.