@moda-ai/cli 1.40.0 → 1.41.1

package/dist/{cli-w9an3j54.js → cli-pr6s56fc.js}

@@ -5595,7 +5595,7 @@ 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 (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)] [--seeds=1-10] [--model=<slug>] [--yes] (legacy alias: --conversations=)");
   }
   assertArmSourceShape(baselineSource, "baseline");
   assertArmSourceShape(candidateSource, "candidate");
@@ -5877,6 +5877,12 @@ async function ensureReplaySet(plan, flags, tenantId, profileOptions) {
   }
   return generated.id;
 }
+function parseTraceRef(ref) {
+  const match = /^(.+)@(\d+)$/.exec(ref.trim());
+  if (!match)
+    return { conversationId: ref.trim() };
+  return { conversationId: match[1], startMsgIndex: Number(match[2]) };
+}
 async function createSetFromConversations(conversationIds, flags, tenantId, profileOptions) {
   const name = (flags.name ?? flags["set-name"] ?? `Prompt A/B ${conversationIds.length} trace(s)`).trim();
   const created = await callControlAPI(`/tenants/${encodeURIComponent(tenantId)}/replay-sets`, {
@@ -5884,7 +5890,8 @@ async function createSetFromConversations(conversationIds, flags, tenantId, prof
     body: JSON.stringify({ name, description: "Created by moda prompts ab" })
   }, profileOptions);
   let position = 0;
-  for (const conversationId of conversationIds) {
+  for (const ref of conversationIds) {
+    const { conversationId, startMsgIndex } = parseTraceRef(ref);
     const scenario = await loadScenarioFromConversation(conversationId);
     await callControlAPI(`/tenants/${encodeURIComponent(tenantId)}/replay-sets/${encodeURIComponent(created.id)}/cases`, {
       method: "POST",
@@ -5892,6 +5899,7 @@ async function createSetFromConversations(conversationIds, flags, tenantId, prof
         title: conversationId,
         scenario,
         sourceConversationId: conversationId,
+        ...startMsgIndex !== undefined ? { sourceStartMsgIndex: startMsgIndex } : {},
         successCriteria: [
           "The agent understands the user request",
           "The agent uses tools appropriately when needed",

package/dist/cli.js

@@ -8,7 +8,7 @@ import {
   runPromptsCommand,
   runSkillsCommand,
   runStatusCommand
-} from "./cli-w9an3j54.js";
+} from "./cli-pr6s56fc.js";
 import {
   ApiError,
   HARNESS_REPORT_APPROVAL_PATH,
@@ -1799,7 +1799,8 @@ function gateFindings(fix) {
 function formatGateScoreboard(gate) {
   const holdout = gate.holdout ?? {};
   const repair = gate.repair ?? {};
-  return `holdout: candidate ${holdout.candidate ?? 0}/${holdout.n ?? 0} vs baseline ${holdout.baseline ?? 0}/${holdout.n ?? 0} ` + `(wins ${holdout.wins ?? 0}, losses ${holdout.losses ?? 0}); ` + `training cases (repair): candidate ${repair.candidate ?? 0}/${repair.n ?? 0} vs baseline ${repair.baseline ?? 0}/${repair.n ?? 0}; ` + `noise floor ${gate.noiseFloor ?? 0}; abstained ${gate.abstained ?? 0}`;
+  const guard = gate.byScope?.regression_guard;
+  return `${gate.byScope ? "in-problem holdout" : "holdout"}: candidate ${holdout.candidate ?? 0}/${holdout.n ?? 0} vs baseline ${holdout.baseline ?? 0}/${holdout.n ?? 0} ` + `(wins ${holdout.wins ?? 0}, losses ${holdout.losses ?? 0}); ` + (guard ? `regression guards: candidate ${guard.candidate ?? 0}/${guard.n ?? 0} vs baseline ${guard.baseline ?? 0}/${guard.n ?? 0} ` + `(losses ${guard.losses ?? 0}, noise floor ${guard.noiseFloor ?? 0}); ` : "") + `training cases (repair): candidate ${repair.candidate ?? 0}/${repair.n ?? 0} vs baseline ${repair.baseline ?? 0}/${repair.n ?? 0}; ` + `noise floor ${gate.noiseFloor ?? 0}; abstained ${gate.abstained ?? 0}`;
 }
 function printHumanGateVerdict(fix, gateRunId, gateResult, cases) {
   const w = (line) => process.stderr.write(`${line}
@@ -7296,6 +7297,7 @@ Examples:
   moda prompts sync
   moda prompts promote support.triage --label=prod --version=pver_abc123
   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@412   # replay from msg 412
   moda skills pull
   moda fixes
   moda fix start <problem_id> --wait
@@ -9168,7 +9170,7 @@ var commandRegistry = createCommandRegistry([
     resetApiRequestCountBeforeRun: true,
     telemetry: "result",
     handler: async (context) => {
-      const { runInit } = await import("./index-1pb2ftqr.js");
+      const { runInit } = await import("./index-ah6yf5x2.js");
       if (context.outputMode === "agent-stream") {
         context.output.writeEvent({
           event: "started",

package/dist/{index-1pb2ftqr.js → index-ah6yf5x2.js}

@@ -3,7 +3,7 @@ import {
   initPrompts,
   runPromptSync,
   runSkillSync
-} from "./cli-w9an3j54.js";
+} from "./cli-pr6s56fc.js";
 import {
   codingAgentDisplayName,
   describeCodingAgentEvent,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@moda-ai/cli",
-  "version": "1.40.0",
+  "version": "1.41.1",
   "description": "CLI for Moda - AI agent analytics and observability",
   "type": "module",
   "bin": {

package/skills/integration/index.json

@@ -1,7 +1,7 @@
 {
   "schema_version": "moda.skill_index.v1",
-  "bundled_at": "2026-10-09T00:19:17.944Z",
-  "cli_version": "1.40.0",
+  "bundled_at": "2026-10-09T01:20:13.374Z",
+  "cli_version": "1.41.1",
   "skills": [
     {
       "id": "integration-cloudflare-think",

package/skills/moda-cli/SKILL.md

@@ -759,7 +759,9 @@ anything they compute planned playouts (cases × seeds × 2 arms) and print it
 (`plannedPlayouts`). Above 200, or when an existing set's case count can't be
 read, they fail (exit 1) with the number and the exact command to re-run with `--yes`.
 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;
+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);
 `--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`,