@moda-ai/cli 1.43.1 → 1.44.0

package/dist/{cli-141ca3yv.js → cli-j2rdgd3p.js}

@@ -29,6 +29,8 @@ import {
 } from "./cli-rycpwqzm.js";
 import {
   CliInputError,
+  EXIT_ERROR,
+  EXIT_OK,
   createCommandContext,
   ensureSecretFilePermissions,
   getActiveProfile,
@@ -5662,7 +5664,7 @@ async function runPromptAb(flags, profileOptions, context) {
         message: enqueue.message ?? "Replay comparison queued",
         plannedPlayouts
       }, writeOptions);
-      return;
+      return EXIT_OK;
     }
     const deadline = Date.now() + timeoutMs;
     let latest = null;
@@ -5696,7 +5698,9 @@ async function runPromptAb(flags, profileOptions, context) {
     if (context.outputMode === "human") {
       printHumanVerdict(result);
     }
-    context.output.writeData(result, writeOptions);
+    const failed = result.status === "error";
+    context.output.writeData(result, { ...writeOptions, ...failed ? { status: "error" } : {} });
+    return failed ? EXIT_ERROR : EXIT_OK;
   } catch (error) {
     if (warnings.length && error && typeof error === "object") {
       Object.assign(error, { warnings });
@@ -6608,8 +6612,7 @@ async function runPromptsCommand(input, legacyFlags, legacyProfileOptions) {
       await promotePrompt(positionals, flags, profileOptions, context);
       break;
     case "ab":
-      await runPromptAb(flags, profileOptions, context);
-      break;
+      return runPromptAb(flags, profileOptions, context);
     case "propose":
       await runPromptPropose(positionals, flags, profileOptions, context);
       break;
@@ -6624,6 +6627,7 @@ async function runPromptsCommand(input, legacyFlags, legacyProfileOptions) {
     default:
       throw new Error(`Unknown prompts command '${subcommand}'. Use init, status, diff, sync, promote, ab, propose, list, show, proposals, usage, replay-runs, or decide.`);
   }
+  return 0;
 }
 async function initPrompts(options = {}) {
   const cwd = process.cwd();
@@ -7284,17 +7288,25 @@ var FIXES_BLOCK_END = "<!-- moda:fixes:end -->";
 function renderFixesBlock() {
   return [
     FIXES_BLOCK_BEGIN,
-    "## Moda Fixes (managed block — edits inside are overwritten by moda-cli)",
+    "## Fixing production issues with Moda (managed block — edits inside are overwritten by moda-cli)",
+    "",
+    "Moda supplies the production evidence; you propose and make the fix yourself.",
     "",
-    "- Before editing prompt or skill files, run `moda fixes` to see the ranked",
-    '  queue of Moda Fixes ("a fix with proof") already drafted for this repo,',
-    "  and `moda fix <fix_id> --packet` for the evidence packet.",
-    "- Gate any prompt/skill change against production evidence with",
-    "  `moda fix verify <fix_id> [--prompt-file=<path>]`. Exit codes: 0 = holdout",
-    "  win vs baseline (pass), 1 = failing (per-case findings in the envelope),",
-    "  3 = degraded/inconclusive — never treat exit 3 as a pass.",
-    "- Fix PRs must carry the magic word `Fixes MODA-FIX-<SHORTREF>` in the PR",
-    "  body so the merge confirms the fix and starts production monitoring.",
+    "- Find what to fix: `moda investigate` ranks tool failures, user frustrations,",
+    "  and top Problems with evidence; `moda problems` ranks cross-signal Problems",
+    "  by root cause; `moda failures` lists tool failures only.",
+    "- Read the evidence before editing: `moda problem <problem_id>` (dossier),",
+    "  `moda problem <problem_id> --evidence` (attribution rationales and anchors),",
+    "  `moda problem <problem_id> --traces` (affected traces),",
+    "  `moda tool-failure-detail <tool_name>` (failing calls in context), and",
+    "  `moda context <conversation_id>` (the messages around a turn; `--all` for",
+    "  the whole trace).",
+    "- 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.",
+    "- After the fix ships, record it with",
+    "  `moda problem-feedback <problem_id> --action=mark_fixed`; if the issue",
+    "  is detected again, the Problem comes back.",
     FIXES_BLOCK_END
   ].join(`
 `);

package/dist/cli.js

@@ -8,7 +8,7 @@ import {
   runPromptsCommand,
   runSkillsCommand,
   runStatusCommand
-} from "./cli-141ca3yv.js";
+} from "./cli-j2rdgd3p.js";
 import {
   ApiError,
   HARNESS_REPORT_APPROVAL_PATH,
@@ -6899,7 +6899,7 @@ function buildManifest(commands) {
     exit_codes: [
       { code: 0, name: "ok", meaning: "Command completed successfully." },
       { code: 1, name: "error", meaning: "Command, input, API, network, or unexpected failure." },
-      { code: 3, name: "degraded", meaning: "Degraded result: `moda ask` local fallback, `moda fix verify` gate inconclusive/blocked-coverage (never treat as a pass), or `moda fixes drive` timeout with fixes still advancing." },
+      { code: 3, name: "degraded", meaning: "Degraded result: `moda ask` local fallback." },
       { code: 4, name: "auth_required", meaning: "Run login_command, then resume." },
       { code: 5, name: "input_required", meaning: "Choose from input_request choices, then resume." },
       { code: 130, name: "cancelled", meaning: "Interrupted or output consumer closed." }
@@ -6923,7 +6923,7 @@ function buildManifest(commands) {
       "moda.intelligence.v1",
       "production_ask.v0.1"
     ],
-    commands: commands.map((command) => ({
+    commands: commands.filter((command) => !command.hidden).map((command) => ({
       name: command.name,
       aliases: command.aliases ?? [],
       summary: command.description,
@@ -6971,8 +6971,6 @@ function event(name, when, terminal, deprecated = false) {
 function exitCodesForCommand(command) {
   if (command === "ask")
     return [0, 1, 3, 4, 5, 130];
-  if (command === "fix" || command === "fixes")
-    return [0, 1, 3, 4, 130];
   if (command === "init" || command === "harness")
     return [0, 1, 4, 5, 130];
   return [0, 1, 4, 130];
@@ -7203,30 +7201,6 @@ Commands:
   skills list|show|policy          List pushed skills, show one, or set --live/--local
   registry push                    Push prompts, skills, and tools from one JSON manifest
 
-  Fixes (a fix with proof):
-  fixes                            Ranked Fix queue (--status=S --limit=N --cursor=TOKEN)
-  fixes draft-batch                Campaign mode: batch-draft fixes for the top-ranked problems
-                                   (--limit=N, 1-25, default 5 — batches cap at 25: LLM spend guard)
-  fixes drive                      Round-robin one advance per active fix per pass until all rest
-                                   (--status=S narrows the server-side scan window to one advancing
-                                   status, --max-fixes=N 1-10 default 3 fixes per run (rest skipped;
-                                   GATING fixes count toward N; least-recently-driven first),
-                                   --timeout=ms default 2h, --pass-interval=ms default 20s;
-                                   exit 3 when the timeout lapses, fixes were skipped, or still advancing)
-  fix <fix_id>                     One Fix: status + gate result; --packet prints moda.fix_packet.v1
-                                   (tool/skill candidates are also written under .moda/fixes/<ref>/);
-                                   --wait drives a still-running pipeline to rest
-  fix start <problem_id>           Draft a Fix from a Problem (--type=auto|prompt); --wait drives
-                                   scope -> propose -> gate and prints the packet
-  fix verify <fix_id>              Re-gate on the frozen holdout; --prompt-file=P gates local content
-                                   (exit 0 pass / 1 fail with per-case findings / 3 degraded-inconclusive)
-                                   Refuses a re-gate within 10 min of the last verdict unless --yes
-  fix checkout <fix_id>            Write the candidate at its target source path (never runs git)
-  fix submit <fix_id>              Deliver: --pr opens a draft PR | --local-ref=<branch> records your branch
-  fix mark-applied <fix_id>        Confirm a candidate you applied in your own stack (--note=TEXT);
-                                   no PR needed — flips to SHIPPED and starts monitoring
-  fix dismiss <fix_id>             Reject with --reason=TEXT (feeds problem feedback)
-
 Options:
   --help                           Show this help message
   --version                        Show version number
@@ -7489,10 +7463,6 @@ 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@412   # replay from msg 412
   moda skills pull
-  moda fixes
-  moda fix start <problem_id> --wait
-  moda fix verify <fix_id> --prompt-file=prompts/agent.prompt.md
-  moda fix mark-applied <fix_id> --note="tool description updated in our agent config"
 
 Run \`moda <command> --help\` for a command's subcommands, and
 \`moda <command> <subcommand> --help\` for that subcommand's flags
@@ -8330,8 +8300,7 @@ async function runCommand(command, positional, flags, positionals = positional ?
   assertHandParsedFlags(command, flags, positionals);
   switch (command) {
     case "prompts": {
-      await runPromptsCommand(context);
-      break;
+      return runPromptsCommand(context);
     }
     case "overview": {
       const args = flagsToArgs(flags);
@@ -9375,7 +9344,7 @@ var commandRegistry = createCommandRegistry([
     resetApiRequestCountBeforeRun: true,
     telemetry: "result",
     handler: async (context) => {
-      const { runInit } = await import("./index-cnfwmzew.js");
+      const { runInit } = await import("./index-nepr7cw7.js");
       if (context.outputMode === "agent-stream") {
         context.output.writeEvent({
           event: "started",
@@ -10204,6 +10173,7 @@ var commandRegistry = createCommandRegistry([
   },
   {
     name: "fixes",
+    hidden: true,
     description: 'List the ranked queue of Fixes ("a fix with proof"); campaign mode batch-drafts and drives them',
     examples: [
       "moda fixes",
@@ -10225,6 +10195,7 @@ var commandRegistry = createCommandRegistry([
   },
   {
     name: "fix",
+    hidden: true,
     description: 'Draft, verify, check out, submit, and dismiss Fixes ("a fix with proof")',
     subcommands: FIX_SUBCOMMAND_HELP,
     examples: [

package/dist/{index-cnfwmzew.js → index-nepr7cw7.js}

@@ -3,7 +3,7 @@ import {
   initPrompts,
   runPromptSync,
   runSkillSync
-} from "./cli-141ca3yv.js";
+} from "./cli-j2rdgd3p.js";
 import {
   codingAgentDisplayName,
   describeCodingAgentEvent,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@moda-ai/cli",
-  "version": "1.43.1",
+  "version": "1.44.0",
   "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": "2026-10-09T03:57:10.818Z",
-  "cli_version": "1.43.1",
+  "bundled_at": "2026-10-09T08:32:09.674Z",
+  "cli_version": "1.44.0",
   "skills": [
     {
       "id": "integration-cloudflare-think",

package/skills/moda-cli/SKILL.md

@@ -900,86 +900,37 @@ rejected, matching what the server accepts. Skill keys are stricter than the
 server (letters, digits, `_`, `-`) because they double as `.claude/skills/<key>/`
 directory names.
 
-### 8. Fixing detected Problems (`moda fix` — a fix with proof)
+### 8. Fixing detected Problems
 
-Moda turns detected Problems into **Fixes**: a routed candidate change plus a
-replay **gate** that proves it on held-out production evidence. Use this loop
-whenever you are asked to fix a detected Problem, or before hand-editing
-prompt/skill files in a repo that talks to Moda.
-
-Fixes are typed by where the evidence points: `PROMPT` (a managed prompt
-revision — the checkout/verify/submit loop below), `TOOL_SCHEMA` (a proposed
-rewrite of one tool's description), and `SKILL` (a drafted SKILL.md for an
-agent-behavior problem). Tool and skill candidates are not repo files: read
-them from the packet (they are also written locally under
-`.moda/fixes/<SHORTREF>/`), apply them in your own stack, then confirm with
-`moda fix mark-applied` — no GitHub integration required.
-
-The fail-to-pass loop:
+Moda does not draft fixes. It supplies the production evidence; you propose
+and make the fix yourself, in the code, prompt, tool definition, or skill that
+causes the issue. Use this loop whenever you are asked to fix a detected
+Problem, or before hand-editing prompt/skill files in a repo that talks to Moda.
 
 ```bash
-moda fixes                                   # ranked queue of fixes with proof
-moda fix <fix_id> --packet                   # moda.fix_packet.v1: cause, target
-                                             # file, evidence links, verify cmd,
-                                             # handback + magic word; tool/skill
-                                             # candidates land in .moda/fixes/
-moda fix start <problem_id> --wait           # draft from a Problem and drive
-                                             # scope -> propose -> gate
-moda fix checkout <fix_id>                   # PROMPT fixes: write the candidate
-                                             # at its target source path
-                                             # (NEVER runs git)
-# ...edit the target file...
-moda fix verify <fix_id> --prompt-file=<path>  # re-gate local content on the
-                                               # frozen holdout (PROMPT only)
-# iterate edit -> verify until exit 0, then hand back:
-moda fix submit <fix_id> --pr                # draft PR via the Moda GitHub App
-moda fix submit <fix_id> --local-ref=<branch>  # or record your own branch
-moda fix mark-applied <fix_id> --note="…"    # tool/skill/no-PR path: confirm a
-                                             # candidate you applied yourself
-moda fix dismiss <fix_id> --reason="why"     # reject (feeds problem feedback)
+moda investigate                             # rank tool failures, frustrations,
+                                             # and top Problems with evidence
+moda problems                                # cross-signal Problems by root cause
+moda problem <problem_id>                    # the dossier: what, where, how often
+moda problem <problem_id> --evidence         # attribution rationales and anchors
+moda problem <problem_id> --traces           # affected traces with anchors
+moda tool-failure-detail <tool_name>         # failing calls with surrounding turns
+moda context <conversation_id> --window=3    # the messages around one turn
+# ...propose and make the change, citing the problem/trace ids in the PR...
+moda problem-feedback <problem_id> --action=mark_fixed   # after it ships
 ```
 
-Every gate replays the holdout, so `moda fix verify` refuses to re-gate a fix
-whose last verdict landed less than 10 minutes ago (`gateResult.completedAt`;
-verdicts without that stamp have no cooldown) and sends nothing. The error
-shows the exact command with `--yes` to gate anyway. Pass `--yes` only when you
-changed the candidate since that verdict. `moda fixes drive` advances at most
-`--max-fixes` fixes per run (default 3, max 10; fixes already `GATING` count
-toward the cap; each run takes the least-recently-driven fixes first), lists the rest in
-`data.skipped`, and exits 3 until a re-run has driven them. `moda fixes
-draft-batch` drafts 5 fixes unless you pass `--limit` (max 25).
-
-**`moda fix verify` exit codes are the contract — branch on them:**
+Rules that keep the fix grounded:
 
-| Exit | Meaning | How to treat it |
-|---|---|---|
-| `0` | gate pass — the candidate beat baseline on the frozen holdout (wins clear the noise floor) | Safe to submit. |
-| `1` | gate fail — per-case findings ride the envelope's `findings[]` (`case:<id>` entries with baseline/candidate pass) | Edit the candidate and re-`verify`. |
-| `3` | degraded/inconclusive — too many abstentions, blocked coverage, or no verdict | **Never treat as a pass.** Re-verify or read `moda fix <fix_id> --packet`. |
-
-Rules that keep the proof honest:
-
-- The verdict reads only **holdout** cases the proposer never saw; repair
-  rows are training cases, never the headline.
-- The gate compares against a noise floor from a baseline-vs-baseline control
-  run — a "win" inside the noise floor does not pass.
-- `checkout` only writes the file. Branching (`moda/fix/<shortref-lower>`),
-  committing, and pushing are yours; the CLI never runs git.
-- Fix PRs end with the magic word `Fixes MODA-FIX-<SHORTREF>` in the body —
-  Moda-authored PRs carry it automatically; hand-authored PRs must include it
-  so the merge confirms the fix and starts production monitoring.
-- `mark-applied` is the confirmation for changes shipped without a PR (tool
-  descriptions, installed skills, hand-applied prompts): it flips the fix to
-  `SHIPPED`, records your `--note`, and starts the same monitoring the merge
-  webhook would. It is idempotent — repeating it is a duplicate no-op.
-- `TOOL_SCHEMA` fixes gate like prompt fixes (both arms pin the same current
-  prompt; the proposed description is the only delta). `SKILL` fixes rest at
-  `PROPOSED` until the skill is registered with the Moda skill harness — an
-  inline drafted SKILL.md cannot enter the gate, so deliver it via the packet
-  and `mark-applied`.
-- The pipeline is advance-on-poll: `--wait` (on `start`, `verify`, or
-  `moda fix <fix_id> --wait`) drives it; a fix left `GATING` will not finish
-  on its own.
+- Read the evidence before editing. Base the change on what the traces show,
+  not on the Problem's title alone.
+- 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).
+- Mark the Problem fixed only after the change ships. If the issue is detected
+  again, the Problem comes back; `moda problem-reopen` reopens it by hand.
 
 ### 9. Custom signals and false positives
 
@@ -1060,7 +1011,7 @@ moda detection-review <problem_id> --conversation-id=<id> \
 ### 10. Everything the dashboard does, from the CLI
 
 Every command below needs only an API key, and the tenant comes from the key.
-Writes listed here accept `--dry-run` (prints the exact request, sends nothing) except `problem-feedback`, `false-positive`, `detection-review`, `signal-label`, `problem-linear-file`, `prompts promote|ab|propose`, and `fix`/`fixes`. Those refuse `--dry-run` before sending anything, so run them only when you mean it.
+Writes listed here accept `--dry-run` (prints the exact request, sends nothing) except `problem-feedback`, `false-positive`, `detection-review`, `signal-label`, `problem-linear-file`, and `prompts promote|ab|propose`. Those refuse `--dry-run` before sending anything, so run them only when you mean it.
 
 | Dashboard | CLI |
 |---|---|
@@ -1077,7 +1028,6 @@ Writes listed here accept `--dry-run` (prints the exact request, sends nothing)
 | Prompts registry | `moda prompts list [--label --search]`, `show <key>`, `proposals <key>`, `usage <key>`, `replay-runs <key>`, `decide <key> <proposal_id> --action=merge\|reject\|reopen` |
 | Skills library | `moda skills list`, `show <key>`, `policy <key> --live\|--local` |
 | Harness pages | `moda harness list`, `show <id>`, `versions <id>`, `diff <id> --from=… --to=…` |
-| Fixes inbox | `moda fixes`, `moda fix <id>`, `fix start <problem_id>`, `fix dismiss` … |
 
 Not available to API keys on purpose (a human does these in the dashboard):
 members/invites/roles, API-key minting, billing, provider credentials,
@@ -1284,14 +1234,6 @@ npx: `npx -p @moda-ai/cli moda <command>`.
 | `moda prompts list\|show\|proposals\|usage\|replay-runs\|decide` | Read the prompt registry; merge/reject proposals |
 | `moda replays sets\|runs\|run\|playouts\|playout` | Replay results: run history, per-seed verdicts, playout transcripts with tool provenance (read-only) |
 | `moda skills list\|show\|policy` / `moda harness list\|show\|versions\|diff` | Skill library and synced harnesses |
-| `moda fixes` | Ranked queue of Fixes ("a fix with proof"); `--status=S`, `--limit=N`, `--cursor=TOKEN` |
-| `moda fix <fix_id>` | One Fix: status + gate result; `--packet` prints `moda.fix_packet.v1` and writes tool/skill candidates under `.moda/fixes/<ref>/`; `--wait` drives a running pipeline to rest |
-| `moda fix start <problem_id>` | Draft a Fix from a Problem (`--type=auto\|prompt`); `--wait` drives scope → propose → gate |
-| `moda fix verify <fix_id>` | Re-gate on the frozen holdout; `--prompt-file=P` gates local content, PROMPT fixes only (exit `0` pass / `1` fail with per-case findings / `3` degraded — never a pass) |
-| `moda fix checkout <fix_id>` | Write the candidate at its target source path (never runs git; PROMPT fixes only) |
-| `moda fix submit <fix_id>` | Deliver: `--pr` opens a draft PR \| `--local-ref=<branch>` records your branch |
-| `moda fix mark-applied <fix_id>` | Confirm a candidate you applied in your own stack (`--note=TEXT`, ≤600 chars); no PR needed — flips to SHIPPED and starts monitoring |
-| `moda fix dismiss <fix_id>` | Reject with `--reason=TEXT` (feeds problem feedback) |
 
 `moda search` flags: `--mode=keyword|semantic|hybrid` (default `hybrid`),
 `--user-id`, `--time-range=<range>`, `--limit=N` (1–100, default 20).