@bli-cockpit/cli 0.2.87 → 0.2.89

package/dist/commands/local-args-tower-admin.js

@@ -68,6 +68,9 @@ export function parseOpsArgs(args) {
             "--memory-days",
             // BLI-3798: the coverage half on its own, with the bucket table.
             "--coverage",
+            // BLI-3912: one line per model — latency, tokens/s, today's spend, last
+            // probe verdict. A SECOND door like --memory, asked for only when named.
+            "--models",
             "--person",
             "--dry-run",
             "--json",
@@ -108,6 +111,7 @@ export function parseOpsArgs(args) {
             coverage: coverageOnly,
             skips: values.booleans.has("--skips"),
             memory: values.booleans.has("--memory") || memoryDays !== undefined,
+            models: values.booleans.has("--models"),
             ...(memoryDays === undefined ? {} : { memoryDays: Number(memoryDays) }),
             ...base,
         };

package/dist/commands/local-args-tower-models.js

@@ -0,0 +1,44 @@
+/**
+ * `cockpit models` — the model-card shelf, in the terminal (BLI-3912).
+ *
+ *   cockpit models                       # every card: documented vs measured
+ *   cockpit models list --json
+ *   cockpit models show openai:gpt-5.6-terra
+ *
+ * Two nouns that look alike and are not: `cockpit model` is the ORG SETTING for
+ * which model a slot runs on (`model show`, `model set`); `cockpit models` is
+ * the shelf that setting picks FROM — windows, prices, reasoning vocabulary,
+ * quirks, probe verdicts and stale marks. The help text for each says so.
+ */
+import { optionalNonEmpty, optionalUrl, parseNamedArgs } from "./local-arg-values.js";
+export function parseModelsArgs(args) {
+    const values = parseNamedArgs(args, {
+        allowedFlags: ["--home", "--dashboard-url", "--json"],
+        valueFlags: ["--home", "--dashboard-url"],
+    });
+    if (values.positionals.length > 2) {
+        throw new Error("models accepts at most an action (list|show) and one model id.");
+    }
+    const [rawAction, rawId] = values.positionals;
+    const base = {
+        homeDir: optionalNonEmpty(values.flags.get("--home")),
+        dashboardUrl: optionalUrl(values.flags.get("--dashboard-url")),
+        json: values.booleans.has("--json"),
+    };
+    if (!rawAction || rawAction === "list") {
+        if (rawAction === "list" && rawId) {
+            throw new Error("models list takes no model id — use `models show <provider:model>`.");
+        }
+        return { kind: "models", action: "list", ...base };
+    }
+    if (rawAction !== "show") {
+        // A bare id is the shape people type first; treat it as `show <id>` rather
+        // than refusing on a technicality.
+        return { kind: "models", action: "show", modelId: rawAction, ...base };
+    }
+    const modelId = optionalNonEmpty(rawId);
+    if (!modelId) {
+        throw new Error("models show needs a model id, e.g. `models show openai:gpt-5.6-terra`.");
+    }
+    return { kind: "models", action: "show", modelId, ...base };
+}

package/dist/commands/local-args-tower.js

@@ -33,5 +33,6 @@ export { SCOUT_MIN_PREFIX_LENGTH, parseScoutArgs, parseOpsArgs, SLACK_WORKSPACE_
 export { parseDocsArgs, parseMsgArgs, } from "./local-args-tower-docs-msg.js";
 export { ISSUE_STATES, parseIssueArgs, parseProjectArgs, } from "./local-args-tower-work.js";
 export { SEARCH_KINDS, parseSearchArgs } from "./local-args-tower-search.js";
+export { parseModelsArgs } from "./local-args-tower-models.js";
 export { parseMailArgs } from "./local-args-tower-mail.js";
 export { parseCalArgs } from "./local-args-tower-cal.js";

package/dist/commands/local-args.js

@@ -16,7 +16,7 @@
  * verbatim, no logic change.
  */
 import { parseAgentRulesArgs, parseAnalyzeArgs, parseAutostartArgs, parseBackfillArgs, parseCleanArgs, parseDoctorArgs, parseInstallArgs, parseLoginArgs, parseMemoryArgs, parseLogoutArgs, parseOnboardArgs, parseReleaseArgs, parseServeArgs, parseSessionsArgs, parseStartArgs, parseStatusArgs, parseSyncArgs, parseUpdateArgs, } from "./local-args-collector.js";
-import { parseBriefArgs, parseCorrectArgs, parseDocsArgs, parseIssueArgs, parseJarvisArgs, parseMailArgs, parseCalArgs, parseModelArgs, parseMsgArgs, parseNotesArgs, parseOpsArgs, parseProjectArgs, parseScoutArgs, parseSearchArgs, parseSettingsArgs, parseSlackArgs, parseTeamArgs, parseWorkbookArgs, } from "./local-args-tower.js";
+import { parseBriefArgs, parseCorrectArgs, parseDocsArgs, parseIssueArgs, parseJarvisArgs, parseMailArgs, parseCalArgs, parseModelArgs, parseMsgArgs, parseNotesArgs, parseOpsArgs, parseProjectArgs, parseModelsArgs, parseScoutArgs, parseSearchArgs, parseSettingsArgs, parseSlackArgs, parseTeamArgs, parseWorkbookArgs, } from "./local-args-tower.js";
 // `normalizeUrl` has always been part of this module's surface — `local.ts` and
 // `local-auth.ts` import it from here — so it stays exported from this address
 // even though it now lives next door. The same goes for the four names the
@@ -63,6 +63,8 @@ export function parseLocalArgs(argv) {
             return parseJarvisArgs(argv.slice(1));
         case "model":
             return parseModelArgs(argv.slice(1));
+        case "models":
+            return parseModelsArgs(argv.slice(1));
         case "scout":
             return parseScoutArgs(argv.slice(1));
         case "ops":

package/dist/commands/local-help-commands.js

@@ -273,6 +273,26 @@ export function localSubcommandHelp(command) {
                 "Run `cockpit login` first if this machine is not paired.",
             ],
         ],
+        [
+            "models",
+            [
+                "Usage: cockpit models [list|show <provider:model>] [--dashboard-url <url>] [--json]",
+                "",
+                "The model-card shelf: one card per model, with what the PROVIDER documents (context",
+                "window, max output, price per million, reasoning modes, tool calling, structured output,",
+                "quirks) beside what OUR machine measured (latency p50/p95, observed tokens/s, the last",
+                "belt and conformance probe verdicts).",
+                "Every value carries the page it was read from; a field the provider does not state reads",
+                "unknown with the reason, never a guess.",
+                "STALE:<field> means a measurement contradicted the documentation — reasoning-off refused",
+                "where the card says it can be turned off, the belt refused where the card says tool",
+                "calling. The line under it is the command that re-reads the provider.",
+                "`cockpit models` is the SHELF; `cockpit model` is the org setting saying which model a",
+                "slot runs on. Different nouns, one letter apart, on purpose.",
+                "--json writes one object to stdout; operational lines stay on stderr.",
+                "Run `cockpit login` first if this machine is not paired.",
+            ],
+        ],
         [
             "scout",
             [
@@ -295,7 +315,7 @@ export function localSubcommandHelp(command) {
         [
             "ops",
             [
-                "Usage: cockpit ops [status [--job <id>] [--coverage] [--skips] [--memory [--memory-days N]] | recompile --person <email|name|id> [--dry-run]] [--json]",
+                "Usage: cockpit ops [status [--job <id>] [--coverage] [--skips] [--memory [--memory-days N]] [--models] | recompile --person <email|name|id> [--dry-run]] [--json]",
                 "",
                 "  cockpit ops status",
                 "      One line per scheduled job: when it last produced something, and whether that",

package/dist/commands/local-help.js

@@ -25,6 +25,7 @@ export const rootCommandNames = new Set([
     "analyze",
     "jarvis",
     "model",
+    "models",
     "scout",
     "ops",
     "slack",
@@ -70,8 +71,9 @@ export function localCommandHelp(command) {
         "  cockpit analyze [--workspace <path>] [--dashboard-url <url>] [--max-depth <n>] [--max-repos <n>] [--json]",
         "  cockpit jarvis [question] [--prompt <question>] [--as <person>] [--date <YYYY-MM-DD>] [--thread <name>] [--model <key>] [--image <path>|--file <path>] [--no-stream] [--threads|--history [--limit <n>]] [--trace <id|last>] [--dashboard-url <url>] [--json [--show-approval-code]]",
         "  cockpit model [show|set <provider:model>] [--json]",
+        "  cockpit models [list|show <provider:model>] [--dashboard-url <url>] [--json]",
         "  cockpit scout [start|dismiss|undo <experiment-id>] [--days <n>] [--dashboard-url <url>] [--json]",
-        "  cockpit ops [status [--job <id>] [--skips] [--memory] | recompile --person <email|name|id> [--dry-run]] [--dashboard-url <url>] [--json]",
+        "  cockpit ops [status [--job <id>] [--skips] [--memory] [--models] | recompile --person <email|name|id> [--dry-run]] [--dashboard-url <url>] [--json]",
         "  cockpit slack [coverage [--workspace bli|blue_pearl] [--stale-only] | read [--person <p>] [--channel <c>] [--query <text>] [--since <YYYY-MM-DD>] [--until <YYYY-MM-DD>] [--limit <n>]] [--json]",
         "  cockpit settings [personal [--chat-model <key>] [--brief-model <key>] | switches [set <key> <value>] | models [set --chat <key>] [--memory <id>] | cli-floor [<version>] | env list|set --project <p> --file <f> --content-stdin|delete --id <uuid> [--yes]] [--json]",
         "  cockpit team [members | invite <email> --role <role> [--team-id <uuid>] | role <userId> --role <role> [--yes] | device [list] | device revoke <id|name> --reason <label> [--note \"<text>\"] [--yes]] [--json]",

package/dist/commands/local.js

@@ -16,6 +16,7 @@ import { runDoctor } from "./doctor.js";
 import { runStatus } from "./status.js";
 import { runSessions } from "./sessions.js";
 import { runJarvis } from "./jarvis.js";
+import { runModels } from "./models.js";
 import { runScout } from "./scout.js";
 import { runOps } from "./ops.js";
 import { runSlack } from "./slack.js";
@@ -104,6 +105,8 @@ export async function runLocalCockpitCli(argv, io = defaultIo()) {
                 return await runJarvis(command, io);
             case "model":
                 return await runModel(command, io);
+            case "models":
+                return await runModels(command, io);
             case "scout":
                 return await runScout(command, io);
             case "ops":

package/dist/commands/memory-hook-counts.js

@@ -54,6 +54,7 @@ export function readMemoryHookCounts(options) {
                 }
                 : {}),
             hook_skipped_trivial_24h: window.skippedTrivial,
+            hook_billing_exhausted_24h: window.billingExhausted,
         },
         reason: "ok",
     };

package/dist/commands/models.js

@@ -0,0 +1,86 @@
+/**
+ * `cockpit models` — what do we know about each model? (BLI-3912)
+ *
+ * Prints the shelf: one line per model with the provider's documented window,
+ * price, reasoning mode and tool support beside OUR measured latency, observed
+ * tokens/s and last probe verdict, and a `STALE:<field>` mark wherever the two
+ * disagree — with the command that fixes it on the next line.
+ *
+ * **Every line comes from the server** (`/api/models/cards`, rendered by
+ * `lib/models/model-lines.ts`). This file decides an exit code and nothing
+ * else. A terminal that formatted a card itself would eventually disagree with
+ * the board about what "stale" means, and the one that is wrong is always the
+ * one somebody is reading.
+ *
+ * **Exit codes.** Zero when the shelf was read, whatever it says — a stale card
+ * is a fact to act on, not a broken command. One when the read failed or when a
+ * named card does not exist, because a script asking about a specific model
+ * needs to know it asked about nothing.
+ */
+import { writeLine } from "./cli-io.js";
+import { loadPairedSession, towerJsonRequest, } from "../tower-client.js";
+/** A card read is checked-in data on the server side; it should never hang a shell. */
+const READ_DEADLINE_MS = 30_000;
+export async function runModels(command, io) {
+    const session = await loadPairedSession("models", command.homeDir);
+    const dashboardUrl = command.dashboardUrl ?? session.dashboard_url;
+    const log = (line) => writeLine(io.stderr, line);
+    const path = command.action === "show" && command.modelId
+        ? `/api/models/cards?id=${encodeURIComponent(command.modelId)}`
+        : "/api/models/cards";
+    const result = await towerJsonRequest({
+        dashboardUrl,
+        path,
+        method: "GET",
+        deviceToken: session.device_token,
+        fetch: io.fetch,
+        label: "models",
+        timeoutMs: READ_DEADLINE_MS,
+        log,
+    });
+    if (!result.ok) {
+        writeFailure(command, io, result);
+        return 1;
+    }
+    const payload = (result.body ?? {});
+    if (command.json) {
+        writeLine(io.stdout, JSON.stringify({ ok: true, ...payload }));
+    }
+    else {
+        for (const line of payload.lines ?? [])
+            writeLine(io.stdout, line);
+    }
+    // Never silent about a card the server could not read: a shelf quietly one
+    // model short is exactly what the card registry exists to prevent.
+    for (const unreadable of payload.unreadable ?? []) {
+        writeLine(io.stderr, `[models cli] card unreadable ${JSON.stringify({ card: unreadable })}`);
+    }
+    writeLine(io.stderr, `[models cli] read ${JSON.stringify({
+        action: command.action,
+        model: command.modelId ?? null,
+        cards: Array.isArray(payload.cards) ? payload.cards.length : payload.card ? 1 : 0,
+        unreadable: payload.unreadable?.length ?? 0,
+    })}`);
+    return 0;
+}
+function writeFailure(command, io, failure) {
+    // `TowerFailure` carries the server's own words in `detail` — a 404 for a
+    // model nobody has carded says exactly which id it was and what to run.
+    const message = failure.detail;
+    if (command.json) {
+        writeLine(io.stdout, JSON.stringify({
+            ok: false,
+            error: failure.reason,
+            message,
+            httpStatus: failure.httpStatus ?? null,
+        }));
+    }
+    else {
+        writeLine(io.stdout, message);
+    }
+    writeLine(io.stderr, `[models cli] not read ${JSON.stringify({
+        reason: failure.reason,
+        http_status: failure.httpStatus ?? null,
+        model: command.modelId ?? null,
+    })}`);
+}

package/dist/commands/ops-render-memory.js

@@ -96,11 +96,14 @@ export function renderMemoryHooks(hooks, dim) {
         const rightKnown = typeof right.runs === "number" ? 0 : 1;
         if (leftKnown !== rightKnown)
             return leftKnown - rightKnown;
-        return (right.timeouts ?? 0) - (left.timeouts ?? 0);
+        // BLI-3891: a billing refusal ranks beside a missed deadline. Both are a
+        // recall the person did not get; only the fix differs.
+        return ((right.timeouts ?? 0) + (right.billingExhausted ?? 0)
+            - ((left.timeouts ?? 0) + (left.billingExhausted ?? 0)));
     });
     for (const device of ranked) {
         const line = `  ${device.line ?? `${device.displayName ?? "?"}: (no line)`}`;
-        lines.push((device.timeouts ?? 0) > 0 ? line : dim(line));
+        lines.push((device.timeouts ?? 0) > 0 || (device.billingExhausted ?? 0) > 0 ? line : dim(line));
         if (device.performanceLine)
             lines.push(`    ${device.performanceLine}`);
         // BLI-3884. Written on the server (`lib/ops/memory-hook-misses.ts`) and

package/dist/commands/ops-render-spend.js

@@ -0,0 +1,36 @@
+/**
+ * `cockpit ops` — WHAT TODAY COST, by model (BLI-3909).
+ *
+ * Split out of `ops-render.ts` the way `ops-render-memory.ts` and
+ * `ops-render-coverage.ts` were, when that file crossed the repo's 500-line
+ * readability floor.
+ *
+ * Same rule as the rest of that family: every sentence here was written on the
+ * SERVER (`lib/ops/model-spend-today.ts`) and is printed VERBATIM. This module
+ * decides order, indentation and column widths and nothing else — a terminal
+ * that composed its own total from the same fields would eventually disagree
+ * with the page about what a $25 day means, and the one that is wrong is
+ * always the one somebody is reading.
+ *
+ * A CLI too old to know this section simply does not print it. Nothing is lost
+ * by that: the section rides the ordinary `/api/ops/status` response, so
+ * `cockpit ops --json` carries it on every version.
+ */
+/** `$18.65`, `$0.0031`, `—` when the server sent no number at all. */
+function usdColumn(usd) {
+    if (typeof usd !== "number")
+        return "—";
+    return `$${usd.toFixed(usd < 0.01 ? 4 : 2)}`;
+}
+/**
+ * The SPEND block: the server's one line, then one row per model. `$314 of a
+ * $357 month on one model` is the thing this exists to make visible before the
+ * invoice does.
+ */
+export function renderSpend(spend, dim) {
+    const lines = ["", `SPEND  ${spend.summary ?? "(no summary)"}`];
+    for (const row of spend.models ?? []) {
+        lines.push(dim(`  ${usdColumn(row.usd).padStart(9)}  ${row.model ?? "?"} · ${row.calls ?? 0} call(s) · ${(row.bases ?? []).join("+") || "?"} · ${(row.sources ?? []).join(",") || "?"}`));
+    }
+    return lines;
+}

package/dist/commands/ops-render.js

@@ -13,6 +13,10 @@
  * somebody is about to conclude something from it.
  */
 import { renderCoverageBuckets, } from "./ops-render-coverage.js";
+// BLI-3909: today's spend by model, its own renderer for the same reason the
+// coverage table has one — this file is at the repo's readability floor.
+import { renderSpend } from "./ops-render-spend.js";
+export { renderSpend } from "./ops-render-spend.js";
 export { renderMemoryHooks, renderMemoryUsage } from "./ops-render-memory.js";
 export { renderCoverageBuckets };
 /** The word a person reads. Short, fixed width, and never a bare colour. */
@@ -124,6 +128,8 @@ options = {}) {
             lines.push(...renderCoverageBuckets(payload.coverage.buckets, dim));
         }
     }
+    if (payload.spend)
+        lines.push(...renderSpend(payload.spend, dim));
     const slack = payload.skips?.slack;
     const external = payload.skips?.external;
     if (slack || external) {

package/dist/commands/ops.js

@@ -77,6 +77,11 @@ async function runOpsStatus(command, io, tower) {
     // status board's exit code — "is the org using memory" going unread is not a
     // pipeline outage.
     const memory = command.memory ? await readMemoryUsage(command, io, tower) : null;
+    // BLI-3912. A THIRD door, asked for only when `--models` was, and never able
+    // to fail the board: "what is each model costing and how fast is it" is a
+    // different question from "is the pipeline healthy", and a card shelf that
+    // could not be read is not a collection outage.
+    const models = command.models ? await readModelBoardLines(io, tower) : null;
     // BLI-3723: `failing` joins the list. An older CLI that has never heard of it
     // simply does not count it — which is why the server keeps the sentence in
     // `detail`, so an un-upgraded terminal still PRINTS the outage even when it
@@ -88,7 +93,13 @@ async function runOpsStatus(command, io, tower) {
         // BLI-3762: a job that ran and wrote less than it owed is not healthy.
         row.verdict === "degraded");
     if (command.json) {
-        writeLine(io.stdout, JSON.stringify(memory ? { ...payload, memory: memory.section, hooks: memory.hooks, experience: memory.experience } : payload));
+        writeLine(io.stdout, JSON.stringify({
+            ...payload,
+            ...(memory
+                ? { memory: memory.section, hooks: memory.hooks, experience: memory.experience }
+                : {}),
+            ...(models ? { models: models.board, modelsReason: models.reason } : {}),
+        }));
     }
     else {
         const styled = colorEnabled(io);
@@ -116,6 +127,17 @@ async function runOpsStatus(command, io, tower) {
             for (const line of memory.experience.lines)
                 writeLine(io.stdout, `  ${line}`);
         }
+        if (models) {
+            writeLine(io.stdout, "");
+            if (models.lines.length > 0) {
+                for (const line of models.lines)
+                    writeLine(io.stdout, line);
+            }
+            else {
+                // Never a silent gap where a section was asked for.
+                writeLine(io.stdout, `MODELS  not read (${models.reason})`);
+            }
+        }
         if (memory && !memory.section) {
             // Never a silent gap where a section was asked for: the reason the gauge
             // could not be read is printed where the gauge would have been.
@@ -152,11 +174,19 @@ async function runOpsStatus(command, io, tower) {
         // BLI-3881: runs the hook declined to search for. A SUBSET of runs, so
         // `runs - skipped_trivial` is how many turns actually asked the door.
         memory_hook_skipped_trivial_24h: memory?.hooks?.totals?.skippedTrivial ?? null,
+        // BLI-3891: null means the fleet reported none of these at all, which is
+        // "not measured" on a machine older than the count — never "no refusals".
+        memory_hook_billing_exhausted_24h: memory?.hooks?.totals?.billingExhausted ?? null,
         memory_hook_devices_reporting: memory?.hooks?.reporting ?? null,
         // BLI-3884: which route answered them, fleet-wide. Null while no machine
         // names one — a rollout fact, not a count of zero.
         memory_hook_via_daemon_24h: memory?.hooks?.via?.daemon ?? null,
         memory_hook_via_direct_24h: memory?.hooks?.via?.direct ?? null,
+        // BLI-3912: whether the model board was asked for, and what it said.
+        models_asked: Boolean(command.models),
+        models_reason: models?.reason ?? null,
+        models_counted: models?.board?.rows?.length ?? null,
+        models_cost_today_usd: models?.board?.total_cost_today_usd ?? null,
     })}`);
     return unhealthy.length > 0 ? 1 : 0;
 }
@@ -192,6 +222,29 @@ async function readMemoryUsage(command, io, tower) {
         return { section: null, hooks: body.hooks ?? null, experience: body.experience ?? null, reason: "memory_section_absent" };
     return { section: body.memory, hooks: body.hooks ?? null, experience: body.experience ?? null, reason: "ok" };
 }
+/**
+ * The per-model board, read through its own door (BLI-3912).
+ *
+ * Same posture as the adoption gauge next door: it never fails the status
+ * board, and when it cannot be read the reason is printed where the section
+ * would have been. The LINES come from the server, so a terminal and the
+ * dashboard cannot disagree about what a model cost today.
+ */
+async function readModelBoardLines(io, tower) {
+    const result = await callTower(tower, { path: "/api/ops/models", label: "ops-models" });
+    if (!result.ok) {
+        writeLine(io.stderr, `[ops cli] model board not read ${JSON.stringify({
+            reason: result.reason,
+            http_status: result.httpStatus ?? null,
+        })}`);
+        return { lines: [], board: null, reason: result.reason };
+    }
+    const body = asRecord(result.body);
+    if (!body.lines || body.lines.length === 0) {
+        return { lines: [], board: body.board ?? null, reason: "models_section_absent" };
+    }
+    return { lines: body.lines, board: body.board ?? null, reason: "ok" };
+}
 async function runOpsRecompile(command, io, tower) {
     const person = command.person ?? "";
     if (!command.json) {

package/dist/commands/public-root.js

@@ -15,7 +15,7 @@ export async function runCockpitCli(argv, io) {
   }
 
   if (command === "--version" || command === "-V" || command === "version") {
-    writeLine(io?.stdout ?? process.stdout, "0.2.87");
+    writeLine(io?.stdout ?? process.stdout, "0.2.89");
     return 0;
   }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bli-cockpit/cli",
-  "version": "0.2.87",
+  "version": "0.2.89",
   "private": false,
   "type": "module",
   "bin": {
@@ -27,8 +27,8 @@
     "test": "node dist/cli.js --help && node ../../scripts/assert-public-cli-routing.mjs && node ../../scripts/assert-public-cli-verb-help.mjs && node ../../scripts/assert-public-cli-runtime-files.mjs && node ../../scripts/assert-public-cli-no-fleet-posts.mjs && node ../../scripts/assert-public-package-pack.mjs --workspace=@bli-cockpit/cli"
   },
   "dependencies": {
-    "@bli-cockpit/memory-mcp": "0.1.19",
-    "@bli-cockpit/mcp": "0.1.20",
-    "@bli-cockpit/telemetry-core": "0.1.36"
+    "@bli-cockpit/memory-mcp": "0.1.20",
+    "@bli-cockpit/mcp": "0.1.22",
+    "@bli-cockpit/telemetry-core": "0.1.37"
   }
 }