@trazum/cli 1.28.0 → 1.29.0

package/dist/index.js

@@ -44,6 +44,7 @@ const VALUE_FLAGS = new Set([
     'max-growth-usd',
     'max-cache-loss-usd',
     'max-day-usd',
+    'max-session-usd',
     'csv-out',
     'csv-shape',
     'what-if',
@@ -258,7 +259,7 @@ const COMMAND_FLAGS = {
     ],
     check: ['max-tokens', 'level', 'exact-tokens', 'markdown-out', 'baseline'],
     baseline: ['model', 'calls', 'output-tokens', 'cache-hit-rate', 'batch', 'exact-tokens', 'out', 'o'],
-    profile: ['json', 'pricing', 'pricing-live', 'against', 'what-if', 'markdown-out', 'csv-out', 'csv-shape', 'max-usd', 'max-growth-usd', 'max-cache-loss-usd', 'max-day-usd', 'label', 'since', 'until'],
+    profile: ['json', 'pricing', 'pricing-live', 'against', 'what-if', 'markdown-out', 'csv-out', 'csv-shape', 'max-usd', 'max-growth-usd', 'max-cache-loss-usd', 'max-day-usd', 'max-session-usd', 'label', 'since', 'until'],
     route: ['prompt-file', 'cases', 'label', 'concurrency', 'json', 'yes', 'pricing', 'pricing-live'],
     eval: ['cases', 'level', 'concurrency', 'export', 'out', 'o', 'model'],
     prune: ['cases', 'concurrency', 'json', 'yes'],
@@ -1730,6 +1731,7 @@ async function commandProfile(args, config, pricing, t) {
             typeof args.flags.get('max-growth-usd') === 'string' ||
             typeof args.flags.get('max-cache-loss-usd') === 'string' ||
             typeof args.flags.get('max-day-usd') === 'string' ||
+            typeof args.flags.get('max-session-usd') === 'string' ||
             config.spend !== undefined;
         if (anyGate) {
             const reasons = [];
@@ -1871,6 +1873,35 @@ async function commandProfile(args, config, pricing, t) {
                 }
             }
         }
+        /**
+         * The per-conversation gate — the unit an agent product actually blows
+         * up in. A month's budget and a day's budget both pass while one
+         * conversation loops its way through $400; the single most expensive
+         * conversation is the number a per-conversation policy has to judge,
+         * and the log already carries it.
+         *
+         * The refusals it inherits: a log with **no sessions** fails rather
+         * than passes ("not measured" is not "under budget"), and a
+         * conversation that started before this log is counted only for the
+         * turns recorded here — so a pass is a floor, and the pass message says
+         * so. The session key itself is never printed, here or anywhere.
+         */
+        if (typeof args.flags.get('max-session-usd') === 'string' || config.spend?.maxSessionUsd !== undefined) {
+            const maxSession = typeof args.flags.get('max-session-usd') === 'string'
+                ? numberFlag(args, 'max-session-usd', 0, t)
+                : config.spend.maxSessionUsd;
+            if (report.sessionSpend === null) {
+                console.error(c.red(t.profile.maxSessionNoSessions()));
+                process.exitCode = 1;
+            }
+            else if (report.sessionSpend.maxUsd > maxSession) {
+                console.error(c.red(t.profile.maxSessionFailed(formatUsd(report.sessionSpend.maxUsd), formatUsd(maxSession), n(report.sessionSpend.sessions))));
+                process.exitCode = 1;
+            }
+            else {
+                console.error(c.dim(t.profile.maxSessionOk(formatUsd(report.sessionSpend.maxUsd), formatUsd(maxSession), n(report.sessionSpend.sessions))));
+            }
+        }
     };
     /**
      * The side files the caller asked for. Written on **both** output paths:
@@ -2463,6 +2494,17 @@ async function commandProfile(args, config, pricing, t) {
             console.log(`  ${c.yellow('!')} ${c.dim(wrap(t.profile.sessionCostTail((shape.p95Usd / shape.medianUsd).toFixed(0)), 74, '    '))}`);
         }
     }
+    /**
+     * The figure that survives a small log. `sessionCosts` refuses slices too
+     * thin for a percentile, and rightly — but a log of four conversations
+     * still has a most expensive one, and that maximum is a fact at any count.
+     * It is also exactly the number `--max-session-usd` judges, so the report
+     * states it rather than going silent where the gate would speak.
+     */
+    if (report.sessionCosts.length === 0 && report.sessionSpend !== null) {
+        console.log();
+        console.log(`  ${c.dim(wrap(t.profile.sessionSpendOnly(n(report.sessionSpend.sessions), formatUsd(report.sessionSpend.maxUsd)), 74, '  '))}`);
+    }
     /**
      * A total that assumed a cache-write rate is a floor, and says so.
      *