trickle-cli 0.1.189 → 0.1.190

package/dist/commands/eval.d.ts

@@ -13,4 +13,5 @@
  */
 export declare function evalCommand(opts: {
     json?: boolean;
+    failUnder?: string;
 }): void;

package/dist/commands/eval.js

@@ -76,7 +76,15 @@ function evalCommand(opts) {
     }
     const result = scoreRun(agentEvents, llmCalls, errors, mcpCalls);
     if (opts.json) {
-        console.log(JSON.stringify(result, null, 2));
+        const threshold = opts.failUnder ? parseInt(opts.failUnder, 10) : undefined;
+        const output = {
+            ...result,
+            ...(threshold !== undefined ? { threshold, passed: result.overallScore >= threshold } : {}),
+        };
+        console.log(JSON.stringify(output, null, 2));
+        if (threshold !== undefined && result.overallScore < threshold) {
+            process.exit(1);
+        }
         return;
     }
     // Pretty print
@@ -104,6 +112,14 @@ function evalCommand(opts) {
         }
     }
     console.log('');
+    // CI mode: exit with non-zero if score below threshold
+    if (opts.failUnder) {
+        const threshold = parseInt(opts.failUnder, 10);
+        if (!isNaN(threshold) && result.overallScore < threshold) {
+            console.log(chalk_1.default.red(`  FAIL: Score ${result.overallScore} is below threshold ${threshold}`));
+            process.exit(1);
+        }
+    }
 }
 function printDimension(name, dim) {
     const bar = renderBar(dim.score);

package/dist/index.js

@@ -918,6 +918,7 @@ program
     .command("eval")
     .description("Score agent runs on reliability — completion, errors, cost efficiency, tool reliability, latency")
     .option("--json", "Output raw JSON for CI integration")
+    .option("--fail-under <score>", "Exit with code 1 if overall score is below this threshold (0-100, for CI)")
     .action(async (opts) => {
     const { evalCommand } = await Promise.resolve().then(() => __importStar(require("./commands/eval")));
     evalCommand(opts);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "trickle-cli",
-  "version": "0.1.189",
+  "version": "0.1.190",
   "description": "CLI for trickle runtime type observability",
   "bin": {
     "trickle": "dist/index.js"

package/src/commands/eval.ts

@@ -36,7 +36,7 @@ function readJsonl(fp: string): any[] {
     .map(l => { try { return JSON.parse(l); } catch { return null; } }).filter(Boolean);
 }
 
-export function evalCommand(opts: { json?: boolean }): void {
+export function evalCommand(opts: { json?: boolean; failUnder?: string }): void {
   const dir = process.env.TRICKLE_LOCAL_DIR || path.join(process.cwd(), '.trickle');
   const agentEvents = readJsonl(path.join(dir, 'agents.jsonl'));
   const llmCalls = readJsonl(path.join(dir, 'llm.jsonl'));
@@ -51,7 +51,15 @@ export function evalCommand(opts: { json?: boolean }): void {
   const result = scoreRun(agentEvents, llmCalls, errors, mcpCalls);
 
   if (opts.json) {
-    console.log(JSON.stringify(result, null, 2));
+    const threshold = opts.failUnder ? parseInt(opts.failUnder, 10) : undefined;
+    const output = {
+      ...result,
+      ...(threshold !== undefined ? { threshold, passed: result.overallScore >= threshold } : {}),
+    };
+    console.log(JSON.stringify(output, null, 2));
+    if (threshold !== undefined && result.overallScore < threshold) {
+      process.exit(1);
+    }
     return;
   }
 
@@ -85,6 +93,15 @@ export function evalCommand(opts: { json?: boolean }): void {
   }
 
   console.log('');
+
+  // CI mode: exit with non-zero if score below threshold
+  if (opts.failUnder) {
+    const threshold = parseInt(opts.failUnder, 10);
+    if (!isNaN(threshold) && result.overallScore < threshold) {
+      console.log(chalk.red(`  FAIL: Score ${result.overallScore} is below threshold ${threshold}`));
+      process.exit(1);
+    }
+  }
 }
 
 function printDimension(name: string, dim: { score: number; detail: string }): void {

package/src/index.ts

@@ -951,6 +951,7 @@ program
   .command("eval")
   .description("Score agent runs on reliability — completion, errors, cost efficiency, tool reliability, latency")
   .option("--json", "Output raw JSON for CI integration")
+  .option("--fail-under <score>", "Exit with code 1 if overall score is below this threshold (0-100, for CI)")
   .action(async (opts) => {
     const { evalCommand } = await import("./commands/eval");
     evalCommand(opts);