@rtrentjones/greenlight 0.5.1 → 0.6.0

package/dist/{agent-web-3FTO2TLJ.js → agent-web-BG5ZIVAB.js}

@@ -1,6 +1,6 @@
 import {
   verifyAgentWeb
-} from "./chunk-KVOI4UL2.js";
+} from "./chunk-IYEIZYI5.js";
 import "./chunk-QFKE5JKC.js";
 export {
   verifyAgentWeb

package/dist/bin.js

@@ -7,13 +7,14 @@ import {
   loadConfig,
   resolveUrl,
   scanSqlFiles,
+  toExportResult,
   verifyAll
-} from "./chunk-OBWWE7GE.js";
+} from "./chunk-FZH2YQPJ.js";
 import "./chunk-HX7VA25D.js";
 import "./chunk-N3IKUCSF.js";
 import "./chunk-KP3Y6WRU.js";
-import "./chunk-KVOI4UL2.js";
-import "./chunk-XWTOJHLV.js";
+import "./chunk-IYEIZYI5.js";
+import "./chunk-3A6F2JNP.js";
 import "./chunk-QFKE5JKC.js";
 
 // src/commands/add.ts
@@ -601,7 +602,7 @@ function tokensForTool(tool) {
 }
 
 // src/version.ts
-var MODULE_REF = "v0.5.1";
+var MODULE_REF = "v0.6.0";
 var MODULE_SOURCE_BASE = "git::https://github.com/RTrentJones/greenlight.git//infra/modules";
 function moduleSource(module, ref = MODULE_REF) {
   return `${MODULE_SOURCE_BASE}/${module}?ref=${ref}`;
@@ -2894,7 +2895,7 @@ import { setTimeout as sleep } from "timers/promises";
 
 // src/commands/verify.ts
 import { spawnSync } from "child_process";
-import { resolve as resolve9 } from "path";
+import { basename, resolve as resolve9 } from "path";
 function defaultSpec(lane) {
   switch (lane) {
     case "astro":
@@ -2918,21 +2919,34 @@ function defaultSpec(lane) {
       };
   }
 }
-function printReport(report) {
-  console.log(`verify ${report.mode} ${report.url}
+function printReport(report, log = console.log) {
+  log(`verify ${report.mode} ${report.url}
 `);
   for (const c of report.checks) {
-    console.log(`  ${c.pass ? "\u2714" : "\u2718"} ${c.name}${c.detail ? ` \u2014 ${c.detail}` : ""}`);
+    log(`  ${c.pass ? "\u2714" : "\u2718"} ${c.name}${c.detail ? ` \u2014 ${c.detail}` : ""}`);
   }
-  console.log(`
+  log(`
 ${report.pass ? "\u2714 PASS" : "\u2718 FAIL"}`);
   if (!report.pass && report.logs) {
-    console.log(`
+    log(`
 --- recent logs (${report.mode}) ---
 ${report.logs}
 --- end logs ---`);
   }
 }
+function emitReports(reports, json, ctx) {
+  const log = json ? console.error : console.log;
+  for (const report of reports) printReport(report, log);
+  const pass = allPass(reports);
+  if (reports.length > 1) log(`
+${pass ? "\u2714 ALL PASS" : "\u2718 FAIL"} (${reports.length} specs)`);
+  if (json) process.stdout.write(`${JSON.stringify(toExportResult(reports, ctx))}
+`);
+  process.exit(pass ? 0 : 1);
+}
+function gitSha() {
+  return process.env.VERCEL_GIT_COMMIT_SHA ?? process.env.GITHUB_SHA ?? null;
+}
 var LOG_TAIL_LINES = 50;
 function redactSecrets(text, env = process.env) {
   let out = text;
@@ -2973,6 +2987,9 @@ function flag6(args, name) {
   const i = args.indexOf(name);
   return i >= 0 ? args[i + 1] : void 0;
 }
+function jsonFlag(args) {
+  return args.includes("--json") || process.env.GREENLIGHT_VERIFY_JSON === "1";
+}
 async function verifyCommand(args) {
   const specPath = flag6(args, "--spec");
   if (specPath) {
@@ -2987,12 +3004,12 @@ async function verifyCommand(args) {
       toolDir: process.cwd()
     });
     attachFailureLogs(reports2, specs2, process.cwd());
-    for (const report of reports2) printReport(report);
-    const pass2 = allPass(reports2);
-    if (reports2.length > 1)
-      console.log(`
-${pass2 ? "\u2714 ALL PASS" : "\u2718 FAIL"} (${reports2.length} specs)`);
-    process.exit(pass2 ? 0 : 1);
+    const tool = flag6(args, "--tool") ?? basename(specPath).replace(/\.config\.[tj]s$/, "");
+    emitReports(reports2, jsonFlag(args), {
+      tool,
+      env: flag6(args, "--env") ?? "preview",
+      gitSha: gitSha()
+    });
   }
   const name = args[0];
   if (!name || name.startsWith("-")) {
@@ -3025,12 +3042,11 @@ ${pass2 ? "\u2714 ALL PASS" : "\u2718 FAIL"} (${reports2.length} specs)`);
   const toolDir = resolve9(process.cwd(), entry.dir ?? ".");
   const reports = await verifyAll(url, specs, { reachableTimeoutMs, toolDir });
   attachFailureLogs(reports, specs, toolDir);
-  for (const report of reports) printReport(report);
-  const pass = allPass(reports);
-  if (reports.length > 1)
-    console.log(`
-${pass ? "\u2714 ALL PASS" : "\u2718 FAIL"} (${reports.length} specs)`);
-  process.exit(pass ? 0 : 1);
+  emitReports(reports, jsonFlag(args), {
+    tool: entry.name ?? name,
+    env: override ? "preview" : flag6(args, "--env") ?? "preview",
+    gitSha: gitSha()
+  });
 }
 
 // src/commands/preview.ts
@@ -3387,7 +3403,7 @@ var HELP = `greenlight <command>
   config                                        load & validate the manifest, then print it
   deploy <name> --env <env>                     build + deploy an entry via its target adapter
   preview <name> [--port <n>]                   build + serve locally + verify (one command)
-  verify <name> [--env <env> | --url <url>]     run the verify harness against the URL
+  verify <name> [--env <env> | --url <url>] [--json]  run the verify harness (--json: standards-shaped result to stdout)
   promote <name> [--perform] [--push]           gated develop -> main fast-forward
   status <name>                                 last ship/deploy/verify run for a tool (via gh)
   secrets gather <name> [--repo o/r] [--env e]  guided, link-first token prompts -> GitHub secrets (no disk/logs)

package/dist/{chunk-XWTOJHLV.js → chunk-3A6F2JNP.js}

@@ -14,6 +14,10 @@ function resultText(res) {
   }
   return JSON.stringify(res);
 }
+var clamp01 = (n) => {
+  const v = typeof n === "number" ? n : Number(n);
+  return Number.isFinite(v) ? Math.min(1, Math.max(0, v)) : 0;
+};
 function llmJudge(model) {
   return async ({ rubric, result }) => {
     if (!process.env.ANTHROPIC_API_KEY) throw new Error("ANTHROPIC_API_KEY not set");
@@ -27,7 +31,7 @@ function llmJudge(model) {
     const resp = await client.messages.create({
       model,
       max_tokens: 512,
-      system: 'You are a strict evaluation judge. Score how well RESULT satisfies RUBRIC on a 1\u20135 scale (5 = fully satisfies). Reply ONLY with JSON: {"score": <1-5>, "pass": <bool>, "reason": "<short>"}.',
+      system: 'You are a strict evaluation judge. Score how well RESULT satisfies RUBRIC on a 0..1 scale (1 = fully satisfies). Reply ONLY with JSON: {"score": <0..1>, "pass": <bool>, "rationale": "<one sentence>"}.',
       messages: [{ role: "user", content: `RUBRIC:
 ${rubric}
 
@@ -38,12 +42,23 @@ ${result}` }]
     const json = text.match(/\{[\s\S]*\}/);
     if (!json) throw new Error(`judge returned no JSON: ${text.slice(0, 120)}`);
     const parsed = JSON.parse(json[0]);
-    return { score: Number(parsed.score) || 0, pass: Boolean(parsed.pass), reason: parsed.reason };
+    return {
+      score: clamp01(parsed.score),
+      pass: Boolean(parsed.pass),
+      rationale: parsed.rationale ?? parsed.reason,
+      // `reason` = deprecated alias, one release
+      tokensIn: resp.usage?.input_tokens,
+      tokensOut: resp.usage?.output_tokens
+    };
   };
 }
 async function verifyEval(baseUrl, spec, judge) {
-  const score = judge ?? llmJudge(spec.model ?? "claude-sonnet-4-6");
+  const model = spec.model ?? "claude-sonnet-4-6";
+  const score = judge ?? llmJudge(model);
   const checks = [];
+  const started = Date.now();
+  let tokensIn = 0;
+  let tokensOut = 0;
   const client = new Client({ name: "greenlight-verify", version: "0.0.0" });
   const transport = new StreamableHTTPClientTransport(new URL(baseUrl));
   try {
@@ -53,15 +68,22 @@ async function verifyEval(baseUrl, spec, judge) {
   }
   try {
     for (const c of spec.cases) {
-      const min = c.minScore ?? 4;
+      const min = c.minScore ?? 0.8;
       try {
         const res = await client.callTool({ name: c.tool, arguments: c.args ?? {} });
-        const verdict = await score({ rubric: c.rubric, result: resultText(res) });
+        const result = resultText(res);
+        const verdict = await score({ rubric: c.rubric, result });
         const pass = verdict.pass && verdict.score >= min;
+        const rationale = verdict.rationale ?? verdict.reason;
+        tokensIn += verdict.tokensIn ?? 0;
+        tokensOut += verdict.tokensOut ?? 0;
         checks.push({
           name: `eval: ${c.name}`,
           pass,
-          detail: `score ${verdict.score}/5 (min ${min})${verdict.reason ? ` \u2014 ${verdict.reason}` : ""}`
+          score: verdict.score,
+          explanation: rationale,
+          output: result,
+          detail: `score ${verdict.score.toFixed(2)} (min ${min})${rationale ? ` \u2014 ${rationale}` : ""}`
         });
       } catch (e) {
         checks.push({ name: `eval: ${c.name}`, pass: false, detail: msg(e) });
@@ -70,10 +92,16 @@ async function verifyEval(baseUrl, spec, judge) {
   } finally {
     await client.close();
   }
-  return report("eval", baseUrl, checks);
+  return {
+    ...report("eval", baseUrl, checks),
+    model,
+    durationMs: Date.now() - started,
+    ...tokensIn || tokensOut ? { tokensIn, tokensOut } : {}
+  };
 }
 
 export {
+  clamp01,
   llmJudge,
   verifyEval
 };

package/dist/{chunk-OBWWE7GE.js → chunk-FZH2YQPJ.js}

@@ -421,6 +421,53 @@ async function verifyApi(baseUrl, spec) {
   );
 }
 
+// ../packages/verify/src/export.ts
+var clamp01 = (n) => Math.min(1, Math.max(0, n));
+function sumDefined(xs) {
+  const present = xs.filter((x) => typeof x === "number");
+  return present.length ? present.reduce((a, b) => a + b, 0) : void 0;
+}
+function toExportResult(reports, ctx) {
+  const checks = [];
+  for (const r of reports) {
+    for (const c of r.checks) {
+      checks.push({
+        name: c.name,
+        passed: c.pass,
+        input: c.input ?? null,
+        expected: c.expected ?? null,
+        output: c.output ?? null,
+        "eval.score": c.score != null ? clamp01(c.score) : c.pass ? 1 : 0,
+        "eval.explanation": c.explanation ?? null
+      });
+    }
+  }
+  const passed = reports.length > 0 && reports.every((r) => r.pass);
+  const passRate = checks.length === 0 ? 0 : checks.filter((c) => c.passed).length / checks.length;
+  const model = reports.find((r) => r.model)?.model;
+  const tokensIn = sumDefined(reports.map((r) => r.tokensIn));
+  const tokensOut = sumDefined(reports.map((r) => r.tokensOut));
+  const cost = sumDefined(reports.map((r) => r.costUsd));
+  const durationMs = sumDefined(reports.map((r) => r.durationMs));
+  const attributes = {};
+  if (model) attributes["gen_ai.request.model"] = model;
+  if (tokensIn != null) attributes["gen_ai.usage.input_tokens"] = tokensIn;
+  if (tokensOut != null) attributes["gen_ai.usage.output_tokens"] = tokensOut;
+  if (cost != null) attributes["gen_ai.response.cost"] = cost;
+  return {
+    schemaVersion: "1",
+    tool: ctx.tool,
+    mode: reports.map((r) => r.mode).join("+") || "verify",
+    env: ctx.env,
+    git_sha: ctx.gitSha ?? null,
+    passed,
+    pass_rate: passRate,
+    duration_ms: durationMs ?? null,
+    ...Object.keys(attributes).length ? { attributes } : {},
+    checks
+  };
+}
+
 // ../packages/verify/src/index.ts
 function defineVerify(spec) {
   return spec;
@@ -456,11 +503,11 @@ async function verify(baseUrl, spec, opts) {
       return verifyTest2(spec, opts?.toolDir ?? process.cwd());
     }
     case "agent-web": {
-      const { verifyAgentWeb: verifyAgentWeb2 } = await import("./agent-web-3FTO2TLJ.js");
+      const { verifyAgentWeb: verifyAgentWeb2 } = await import("./agent-web-BG5ZIVAB.js");
       return verifyAgentWeb2(baseUrl, spec);
     }
     case "eval": {
-      const { verifyEval: verifyEval2 } = await import("./eval-44S2BATV.js");
+      const { verifyEval: verifyEval2 } = await import("./eval-YZXJSUKH.js");
       return verifyEval2(baseUrl, spec);
     }
   }
@@ -487,6 +534,7 @@ export {
   loadConfig,
   resolveUrl,
   scanSqlFiles,
+  toExportResult,
   defineVerify,
   verifyAll,
   allPass

package/dist/{chunk-KVOI4UL2.js → chunk-IYEIZYI5.js}

@@ -124,6 +124,8 @@ async function runScenario(client, page, base, spec, scenario) {
   const messages = [{ role: "user", content: `Task: ${scenario.task}` }];
   const maxSteps = spec.maxSteps ?? 12;
   let finish = null;
+  let tokensIn = 0;
+  let tokensOut = 0;
   for (let step = 0; step < maxSteps && !finish; step++) {
     const resp = await client.messages.create({
       model: spec.model ?? "claude-sonnet-4-6",
@@ -132,6 +134,8 @@ async function runScenario(client, page, base, spec, scenario) {
       tools: TOOLS,
       messages
     });
+    tokensIn += resp.usage?.input_tokens ?? 0;
+    tokensOut += resp.usage?.output_tokens ?? 0;
     const blocks = resp.content;
     messages.push({ role: "assistant", content: blocks });
     const toolUses = blocks.filter((b) => b.type === "tool_use");
@@ -160,7 +164,7 @@ async function runScenario(client, page, base, spec, scenario) {
     checks.push({ ...c, name: `${tag} ${c.name}` });
   }
   if (checks.length === 0) checks.push({ name: `${tag} agent succeeded`, pass: true });
-  return checks;
+  return { checks, tokensIn, tokensOut };
 }
 async function verifyAgentWeb(baseUrl, spec) {
   const base = baseUrl.replace(/\/+$/, "");
@@ -213,11 +217,17 @@ async function verifyAgentWeb(baseUrl, spec) {
     ]);
   }
   const checks = [];
+  const started = Date.now();
+  let tokensIn = 0;
+  let tokensOut = 0;
   try {
     for (const scenario of spec.scenarios) {
       const page = await browser.newPage();
       try {
-        checks.push(...await runScenario(client, page, base, spec, scenario));
+        const r = await runScenario(client, page, base, spec, scenario);
+        checks.push(...r.checks);
+        tokensIn += r.tokensIn;
+        tokensOut += r.tokensOut;
       } catch (e) {
         checks.push({ name: `[${scenario.name}]`, pass: false, detail: msg(e) });
       } finally {
@@ -227,7 +237,12 @@ async function verifyAgentWeb(baseUrl, spec) {
   } finally {
     await browser.close();
   }
-  return report("agent-web", baseUrl, checks);
+  return {
+    ...report("agent-web", baseUrl, checks),
+    model: spec.model ?? "claude-sonnet-4-6",
+    durationMs: Date.now() - started,
+    ...tokensIn || tokensOut ? { tokensIn, tokensOut } : {}
+  };
 }
 
 export {

package/dist/{eval-44S2BATV.js → eval-YZXJSUKH.js}

@@ -1,9 +1,11 @@
 import {
+  clamp01,
   llmJudge,
   verifyEval
-} from "./chunk-XWTOJHLV.js";
+} from "./chunk-3A6F2JNP.js";
 import "./chunk-QFKE5JKC.js";
 export {
+  clamp01,
   llmJudge,
   verifyEval
 };

package/dist/index.js

@@ -2,12 +2,12 @@ import {
   defineConfig,
   defineVerify,
   loadConfig
-} from "./chunk-OBWWE7GE.js";
+} from "./chunk-FZH2YQPJ.js";
 import "./chunk-HX7VA25D.js";
 import "./chunk-N3IKUCSF.js";
 import "./chunk-KP3Y6WRU.js";
-import "./chunk-KVOI4UL2.js";
-import "./chunk-XWTOJHLV.js";
+import "./chunk-IYEIZYI5.js";
+import "./chunk-3A6F2JNP.js";
 import "./chunk-QFKE5JKC.js";
 export {
   defineConfig,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rtrentjones/greenlight",
-  "version": "0.5.1",
+  "version": "0.6.0",
   "description": "Greenlight CLI — setup and lifecycle for the harness.",
   "license": "MIT",
   "repository": {
@@ -31,10 +31,10 @@
     "@anthropic-ai/sdk": "^0.69.0"
   },
   "devDependencies": {
-    "@rtrentjones/greenlight-adapters": "0.5.1",
-    "@rtrentjones/greenlight-loop": "0.5.1",
-    "@rtrentjones/greenlight-verify": "0.5.1",
-    "@rtrentjones/greenlight-shared": "0.5.1"
+    "@rtrentjones/greenlight-adapters": "0.6.0",
+    "@rtrentjones/greenlight-shared": "0.6.0",
+    "@rtrentjones/greenlight-verify": "0.6.0",
+    "@rtrentjones/greenlight-loop": "0.6.0"
   },
   "scripts": {
     "build": "node scripts/copy-assets.mjs && tsup",