apiblaze 0.21.75 → 0.21.76

package/dist/index.js

@@ -34,7 +34,7 @@ var __toESM = (mod, isNodeMode, target2) => (target2 = mod != null ? __create(__
 var version;
 var init_package = __esm({
   "package.json"() {
-    version = "0.21.75";
+    version = "0.21.76";
   }
 });
 
@@ -5577,6 +5577,7 @@ __export(directory_exports, {
   stepOpen: () => stepOpen,
   stepRunnable: () => stepRunnable,
   summaryOf: () => summaryOf,
+  testCasesText: () => testCasesText,
   withBrandSource: () => withBrandSource,
   writeJsonLine: () => writeJsonLine
 });
@@ -5675,6 +5676,22 @@ function directoryBodyFromFlags(f2) {
   }
   return out;
 }
+function testCasesText(r, platforms) {
+  const cases = r?.target?.metadata?.portalAnswers?.test_cases ?? [];
+  if (!cases.length) return null;
+  const dirs = ["claude", "chatgpt"].filter((x) => platforms.includes(x));
+  const lines = ["Test cases (what reviewers try; results are our pre-check, not their review)"];
+  for (const c of cases) {
+    lines.push(`  "${c.prompt}" (${c.kind === "refuse" ? "should say no" : c.tool ? `should use ${c.tool}` : "pick the tool it should use"})`);
+    for (const dir of dirs) {
+      const x = (r?.case_results ?? []).find((y) => y.caseId === c.id && y.directory === dir);
+      const word = !x ? "not run yet" : !x.current ? "out of date, run again" : `${x.pass ? "\u2713" : "\u2717"} ${CASE_STATE[x.state]}${x.state === "other_tool" || x.state === "acted" ? `: ${x.tools.join(", ")}` : ""} \xB7 ${CASE_SOURCE[x.source]}, ${x.at.slice(0, 10)}`;
+      lines.push(`     for ${dir === "claude" ? "Claude" : "ChatGPT"}: ${word}`);
+    }
+  }
+  lines.push("  Run them on the Testing tab, or with the Claude Code lines below.");
+  return lines.join("\n");
+}
 function withBrandSource(url, cli = B.cli) {
   return cli === "mcpblaze" ? `${url}?source=mcpblazeCli` : url;
 }
@@ -5821,6 +5838,7 @@ function claudeCodeTests(r, proj2) {
     seen.add(p);
     cases.push({ prompt: p, line: askLine(name, p), expect, refuse });
   };
+  for (const c of t?.metadata?.portalAnswers?.test_cases ?? []) add(c.prompt, c.kind === "refuse" ? `it says it can't help${c.why ? ` (${c.why})` : ""}, and calls no tool` : c.expect || "it calls your tools and answers from what comes back", c.kind === "refuse");
   for (const p of t?.metadata?.examplePrompts ?? []) add(p, "it calls your tools and answers from what comes back", false);
   const tc = t?.metadata?.portalAnswers?.chatgpt_test_cases;
   for (const c of tc?.positive ?? []) add(c.prompt, c.expected_behavior || "it calls your tools and answers from what comes back", false);
@@ -5840,7 +5858,7 @@ function claudeCodeText(cc) {
   else lines.push(`  ${n}. One line per use case:`, ...cc.cases.flatMap((c) => [`     ${c.line}`, `       expect: ${c.expect}`]));
   return lines.join("\n");
 }
-var import_node_fs, import_chalk17, PLATFORMS, ATTESTATIONS, OPENAI_CATEGORIES, AUTH_MODES, DATA_HANDLING_MODES, STEP_IDS, reportPath, DESCRIPTION_MIN, ccServerName, shellQuote2, askLine;
+var import_node_fs, import_chalk17, PLATFORMS, ATTESTATIONS, OPENAI_CATEGORIES, AUTH_MODES, DATA_HANDLING_MODES, STEP_IDS, CASE_SOURCE, CASE_STATE, reportPath, DESCRIPTION_MIN, ccServerName, shellQuote2, askLine;
 var init_directory = __esm({
   "src/lib/directory.ts"() {
     "use strict";
@@ -5872,6 +5890,8 @@ var init_directory = __esm({
     AUTH_MODES = ["signin", "open"];
     DATA_HANDLING_MODES = ["own", "proxied", "third-party"];
     STEP_IDS = ["mode", "address", "access", "tools", "listing", "answers", "attestations", "prove", "signin", "token", "packet", "submit"];
+    CASE_SOURCE = { "our-claude": "our test, Claude model", "our-gpt": "our test, GPT model", "claude-code": "your Claude Code", codex: "your Codex CLI", chatgpt: "ChatGPT", claude: "Claude" };
+    CASE_STATE = { passed: "used the right tool", chose_write: "chose the right tool (not run: it changes data)", other_tool: "used a different tool", no_tool: "didn't use a tool", tool_failed: "the tool failed", said_no: "said no", acted: "used a tool when it shouldn't" };
     reportPath = (proj2) => `/projects/${encodeURIComponent(proj2.projectId)}/${encodeURIComponent(proj2.apiVersion)}/directory-report`;
     DESCRIPTION_MIN = 300;
     ccServerName = (projectId) => projectId.toLowerCase().replace(/[^a-z0-9_-]+/g, "-").replace(/^-+|-+$/g, "") || "app";
@@ -8650,8 +8670,12 @@ async function runConnectorExercise(project, opts = {}) {
     const codex = deps.codexInstalled !== void 0 ? { installed: deps.codexInstalled, loggedIn: deps.codexInstalled } : clients.includes("codex") ? detectCodex(deps.spawn ?? import_child_process5.spawnSync) : { installed: false, loggedIn: false };
     const proven = opts.provenPromptNotes ?? [];
     const provenFor = (client) => new Map(proven.filter((n) => n.kind === "prompt" && (n.client ?? "claude") === client).map((n) => [n.prompt.trim(), n]));
-    const passPrompts = (opts.promptsOnly ?? target2.metadata?.examplePrompts ?? []).filter((p) => typeof p === "string" && p.trim()).slice(0, 5);
-    const negatives = (target2.metadata?.portalAnswers?.chatgpt_test_cases?.negative ?? []).map((c) => c.prompt).filter((p) => typeof p === "string" && p.trim()).slice(0, 3);
+    const testCases = target2.metadata?.portalAnswers?.test_cases ?? [];
+    const readTools = new Set(readOnlyToolNames(tools));
+    const casePrompts = testCases.filter((c) => c.kind === "use" && (!c.tool || readTools.has(c.tool))).map((c) => c.prompt);
+    const passPrompts = opts.promptsOnly ?? [...new Set([...(target2.metadata?.examplePrompts ?? []).slice(0, 5), ...casePrompts].filter((p) => typeof p === "string" && p.trim()).map((p) => p.trim()))].slice(0, 10);
+    const refusals = testCases.filter((c) => c.kind === "refuse");
+    const negatives = (refusals.length ? [...refusals.filter((c) => c.send), ...refusals.filter((c) => !c.send)].map((c) => c.prompt) : (target2.metadata?.portalAnswers?.chatgpt_test_cases?.negative ?? []).map((c) => c.prompt)).filter((p) => typeof p === "string" && p.trim()).slice(0, 3);
     const readOnly2 = readOnlyToolNames(tools);
     const requiredCount = (t) => Object.values(t.inputProperties ?? {}).filter((p) => p.required).length;
     const firstRead = [...tools].filter((t) => toolRole(t) === "read").sort((a, b) => requiredCount(a) - requiredCount(b))[0] ?? tools[0];
@@ -13956,12 +13980,15 @@ async function runConnector(target2, opts) {
     if (opts.status) {
       const cc = claudeCodeTests(r, proj2);
       if (json) {
-        writeJsonLine({ ready: isReady(r, platforms), project: proj2.projectId, headline: r.rendered.headline, summary: r.progress?.summary ?? null, progress: r.progress ?? null, claude_code: cc, report: r.report });
+        writeJsonLine({ ready: isReady(r, platforms), project: proj2.projectId, headline: r.rendered.headline, summary: r.progress?.summary ?? null, progress: r.progress ?? null, claude_code: cc, test_cases: r.case_results ?? null, report: r.report });
         process.exit(isReady(r, platforms) ? 0 : 1);
       }
       const card2 = scorecardText(r);
       if (card2) console.log(card2);
       else printReport(r, false);
+      const tc = testCasesText(r, platforms);
+      if (tc) console.log(`
+${tc}`);
       if (cc) console.log(`
 ${claudeCodeText(cc)}`);
       process.exit(isReady(r, platforms) ? 0 : 1);
@@ -15857,8 +15884,11 @@ saved as it is given; \`npx ${cli} connector <name> --status\` says what is left
    and \`npx ${cli} connector answers <name> --use-cases "\u2026" --prerequisites "\u2026" --category "\u2026" --reviewer-instructions "\u2026"\`.
 6. **Test**: \`npx ${cli} connector exercise <name>\` calls every tool as the reviewer. Then
    \`npx ${cli} connector <name> --status\` prints the Claude Code lines: one \`claude mcp add\`, one sign-in, and one
-   \`claude -p '\u2026' --allowedTools 'mcp__<name>'\` per example prompt, each with what should come back. Run each, tell the user
+   \`claude -p '\u2026' --allowedTools 'mcp__<name>'\` per test case and example prompt, each with what should come back. Run each, tell the user
    what came back, and fix a tool's description (\`npx ${cli} mcp ops set\`) when a prompt did not reach the right tool.
+   \`--status\` also lists the **test cases** (what reviewers try: five that should use a tool and three that should say no
+   for ChatGPT) with each one's latest result and where it ran. Those results are a pre-check, not the directories' review:
+   never tell the user a case "passed in ChatGPT" or "in Claude". The user writes and runs the cases on the page's Testing tab.
 7. **Send**: when every check passes it writes the packet. The user submits it on claude.ai/directory/manage (and the
    OpenAI portal for ChatGPT).
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "apiblaze",
-  "version": "0.21.75",
+  "version": "0.21.76",
   "description": "APIblaze CLI — Chat with your APIs, Manage your API keys, users and groups with the APIblaze serverless proxy",
   "keywords": [
     "apiblaze",