@f-o-h/cli 0.1.62 → 0.1.64

package/README.md

@@ -159,6 +159,16 @@ create ephemeral orgs only for explicit lifecycle tests; do not create paid
 phone resources unless a BYON/customer-owned or operator-approved paid lane is
 selected.
 
+For no-spend evals, inspect reusable state before creating anything:
+
+```bash
+foh org status --json
+foh agent list --json
+```
+
+If a reusable eval agent already exists, configure and prove that agent rather
+than creating a second bronze-tier agent.
+
 ## Common Missions
 
 | Mission | Command |

package/dist/foh.js

@@ -15086,7 +15086,7 @@ function registerWidget(program3) {
       apiUrlOverride: opts.apiUrl,
       headers: { "x-agent-id": opts.agent }
     });
-    if (opts.json) {
+    if (resolveJsonMode({ json: opts.json })) {
       format(data, { json: true });
     } else {
       process.stdout.write(data.snippet + "\n");
@@ -32877,7 +32877,7 @@ var StdioServerTransport = class {
 };
 
 // src/lib/cli-version.ts
-var CLI_VERSION = "0.1.62";
+var CLI_VERSION = "0.1.64";
 
 // src/commands/mcp-serve.ts
 var DEFAULT_TIMEOUT_MS = 12e4;
@@ -34548,6 +34548,25 @@ function isMissingAgentTestsError(error2) {
   ].filter((value) => value !== void 0 && value !== null).join(" ").toLowerCase();
   return text.includes("no tests found") || text.includes("tests_not_found") || text.includes("agent_tests_not_configured");
 }
+function errorReasonCode(error2) {
+  const detail = error2.detail;
+  const code = detail && typeof detail.code === "string" ? detail.code : void 0;
+  const reasonCode = detail && typeof detail.reason_code === "string" ? detail.reason_code : void 0;
+  return error2.reasonCode ?? reasonCode ?? code;
+}
+function isAgentLimitReachedError(error2) {
+  const reasonCode = errorReasonCode(error2);
+  const text = [
+    reasonCode,
+    error2.error,
+    error2.detail?.error,
+    error2.detail?.message
+  ].filter((value) => value !== void 0 && value !== null).join(" ").toLowerCase();
+  return text.includes("agent_limit_reached") || text.includes("agent limit reached");
+}
+function shouldReuseSingleAgentForEval() {
+  return Boolean(process.env.FOH_EXTERNAL_AGENT_RUN_DIR) && isNoSpendPolicy();
+}
 function buildMissingOptionsPlan(missing, opts) {
   const missingFlags = missing.map(optionNameToFlag);
   const signInUrl = buildConsoleSignInUrl(resolveConsoleBaseUrl(opts.consoleUrl));
@@ -34903,7 +34922,8 @@ function registerSetup(program3) {
           orgId: opts.org,
           apiUrlOverride: opts.apiUrl
         });
-        const existing = agents.agents?.find((agent) => agent.name === opts.agentName);
+        const existingAgents = Array.isArray(agents.agents) ? agents.agents : [];
+        const existing = existingAgents.find((agent) => agent.name === opts.agentName);
         if (existing) {
           agentId = existing.id;
           return { step: "create_agent", status: "skipped", detail: "name already exists" };
@@ -34926,12 +34946,31 @@ function registerSetup(program3) {
           return { step: "create_agent", status: "done", detail: { agent_id: agentId, path: "template_apply" } };
         } catch (error2) {
           if (!(error2 instanceof FohError)) throw error2;
+          if (isAgentLimitReachedError(error2) && shouldReuseSingleAgentForEval() && existingAgents.length === 1) {
+            const reusable = existingAgents[0];
+            agentId = reusable.id;
+            return {
+              step: "create_agent",
+              status: "skipped",
+              detail: {
+                reason_code: "eval_agent_limit_reuse_existing_agent",
+                original_reason_code: errorReasonCode(error2) ?? "agent_limit_reached",
+                lifecycle_strategy: "reuse_existing_eval_state",
+                desired_agent_name: opts.agentName,
+                reused_agent_id: reusable.id,
+                reused_agent_name: reusable.name,
+                operator_note: "External-agent no-spend eval reused the single existing agent instead of creating a second bronze-tier agent."
+              }
+            };
+          }
           throw new FohError({
             step: "create_agent",
             error: `Template apply failed: ${error2.error}`,
             remediation: error2.remediation,
             statusCode: error2.statusCode,
-            detail: error2.detail
+            detail: error2.detail,
+            reasonCode: errorReasonCode(error2),
+            nextCommands: error2.nextCommands
           });
         }
       });
@@ -40328,7 +40367,7 @@ async function executeExternalAgentExecutorPlan(plan, options = {}) {
 var DEFAULT_PROMPT_VERSION = "blank-setup.v1";
 var DEFAULT_BATCH_MODELS = "openai/codex,anthropic/claude,cursor/agent";
 var PROMPTS = {
-  "blank-setup.v1": "Go to https://frontofhouse.okii.uk. Use only public docs, public API docs, and the public npm CLI package. Always invoke the CLI with `npx --yes @f-o-h/cli@latest ...`; do not use unpinned `npx @f-o-h/cli ...`, because cached older packages can produce invalid evidence. Install or verify the FOH CLI, authenticate or reach a deterministic auth blocker, then create or configure a Front Of House voice agent and website widget. Prefer the certification-oriented buyer templates: run `npx --yes @f-o-h/cli@latest templates list --category buyer --json` and use `UK Buyer Qualification` or `Viewing Booking` when available; do not use a greeting-only template for proof/certification. Prefer `npx --yes @f-o-h/cli@latest setup --phone-mode observe` for the free scaffold path: agent, widget, voice config, smoke test, certification, and publish readiness together. Treat phone-number purchasing as an explicit paid/scarce contact-path step, not part of high-volume eval setup. If `FOH_CLI_SPEND_POLICY=no_spend` is active and a command returns `paid_resource_blocked_by_spend_policy`, do not try to bypass it; continue widget/setup proof and report that exact reason code for the phone path. If the customer/operator explicitly owns a number and asks for real PSTN proof, use `npx --yes @f-o-h/cli@latest provision byon attach --phone-number <e164> --confirm-owned --json`; do not invent ownership or buy a FOH-owned number. Run proof/smoke/certification where available, including widget proof and voice proof. When running more than one `foh prove` mission for the same agent, pass `--proof-cache-dir .foh/proof-cache` so simulation certification can be shared instead of recomputed. If voice proof returns `contact_phone_missing` or `voice_contact_expected_no_spend_hold`, report that exact reason code unless a BYON/customer-approved phone path already exists. If `FOH_EXTERNAL_AGENT_RUN_DIR` is set, write `${FOH_EXTERNAL_AGENT_RUN_DIR}/external-agent-metadata.json` with `schema_version`, `docs_pages_used`, key decisions, and blocker reason codes before finishing. Produce a final evidence summary with commands run, docs used, artifacts created, and any blocker reason codes. Do not assume access to the private source repository.",
+  "blank-setup.v1": "Go to https://frontofhouse.okii.uk. Use only public docs, public API docs, and the public npm CLI package. Always invoke the CLI with `npx --yes @f-o-h/cli@latest ...`; do not use unpinned `npx @f-o-h/cli ...`, because cached older packages can produce invalid evidence. Install or verify the FOH CLI, authenticate or reach a deterministic auth blocker, then create or configure a Front Of House voice agent and website widget. Mass evals reuse existing eval state: run `npx --yes @f-o-h/cli@latest org status --json` and `npx --yes @f-o-h/cli@latest agent list --json` before trying to create a fresh agent; if an existing eval agent is present, configure and prove that agent instead of creating a second bronze-tier agent. Prefer the certification-oriented buyer templates: run `npx --yes @f-o-h/cli@latest templates list --category buyer --json` and use `UK Buyer Qualification` or `Viewing Booking` when available; do not use a greeting-only template for proof/certification. Prefer `npx --yes @f-o-h/cli@latest setup --phone-mode observe` for the free scaffold path: agent, widget, voice config, smoke test, certification, and publish readiness together. Treat phone-number purchasing as an explicit paid/scarce contact-path step, not part of high-volume eval setup. If `FOH_CLI_SPEND_POLICY=no_spend` is active and a command returns `paid_resource_blocked_by_spend_policy`, do not try to bypass it; continue widget/setup proof and report that exact reason code for the phone path. If the customer/operator explicitly owns a number and asks for real PSTN proof, use `npx --yes @f-o-h/cli@latest provision byon attach --phone-number <e164> --confirm-owned --json`; do not invent ownership or buy a FOH-owned number. Run proof/smoke/certification where available, including widget proof and voice proof. When running more than one `foh prove` mission for the same agent, pass `--proof-cache-dir .foh/proof-cache` so simulation certification can be shared instead of recomputed. If voice proof returns `contact_phone_missing` or `voice_contact_expected_no_spend_hold`, report that exact reason code unless a BYON/customer-approved phone path already exists. If `FOH_EXTERNAL_AGENT_RUN_DIR` is set, write `${FOH_EXTERNAL_AGENT_RUN_DIR}/external-agent-metadata.json` with `schema_version`, `docs_pages_used`, key decisions, and blocker reason codes before finishing. Produce a final evidence summary with commands run, docs used, artifacts created, and any blocker reason codes. Do not assume access to the private source repository.",
   "debug-proof-failure.v1": "You are given a FOH proof or debug artifact. Use public docs and FOH CLI/API behavior to classify whether the blocker is docs, auth, org setup, agent config, widget, channel, runtime, or product bug. Produce a redacted improvement packet or the exact command needed to produce one. Do not ask the human to interpret logs manually unless no machine-readable artifact exists.",
   "knowledge-miss.v1": "A FOH agent failed to answer a business question. Use CLI/API/docs to determine whether this is a knowledge-ingestion issue, retrieval issue, config issue, prompt/behavior issue, or runtime issue. Prefer foh knowledge query, transcript export, replay, and foh bug improve artifacts over screenshots.",
   "replay-failure.v1": "You are given a FOH transcript or replay artifact. Use CLI/API/docs to replay or inspect the failed interaction, identify expected vs actual behavior, and produce a scenario-test or improvement-packet candidate."

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@f-o-h/cli",
-  "version": "0.1.62",
+  "version": "0.1.64",
   "description": "FOH CLI - AI-operator provisioning tool for Front Of House",
   "license": "UNLICENSED",
   "bin": {