mira-harness 0.2.0 → 0.2.1

package/README.md

@@ -101,7 +101,7 @@ a "typing…" fallback for a slow bot — replies run 5–62s) and capture, per
 | `login` | One-time interactive login → prints `TG_SESSION` |
 | `doctor` | Check `.env` / session / connectivity / @mira resolution (read-only) |
 | `send [message...]` | One probe → full reply as JSON (message via arg or stdin). `--quiet --settle --timeout --no-log` |
-| `loop` | Run the catalog paced. `--category --max --confirm --peer --gap --settle --timeout --list --catalog --quiet` |
+| `loop` | Run the catalog paced; grades `expect` probes (exit 1 on failure). `--category --max --confirm --peer --gap --settle --timeout --list --catalog --no-fail --quiet` |
 | `catalog` | List the catalog (no sends). `--category --catalog --json` |
 | `watch` | Live-tail @mira's messages (observe-only). `--peer` |
 | `report` | Distill the run log into Markdown. `--in --out --category` |
@@ -130,13 +130,35 @@ per-feature flags — `MIRA_NO_BANNER=1` (mascot + tip), `MIRA_NO_NOTIFY=1` (com
 The built-in catalog (27 probes: `core` / `skills` / `generation` / `wallet`) is just a
 default. Point `--catalog <file.json>` (CLI) or `catalogFile` (MCP) at your own probe set
 to probe any bot — each entry needs `id` + `send` (`category` / `hypothesis` / `slow` /
-`confirm` / `note` optional). See [`examples/catalog.sample.json`](examples/catalog.sample.json):
+`confirm` / `note` / `expect` optional). See [`examples/catalog.sample.json`](examples/catalog.sample.json):
 
 ```bash
 mira-harness loop --catalog ./examples/catalog.sample.json
 mira-harness catalog --catalog ./examples/catalog.sample.json --json
 ```
 
+### Assertions (PASS/FAIL)
+
+Give a probe an optional `expect` block and `loop` grades it ✓/✗. The checks are **structural**
+— @mira is an LLM (non-deterministic), so exact-text matches would flake:
+
+| Check | Means |
+|---|---|
+| `replies: true` | a reply arrived (no timeout) |
+| `textMatches: "<regex>"` | some message text matches (case-insensitive) |
+| `minButtons` / `minLinks` | at least N inline buttons / links across messages |
+| `hasWebApp: true` | a Mini App (`web_app` / startapp) "Launch" button is present |
+| `media: "photo"` | a message carries media of that kind (`photo`/`video`/`audio`/…) |
+| `maxFirstReplyMs` | first-reply latency within the bound |
+| `json: true` | the first message text parses as JSON |
+
+```json
+{ "id": "json-strict", "send": "Reply with ONLY {\"ok\":true}", "expect": { "json": true } }
+```
+
+Probes without `expect` stay observe-only (informational). `loop` **exits non-zero** if any
+graded probe fails — so it drops straight into CI. Add `--no-fail` to report without failing.
+
 ## Use as a library
 
 The CLI is a thin frontend over an exported core:

package/dist/assert.d.ts

@@ -0,0 +1,41 @@
+/**
+ * Probe assertions — turn an observed reply into a PASS/FAIL verdict.
+ *
+ * A probe's optional `expect` declares machine-checkable expectations; `evaluate`
+ * runs them against the captured ProbeResult. @mira is an LLM (non-deterministic),
+ * so the checks are deliberately STRUCTURAL and loose — a reply arrived, >= N
+ * links, valid JSON, a latency bound — never exact text, which would flake.
+ *
+ * Probes WITHOUT `expect` are not graded: they stay observe-only / informational.
+ */
+import { z } from "zod";
+import type { ProbeResult } from "./capture.js";
+export declare const ExpectSchema: z.ZodObject<{
+    replies: z.ZodOptional<z.ZodBoolean>;
+    textMatches: z.ZodOptional<z.ZodString>;
+    minButtons: z.ZodOptional<z.ZodNumber>;
+    minLinks: z.ZodOptional<z.ZodNumber>;
+    hasWebApp: z.ZodOptional<z.ZodBoolean>;
+    media: z.ZodOptional<z.ZodEnum<{
+        photo: "photo";
+        video: "video";
+        audio: "audio";
+        document: "document";
+        webpage: "webpage";
+        other: "other";
+    }>>;
+    maxFirstReplyMs: z.ZodOptional<z.ZodNumber>;
+    json: z.ZodOptional<z.ZodBoolean>;
+}, z.core.$strip>;
+export type Expect = z.infer<typeof ExpectSchema>;
+export interface Check {
+    name: string;
+    ok: boolean;
+    detail: string;
+}
+export interface Verdict {
+    ok: boolean;
+    checks: Check[];
+}
+/** Run a probe's expectations against its captured result. Pure — no network. */
+export declare function evaluate(expect: Expect, result: ProbeResult): Verdict;

package/dist/catalog.d.ts

@@ -1,3 +1,4 @@
+import { type Expect } from "./assert.js";
 /** Built-in categories. Custom catalogs may use any category string. */
 export type ProbeCategory = "core" | "skills" | "generation" | "wallet";
 export interface Probe {
@@ -15,6 +16,8 @@ export interface Probe {
      */
     confirm?: boolean;
     note?: string;
+    /** Optional machine-checkable expectations — graded by `loop` (PASS/FAIL). */
+    expect?: Expect;
 }
 export declare const CATEGORIES: ProbeCategory[];
 /**

package/dist/cli.js

@@ -6,16 +6,87 @@ import { Command } from "commander";
 // src/catalog.ts
 import { readFileSync } from "node:fs";
 import { resolve } from "node:path";
+import { z as z2 } from "zod";
+
+// src/assert.ts
 import { z } from "zod";
+var ExpectSchema = z.object({
+  replies: z.boolean().optional(),
+  textMatches: z.string().optional(),
+  minButtons: z.number().int().nonnegative().optional(),
+  minLinks: z.number().int().nonnegative().optional(),
+  hasWebApp: z.boolean().optional(),
+  media: z.enum(["photo", "video", "audio", "document", "webpage", "other"]).optional(),
+  maxFirstReplyMs: z.number().positive().optional(),
+  json: z.boolean().optional()
+});
+function evaluate(expect, result) {
+  const checks = [];
+  const texts = result.messages.map((m) => m.text).filter((t) => t.length > 0);
+  const firstText = texts[0] ?? "";
+  const buttons = result.messages.reduce((n, m) => n + m.buttons.length, 0);
+  const links = result.messages.reduce((n, m) => n + m.links.length, 0);
+  const webApp = result.messages.some((m) => m.buttons.some((b) => b.webAppUrl !== undefined));
+  const mediaKinds = result.messages.map((m) => m.media?.kind).filter((k) => k !== undefined);
+  const add = (name, ok, detail) => {
+    checks.push({ name, ok, detail });
+  };
+  if (expect.replies !== undefined) {
+    const replied = !result.timedOut && result.messages.length > 0;
+    add("replies", replied === expect.replies, result.timedOut ? "timed out" : `${result.messages.length} message(s)`);
+  }
+  if (expect.textMatches !== undefined) {
+    const pattern = expect.textMatches;
+    let compiled;
+    try {
+      compiled = new RegExp(pattern, "i");
+    } catch {
+      compiled = undefined;
+    }
+    const rx = compiled;
+    const ok = rx ? texts.some((t) => rx.test(t)) : false;
+    add("textMatches", ok, rx ? `/${pattern}/i` : `invalid regex: ${pattern}`);
+  }
+  if (expect.minButtons !== undefined) {
+    add("minButtons", buttons >= expect.minButtons, `${buttons} >= ${expect.minButtons}`);
+  }
+  if (expect.minLinks !== undefined) {
+    add("minLinks", links >= expect.minLinks, `${links} >= ${expect.minLinks}`);
+  }
+  if (expect.hasWebApp !== undefined) {
+    add("hasWebApp", webApp === expect.hasWebApp, `webApp=${webApp}`);
+  }
+  if (expect.media !== undefined) {
+    add("media", mediaKinds.includes(expect.media), `kinds=[${mediaKinds.join(",")}] want ${expect.media}`);
+  }
+  if (expect.maxFirstReplyMs !== undefined) {
+    const ms = result.firstReplyMs;
+    add("maxFirstReplyMs", ms !== null && ms <= expect.maxFirstReplyMs, ms === null ? "no reply" : `${ms}ms <= ${expect.maxFirstReplyMs}ms`);
+  }
+  if (expect.json === true) {
+    let ok = false;
+    try {
+      JSON.parse(firstText);
+      ok = firstText.length > 0;
+    } catch {
+      ok = false;
+    }
+    add("json", ok, ok ? "valid JSON" : "not valid JSON");
+  }
+  return { ok: checks.every((ch) => ch.ok), checks };
+}
+
+// src/catalog.ts
 var CATEGORIES = ["core", "skills", "generation", "wallet"];
-var ProbeSchema = z.object({
-  id: z.string().min(1),
-  category: z.string().default("custom"),
-  hypothesis: z.string().default(""),
-  send: z.string().min(1),
-  slow: z.boolean().optional(),
-  confirm: z.boolean().optional(),
-  note: z.string().optional()
+var ProbeSchema = z2.object({
+  id: z2.string().min(1),
+  category: z2.string().default("custom"),
+  hypothesis: z2.string().default(""),
+  send: z2.string().min(1),
+  slow: z2.boolean().optional(),
+  confirm: z2.boolean().optional(),
+  note: z2.string().optional(),
+  expect: ExpectSchema.optional()
 });
 function loadCatalog(file) {
   const path = resolve(process.cwd(), file);
@@ -26,7 +97,7 @@ function loadCatalog(file) {
   } catch {
     throw new Error(`catalog ${file} is not valid JSON`);
   }
-  const parsed = z.array(ProbeSchema).safeParse(data);
+  const parsed = z2.array(ProbeSchema).safeParse(data);
   if (!parsed.success) {
     throw new Error(`invalid catalog ${file}: ${parsed.error.issues[0]?.message ?? "schema error"}`);
   }
@@ -39,7 +110,8 @@ var CATALOG = [
     id: "core-model",
     category: "core",
     hypothesis: "Which underlying model is Mira on? (docs say MiniMax M2.5)",
-    send: "What AI model are you running on right now? Answer with just the model name."
+    send: "What AI model are you running on right now? Answer with just the model name.",
+    expect: { replies: true }
   },
   {
     id: "core-mem-set",
@@ -51,13 +123,15 @@ var CATALOG = [
     id: "core-mem-get",
     category: "core",
     hypothesis: "Does the fact from core-mem-set persist across separate sends?",
-    send: "What did I tell you my favorite TON token is? Answer with one word."
+    send: "What did I tell you my favorite TON token is? Answer with one word.",
+    expect: { replies: true }
   },
   {
     id: "core-json",
     category: "core",
     hypothesis: "Will Mira return STRICT JSON only (usable for structured workflows)?",
-    send: 'Reply with ONLY this JSON and nothing else: {"ok":true,"n":42}'
+    send: 'Reply with ONLY this JSON and nothing else: {"ok":true,"n":42}',
+    expect: { json: true }
   },
   {
     id: "core-reason",
@@ -69,7 +143,8 @@ var CATALOG = [
     id: "core-commands",
     category: "core",
     hypothesis: "What commands does Mira expose? (/help surface, buttons)",
-    send: "/help"
+    send: "/help",
+    expect: { replies: true }
   },
   {
     id: "core-ja",
@@ -81,7 +156,8 @@ var CATALOG = [
     id: "skill-capabilities",
     category: "skills",
     hypothesis: "Does Mira enumerate its capabilities and surface buttons/links?",
-    send: "What can you do? List your main capabilities."
+    send: "What can you do? List your main capabilities.",
+    expect: { replies: true }
   },
   {
     id: "skill-research",
@@ -89,7 +165,8 @@ var CATALOG = [
     hypothesis: "Web research returns source links (text_url) the harness can capture.",
     send: "Search the web for the latest TON ecosystem news and give me 3 source links.",
     slow: true,
-    note: "expect links[] (text_url source links) from deep research"
+    note: "expect links[] (text_url source links) from deep research",
+    expect: { minLinks: 1 }
   },
   {
     id: "skill-translate",
@@ -116,7 +193,8 @@ var CATALOG = [
     send: "Generate an image: a friendly teal robot mascot holding a glowing TON diamond, flat vector style.",
     slow: true,
     confirm: true,
-    note: "expect a Confirm card; with --confirm the runner presses ✅ Confirm -> media.kind=photo"
+    note: "expect a Confirm card; with --confirm the runner presses ✅ Confirm -> media.kind=photo",
+    expect: { replies: true, minButtons: 1 }
   },
   {
     id: "gen-music",
@@ -340,7 +418,8 @@ function listCatalog(opts = {}) {
       hypothesis: p.hypothesis,
       send: p.send,
       slow: Boolean(p.slow),
-      confirm: Boolean(p.confirm)
+      confirm: Boolean(p.confirm),
+      expect: p.expect ?? null
     })), null, 2));
     return;
   }
@@ -751,13 +830,15 @@ function findConfirmButton(result) {
   }
   return;
 }
-function summarize(id, r) {
+function summarize(id, r, verdict) {
   const btns = r.messages.reduce((n, m) => n + m.buttons.length, 0);
   const links = r.messages.reduce((n, m) => n + m.links.length, 0);
   const media = r.messages.filter((m) => m.media).map((m) => m.media?.kind).join(",");
   const head = r.timedOut ? c.yellow("TIMEOUT") : c.green(`${r.messages.length}msg`);
   const latency = r.firstReplyMs === null ? c.dim("—") : c.dim(`${(r.firstReplyMs / 1000).toFixed(1)}s`);
+  const mark = verdict ? verdict.ok ? c.green("✓") : c.red("✗") : c.dim("·");
   const parts = [
+    mark,
     c.cyan(id),
     head,
     latency,
@@ -809,6 +890,8 @@ async function loop(opts) {
   note(c.bold(`running ${probes.length} probe(s)${opts.category ? ` [${opts.category}]` : ""} -> @${peer}` + c.dim(`  gap ${gap / 1000}s${opts.confirm ? "  --confirm ON" : ""}`)));
   const client = await connect(session);
   let ran = 0;
+  let graded = 0;
+  let failed = 0;
   try {
     for (const [i, p] of probes.entries()) {
       if (existsSync(STOP_FILE)) {
@@ -818,9 +901,25 @@ async function loop(opts) {
       if (!opts.quiet)
         setTitle(`mira loop ${i + 1}/${probes.length} · ${p.category}`);
       const result = await withProgress(`${p.id} -> @${peer}`, () => sendAndCollect(client, peer, p.send, collectFor(p, opts)), opts.quiet);
-      await appendRun(result, { probeId: p.id, category: p.category, hypothesis: p.hypothesis });
+      const verdict = p.expect ? evaluate(p.expect, result) : undefined;
+      await appendRun(result, {
+        probeId: p.id,
+        category: p.category,
+        hypothesis: p.hypothesis,
+        ...verdict ? { assert: verdict } : {}
+      });
       ran += 1;
-      console.log(summarize(p.id, result));
+      if (verdict) {
+        graded += 1;
+        if (!verdict.ok)
+          failed += 1;
+      }
+      console.log(summarize(p.id, result, verdict));
+      if (verdict && !verdict.ok) {
+        for (const ch of verdict.checks.filter((x) => !x.ok)) {
+          note(c.dim(`    ✗ ${ch.name}: ${ch.detail}`));
+        }
+      }
       if (opts.confirm && p.confirm) {
         const found = findConfirmButton(result);
         if (found && existsSync(STOP_FILE)) {
@@ -845,8 +944,12 @@ async function loop(opts) {
     clearTitle();
     await client.disconnect();
   }
-  note(c.green(`done — ${ran} probe(s) logged.`));
+  note(c.green(`done — ${ran} probe(s) logged.`) + (graded ? c.dim(`  ·  ${graded - failed}/${graded} assertions passed`) : ""));
   notify(`mira loop done — ${ran} probe(s) logged`, { quiet: opts.quiet });
+  if (failed > 0 && !opts.noFail) {
+    note(c.red(`${failed} probe(s) failed their assertions.`));
+    process.exit(1);
+  }
 }
 
 // src/commands/report.ts
@@ -890,16 +993,24 @@ function gist(r) {
   const first = r.messages.find((m) => m.text)?.text ?? (r.messages.length ? "(non-text)" : "(no reply)");
   return truncate(first, 90);
 }
+function verdictCell(r) {
+  if (!r.assert)
+    return "—";
+  if (r.assert.ok)
+    return "✅";
+  const failed = r.assert.checks.filter((ch) => !ch.ok).map((ch) => ch.name).join(", ");
+  return `❌ ${cell(failed)}`;
+}
 function table(rows) {
-  const head = `| Probe | Sent | Reply (gist) | Signals | First reply | Settled |
-` + "|---|---|---|---|---|---|";
+  const head = `| Probe | Test | Sent | Reply (gist) | Signals | First reply | Settled |
+` + "|---|---|---|---|---|---|---|";
   const body = rows.map((r) => {
     const id = r.probeId ?? "—";
     const sent = cell(truncate(r.sent, 40));
     const reply = r.timedOut ? "**TIMEOUT**" : cell(gist(r));
     const first = r.firstReplyMs === null ? "—" : `${(r.firstReplyMs / 1000).toFixed(1)}s`;
     const total = `${(r.totalMs / 1000).toFixed(1)}s`;
-    return `| \`${cell(id)}\` | ${sent} | ${reply} | ${cell(signals(r))} | ${first} | ${total} |`;
+    return `| \`${cell(id)}\` | ${verdictCell(r)} | ${sent} | ${reply} | ${cell(signals(r))} | ${first} | ${total} |`;
   }).join(`
 `);
   return `${head}
@@ -910,12 +1021,14 @@ function buildReport(records) {
   const latencies = replied.map((r) => r.firstReplyMs ?? 0).filter((n) => n > 0);
   const avg = latencies.length ? latencies.reduce((a, b) => a + b, 0) / latencies.length : 0;
   const max = latencies.length ? Math.max(...latencies) : 0;
+  const graded = records.filter((r) => r.assert);
+  const passed = graded.filter((r) => r.assert?.ok);
   const out = [];
   out.push("# Mira(@mira) probe report (auto-generated)");
   out.push("");
   out.push("> Distilled from the run log by `mira-harness report`.");
   out.push("");
-  out.push(`**${records.length}** probes · **${replied.length}** replied · **${records.length - replied.length}** timed out · ` + `first-reply avg **${(avg / 1000).toFixed(1)}s** / max **${(max / 1000).toFixed(1)}s**.`);
+  out.push(`**${records.length}** probes · **${replied.length}** replied · **${records.length - replied.length}** timed out · ` + (graded.length ? `**${passed.length}/${graded.length}** assertions passed · ` : "") + `first-reply avg **${(avg / 1000).toFixed(1)}s** / max **${(max / 1000).toFixed(1)}s**.`);
   out.push("");
   const builtin = CATEGORIES.filter((c2) => records.some((r) => r.category === c2));
   const custom = [];
@@ -1063,6 +1176,8 @@ function stats(opts = {}) {
   const replied = records.filter((r) => !r.timedOut);
   const timedOut = records.length - replied.length;
   const lat = latencyStats(records);
+  const graded = records.filter((r) => r.assert);
+  const passed = graded.filter((r) => r.assert?.ok);
   const byCat = new Map;
   for (const r of records) {
     const key = r.category ?? "(uncategorized)";
@@ -1078,6 +1193,7 @@ function stats(opts = {}) {
       probes: records.length,
       replied: replied.length,
       timedOut,
+      assertions: graded.length ? { graded: graded.length, passed: passed.length } : null,
       latencyMs: lat.count ? { min: lat.min, median: lat.median, p95: lat.p95, max: lat.max } : null,
       byCategory: Object.fromEntries([...byCat].map(([k, v]) => [k, v]))
     }, null, 2));
@@ -1086,6 +1202,10 @@ function stats(opts = {}) {
   const rate = (n, total) => `${total ? Math.round(n / total * 100) : 0}%`;
   note(c.bold(`\uD83D\uDCCA mira stats${opts.category ? ` [${opts.category}]` : ""}`));
   note(`  ${c.cyan(String(records.length))} probes · ${c.green(`${replied.length} replied`)} · ` + (timedOut ? c.yellow(`${timedOut} timed out`) : c.dim("0 timed out")) + c.dim(`  (${rate(replied.length, records.length)} reply rate)`));
+  if (graded.length) {
+    const allPass = passed.length === graded.length;
+    note(`  ${(allPass ? c.green : c.yellow)(`${passed.length}/${graded.length}`)} assertions passed` + c.dim(`  (${rate(passed.length, graded.length)})`));
+  }
   if (lat.count) {
     note("");
     note(c.bold("  first-reply latency"));
@@ -1193,7 +1313,7 @@ program.command("send").description("Send one probe to @mira and print the full
     noLog: opts.log === false
   });
 });
-program.command("loop").description("Run the experiment catalog against @mira at a human pace").option("-c, --category <category>", `only this category (${CATEGORIES.join(" | ")})`).option("-m, --max <n>", "max probes this run", "6").option("--confirm", "press a safe ✅ Confirm on generation probes (spends credits)", false).option("--peer <peer>", "'experiment'|'group' for TG_EXPERIMENT_CHAT, or a literal allowlisted peer").option("--gap <ms>", "delay between sends").option("--settle <ms>", "quiet window before concluding a reply").option("--timeout <ms>", "give up if no reply by this many ms").option("--list", "list the probes that would run, then exit (no sends)", false).option("--catalog <file>", "custom catalog JSON file (instead of the built-in catalog)").option("-q, --quiet", "suppress progress spinners", false).action(async (opts) => {
+program.command("loop").description("Run the experiment catalog against @mira at a human pace").option("-c, --category <category>", `only this category (${CATEGORIES.join(" | ")})`).option("-m, --max <n>", "max probes this run", "6").option("--confirm", "press a safe ✅ Confirm on generation probes (spends credits)", false).option("--peer <peer>", "'experiment'|'group' for TG_EXPERIMENT_CHAT, or a literal allowlisted peer").option("--gap <ms>", "delay between sends").option("--settle <ms>", "quiet window before concluding a reply").option("--timeout <ms>", "give up if no reply by this many ms").option("--list", "list the probes that would run, then exit (no sends)", false).option("--catalog <file>", "custom catalog JSON file (instead of the built-in catalog)").option("--no-fail", "report failed assertions but still exit 0 (default: exit 1 on failure)").option("-q, --quiet", "suppress progress spinners", false).action(async (opts) => {
   await loop({
     category: opts.category,
     max: posInt("max", opts.max),
@@ -1204,6 +1324,7 @@ program.command("loop").description("Run the experiment catalog against @mira at
     timeout: ms("timeout", opts.timeout),
     list: opts.list,
     catalog: opts.catalog,
+    noFail: opts.fail === false,
     quiet: opts.quiet
   });
 });

package/dist/commands/report.d.ts

@@ -1,8 +1,10 @@
+import type { Verdict } from "../assert.js";
 import type { ProbeResult } from "../capture.js";
 export type RunRecord = ProbeResult & {
     probeId?: string;
     category?: string;
     hypothesis?: string;
+    assert?: Verdict;
 };
 /**
  * Read the JSONL run log into records (malformed lines skipped). Throws ENOENT when

package/dist/index.js

@@ -92,16 +92,87 @@ function extractMessage(msg, editCount = 0) {
 // src/catalog.ts
 import { readFileSync } from "node:fs";
 import { resolve } from "node:path";
+import { z as z2 } from "zod";
+
+// src/assert.ts
 import { z } from "zod";
+var ExpectSchema = z.object({
+  replies: z.boolean().optional(),
+  textMatches: z.string().optional(),
+  minButtons: z.number().int().nonnegative().optional(),
+  minLinks: z.number().int().nonnegative().optional(),
+  hasWebApp: z.boolean().optional(),
+  media: z.enum(["photo", "video", "audio", "document", "webpage", "other"]).optional(),
+  maxFirstReplyMs: z.number().positive().optional(),
+  json: z.boolean().optional()
+});
+function evaluate(expect, result) {
+  const checks = [];
+  const texts = result.messages.map((m) => m.text).filter((t) => t.length > 0);
+  const firstText = texts[0] ?? "";
+  const buttons = result.messages.reduce((n, m) => n + m.buttons.length, 0);
+  const links = result.messages.reduce((n, m) => n + m.links.length, 0);
+  const webApp = result.messages.some((m) => m.buttons.some((b) => b.webAppUrl !== undefined));
+  const mediaKinds = result.messages.map((m) => m.media?.kind).filter((k) => k !== undefined);
+  const add = (name, ok, detail) => {
+    checks.push({ name, ok, detail });
+  };
+  if (expect.replies !== undefined) {
+    const replied = !result.timedOut && result.messages.length > 0;
+    add("replies", replied === expect.replies, result.timedOut ? "timed out" : `${result.messages.length} message(s)`);
+  }
+  if (expect.textMatches !== undefined) {
+    const pattern = expect.textMatches;
+    let compiled;
+    try {
+      compiled = new RegExp(pattern, "i");
+    } catch {
+      compiled = undefined;
+    }
+    const rx = compiled;
+    const ok = rx ? texts.some((t) => rx.test(t)) : false;
+    add("textMatches", ok, rx ? `/${pattern}/i` : `invalid regex: ${pattern}`);
+  }
+  if (expect.minButtons !== undefined) {
+    add("minButtons", buttons >= expect.minButtons, `${buttons} >= ${expect.minButtons}`);
+  }
+  if (expect.minLinks !== undefined) {
+    add("minLinks", links >= expect.minLinks, `${links} >= ${expect.minLinks}`);
+  }
+  if (expect.hasWebApp !== undefined) {
+    add("hasWebApp", webApp === expect.hasWebApp, `webApp=${webApp}`);
+  }
+  if (expect.media !== undefined) {
+    add("media", mediaKinds.includes(expect.media), `kinds=[${mediaKinds.join(",")}] want ${expect.media}`);
+  }
+  if (expect.maxFirstReplyMs !== undefined) {
+    const ms = result.firstReplyMs;
+    add("maxFirstReplyMs", ms !== null && ms <= expect.maxFirstReplyMs, ms === null ? "no reply" : `${ms}ms <= ${expect.maxFirstReplyMs}ms`);
+  }
+  if (expect.json === true) {
+    let ok = false;
+    try {
+      JSON.parse(firstText);
+      ok = firstText.length > 0;
+    } catch {
+      ok = false;
+    }
+    add("json", ok, ok ? "valid JSON" : "not valid JSON");
+  }
+  return { ok: checks.every((ch) => ch.ok), checks };
+}
+
+// src/catalog.ts
 var CATEGORIES = ["core", "skills", "generation", "wallet"];
-var ProbeSchema = z.object({
-  id: z.string().min(1),
-  category: z.string().default("custom"),
-  hypothesis: z.string().default(""),
-  send: z.string().min(1),
-  slow: z.boolean().optional(),
-  confirm: z.boolean().optional(),
-  note: z.string().optional()
+var ProbeSchema = z2.object({
+  id: z2.string().min(1),
+  category: z2.string().default("custom"),
+  hypothesis: z2.string().default(""),
+  send: z2.string().min(1),
+  slow: z2.boolean().optional(),
+  confirm: z2.boolean().optional(),
+  note: z2.string().optional(),
+  expect: ExpectSchema.optional()
 });
 function loadCatalog(file) {
   const path = resolve(process.cwd(), file);
@@ -112,7 +183,7 @@ function loadCatalog(file) {
   } catch {
     throw new Error(`catalog ${file} is not valid JSON`);
   }
-  const parsed = z.array(ProbeSchema).safeParse(data);
+  const parsed = z2.array(ProbeSchema).safeParse(data);
   if (!parsed.success) {
     throw new Error(`invalid catalog ${file}: ${parsed.error.issues[0]?.message ?? "schema error"}`);
   }
@@ -125,7 +196,8 @@ var CATALOG = [
     id: "core-model",
     category: "core",
     hypothesis: "Which underlying model is Mira on? (docs say MiniMax M2.5)",
-    send: "What AI model are you running on right now? Answer with just the model name."
+    send: "What AI model are you running on right now? Answer with just the model name.",
+    expect: { replies: true }
   },
   {
     id: "core-mem-set",
@@ -137,13 +209,15 @@ var CATALOG = [
     id: "core-mem-get",
     category: "core",
     hypothesis: "Does the fact from core-mem-set persist across separate sends?",
-    send: "What did I tell you my favorite TON token is? Answer with one word."
+    send: "What did I tell you my favorite TON token is? Answer with one word.",
+    expect: { replies: true }
   },
   {
     id: "core-json",
     category: "core",
     hypothesis: "Will Mira return STRICT JSON only (usable for structured workflows)?",
-    send: 'Reply with ONLY this JSON and nothing else: {"ok":true,"n":42}'
+    send: 'Reply with ONLY this JSON and nothing else: {"ok":true,"n":42}',
+    expect: { json: true }
   },
   {
     id: "core-reason",
@@ -155,7 +229,8 @@ var CATALOG = [
     id: "core-commands",
     category: "core",
     hypothesis: "What commands does Mira expose? (/help surface, buttons)",
-    send: "/help"
+    send: "/help",
+    expect: { replies: true }
   },
   {
     id: "core-ja",
@@ -167,7 +242,8 @@ var CATALOG = [
     id: "skill-capabilities",
     category: "skills",
     hypothesis: "Does Mira enumerate its capabilities and surface buttons/links?",
-    send: "What can you do? List your main capabilities."
+    send: "What can you do? List your main capabilities.",
+    expect: { replies: true }
   },
   {
     id: "skill-research",
@@ -175,7 +251,8 @@ var CATALOG = [
     hypothesis: "Web research returns source links (text_url) the harness can capture.",
     send: "Search the web for the latest TON ecosystem news and give me 3 source links.",
     slow: true,
-    note: "expect links[] (text_url source links) from deep research"
+    note: "expect links[] (text_url source links) from deep research",
+    expect: { minLinks: 1 }
   },
   {
     id: "skill-translate",
@@ -202,7 +279,8 @@ var CATALOG = [
     send: "Generate an image: a friendly teal robot mascot holding a glowing TON diamond, flat vector style.",
     slow: true,
     confirm: true,
-    note: "expect a Confirm card; with --confirm the runner presses ✅ Confirm -> media.kind=photo"
+    note: "expect a Confirm card; with --confirm the runner presses ✅ Confirm -> media.kind=photo",
+    expect: { replies: true, minButtons: 1 }
   },
   {
     id: "gen-music",
@@ -538,16 +616,24 @@ function gist(r) {
   const first = r.messages.find((m) => m.text)?.text ?? (r.messages.length ? "(non-text)" : "(no reply)");
   return truncate(first, 90);
 }
+function verdictCell(r) {
+  if (!r.assert)
+    return "—";
+  if (r.assert.ok)
+    return "✅";
+  const failed = r.assert.checks.filter((ch) => !ch.ok).map((ch) => ch.name).join(", ");
+  return `❌ ${cell(failed)}`;
+}
 function table(rows) {
-  const head = `| Probe | Sent | Reply (gist) | Signals | First reply | Settled |
-` + "|---|---|---|---|---|---|";
+  const head = `| Probe | Test | Sent | Reply (gist) | Signals | First reply | Settled |
+` + "|---|---|---|---|---|---|---|";
   const body = rows.map((r) => {
     const id = r.probeId ?? "—";
     const sent = cell(truncate(r.sent, 40));
     const reply = r.timedOut ? "**TIMEOUT**" : cell(gist(r));
     const first = r.firstReplyMs === null ? "—" : `${(r.firstReplyMs / 1000).toFixed(1)}s`;
     const total = `${(r.totalMs / 1000).toFixed(1)}s`;
-    return `| \`${cell(id)}\` | ${sent} | ${reply} | ${cell(signals(r))} | ${first} | ${total} |`;
+    return `| \`${cell(id)}\` | ${verdictCell(r)} | ${sent} | ${reply} | ${cell(signals(r))} | ${first} | ${total} |`;
   }).join(`
 `);
   return `${head}
@@ -558,12 +644,14 @@ function buildReport(records) {
   const latencies = replied.map((r) => r.firstReplyMs ?? 0).filter((n) => n > 0);
   const avg = latencies.length ? latencies.reduce((a, b) => a + b, 0) / latencies.length : 0;
   const max = latencies.length ? Math.max(...latencies) : 0;
+  const graded = records.filter((r) => r.assert);
+  const passed = graded.filter((r) => r.assert?.ok);
   const out = [];
   out.push("# Mira(@mira) probe report (auto-generated)");
   out.push("");
   out.push("> Distilled from the run log by `mira-harness report`.");
   out.push("");
-  out.push(`**${records.length}** probes · **${replied.length}** replied · **${records.length - replied.length}** timed out · ` + `first-reply avg **${(avg / 1000).toFixed(1)}s** / max **${(max / 1000).toFixed(1)}s**.`);
+  out.push(`**${records.length}** probes · **${replied.length}** replied · **${records.length - replied.length}** timed out · ` + (graded.length ? `**${passed.length}/${graded.length}** assertions passed · ` : "") + `first-reply avg **${(avg / 1000).toFixed(1)}s** / max **${(max / 1000).toFixed(1)}s**.`);
   out.push("");
   const builtin = CATEGORIES.filter((c) => records.some((r) => r.category === c));
   const custom = [];

package/dist/log.d.ts

@@ -1,3 +1,4 @@
+import type { Verdict } from "./assert.js";
 import type { ProbeResult } from "./capture.js";
 export declare const RUNS_FILE: string;
 /** Optional provenance attached by the loop runner (manual sends omit it). */
@@ -5,5 +6,7 @@ export interface RunMeta {
     probeId?: string;
     category?: string;
     hypothesis?: string;
+    /** Assertion verdict, present only when the probe declared `expect`. */
+    assert?: Verdict;
 }
 export declare function appendRun(result: ProbeResult, meta?: RunMeta): Promise<void>;

package/dist/mcp.js

@@ -4,21 +4,92 @@
 import { existsSync } from "node:fs";
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
-import { z as z2 } from "zod";
+import { z as z3 } from "zod";
 
 // src/catalog.ts
 import { readFileSync } from "node:fs";
 import { resolve } from "node:path";
+import { z as z2 } from "zod";
+
+// src/assert.ts
 import { z } from "zod";
+var ExpectSchema = z.object({
+  replies: z.boolean().optional(),
+  textMatches: z.string().optional(),
+  minButtons: z.number().int().nonnegative().optional(),
+  minLinks: z.number().int().nonnegative().optional(),
+  hasWebApp: z.boolean().optional(),
+  media: z.enum(["photo", "video", "audio", "document", "webpage", "other"]).optional(),
+  maxFirstReplyMs: z.number().positive().optional(),
+  json: z.boolean().optional()
+});
+function evaluate(expect, result) {
+  const checks = [];
+  const texts = result.messages.map((m) => m.text).filter((t) => t.length > 0);
+  const firstText = texts[0] ?? "";
+  const buttons = result.messages.reduce((n, m) => n + m.buttons.length, 0);
+  const links = result.messages.reduce((n, m) => n + m.links.length, 0);
+  const webApp = result.messages.some((m) => m.buttons.some((b) => b.webAppUrl !== undefined));
+  const mediaKinds = result.messages.map((m) => m.media?.kind).filter((k) => k !== undefined);
+  const add = (name, ok, detail) => {
+    checks.push({ name, ok, detail });
+  };
+  if (expect.replies !== undefined) {
+    const replied = !result.timedOut && result.messages.length > 0;
+    add("replies", replied === expect.replies, result.timedOut ? "timed out" : `${result.messages.length} message(s)`);
+  }
+  if (expect.textMatches !== undefined) {
+    const pattern = expect.textMatches;
+    let compiled;
+    try {
+      compiled = new RegExp(pattern, "i");
+    } catch {
+      compiled = undefined;
+    }
+    const rx = compiled;
+    const ok = rx ? texts.some((t) => rx.test(t)) : false;
+    add("textMatches", ok, rx ? `/${pattern}/i` : `invalid regex: ${pattern}`);
+  }
+  if (expect.minButtons !== undefined) {
+    add("minButtons", buttons >= expect.minButtons, `${buttons} >= ${expect.minButtons}`);
+  }
+  if (expect.minLinks !== undefined) {
+    add("minLinks", links >= expect.minLinks, `${links} >= ${expect.minLinks}`);
+  }
+  if (expect.hasWebApp !== undefined) {
+    add("hasWebApp", webApp === expect.hasWebApp, `webApp=${webApp}`);
+  }
+  if (expect.media !== undefined) {
+    add("media", mediaKinds.includes(expect.media), `kinds=[${mediaKinds.join(",")}] want ${expect.media}`);
+  }
+  if (expect.maxFirstReplyMs !== undefined) {
+    const ms = result.firstReplyMs;
+    add("maxFirstReplyMs", ms !== null && ms <= expect.maxFirstReplyMs, ms === null ? "no reply" : `${ms}ms <= ${expect.maxFirstReplyMs}ms`);
+  }
+  if (expect.json === true) {
+    let ok = false;
+    try {
+      JSON.parse(firstText);
+      ok = firstText.length > 0;
+    } catch {
+      ok = false;
+    }
+    add("json", ok, ok ? "valid JSON" : "not valid JSON");
+  }
+  return { ok: checks.every((ch) => ch.ok), checks };
+}
+
+// src/catalog.ts
 var CATEGORIES = ["core", "skills", "generation", "wallet"];
-var ProbeSchema = z.object({
-  id: z.string().min(1),
-  category: z.string().default("custom"),
-  hypothesis: z.string().default(""),
-  send: z.string().min(1),
-  slow: z.boolean().optional(),
-  confirm: z.boolean().optional(),
-  note: z.string().optional()
+var ProbeSchema = z2.object({
+  id: z2.string().min(1),
+  category: z2.string().default("custom"),
+  hypothesis: z2.string().default(""),
+  send: z2.string().min(1),
+  slow: z2.boolean().optional(),
+  confirm: z2.boolean().optional(),
+  note: z2.string().optional(),
+  expect: ExpectSchema.optional()
 });
 function loadCatalog(file) {
   const path = resolve(process.cwd(), file);
@@ -29,7 +100,7 @@ function loadCatalog(file) {
   } catch {
     throw new Error(`catalog ${file} is not valid JSON`);
   }
-  const parsed = z.array(ProbeSchema).safeParse(data);
+  const parsed = z2.array(ProbeSchema).safeParse(data);
   if (!parsed.success) {
     throw new Error(`invalid catalog ${file}: ${parsed.error.issues[0]?.message ?? "schema error"}`);
   }
@@ -42,7 +113,8 @@ var CATALOG = [
     id: "core-model",
     category: "core",
     hypothesis: "Which underlying model is Mira on? (docs say MiniMax M2.5)",
-    send: "What AI model are you running on right now? Answer with just the model name."
+    send: "What AI model are you running on right now? Answer with just the model name.",
+    expect: { replies: true }
   },
   {
     id: "core-mem-set",
@@ -54,13 +126,15 @@ var CATALOG = [
     id: "core-mem-get",
     category: "core",
     hypothesis: "Does the fact from core-mem-set persist across separate sends?",
-    send: "What did I tell you my favorite TON token is? Answer with one word."
+    send: "What did I tell you my favorite TON token is? Answer with one word.",
+    expect: { replies: true }
   },
   {
     id: "core-json",
     category: "core",
     hypothesis: "Will Mira return STRICT JSON only (usable for structured workflows)?",
-    send: 'Reply with ONLY this JSON and nothing else: {"ok":true,"n":42}'
+    send: 'Reply with ONLY this JSON and nothing else: {"ok":true,"n":42}',
+    expect: { json: true }
   },
   {
     id: "core-reason",
@@ -72,7 +146,8 @@ var CATALOG = [
     id: "core-commands",
     category: "core",
     hypothesis: "What commands does Mira expose? (/help surface, buttons)",
-    send: "/help"
+    send: "/help",
+    expect: { replies: true }
   },
   {
     id: "core-ja",
@@ -84,7 +159,8 @@ var CATALOG = [
     id: "skill-capabilities",
     category: "skills",
     hypothesis: "Does Mira enumerate its capabilities and surface buttons/links?",
-    send: "What can you do? List your main capabilities."
+    send: "What can you do? List your main capabilities.",
+    expect: { replies: true }
   },
   {
     id: "skill-research",
@@ -92,7 +168,8 @@ var CATALOG = [
     hypothesis: "Web research returns source links (text_url) the harness can capture.",
     send: "Search the web for the latest TON ecosystem news and give me 3 source links.",
     slow: true,
-    note: "expect links[] (text_url source links) from deep research"
+    note: "expect links[] (text_url source links) from deep research",
+    expect: { minLinks: 1 }
   },
   {
     id: "skill-translate",
@@ -119,7 +196,8 @@ var CATALOG = [
     send: "Generate an image: a friendly teal robot mascot holding a glowing TON diamond, flat vector style.",
     slow: true,
     confirm: true,
-    note: "expect a Confirm card; with --confirm the runner presses ✅ Confirm -> media.kind=photo"
+    note: "expect a Confirm card; with --confirm the runner presses ✅ Confirm -> media.kind=photo",
+    expect: { replies: true, minButtons: 1 }
   },
   {
     id: "gen-music",
@@ -549,16 +627,24 @@ function gist(r) {
   const first = r.messages.find((m) => m.text)?.text ?? (r.messages.length ? "(non-text)" : "(no reply)");
   return truncate(first, 90);
 }
+function verdictCell(r) {
+  if (!r.assert)
+    return "—";
+  if (r.assert.ok)
+    return "✅";
+  const failed = r.assert.checks.filter((ch) => !ch.ok).map((ch) => ch.name).join(", ");
+  return `❌ ${cell(failed)}`;
+}
 function table(rows) {
-  const head = `| Probe | Sent | Reply (gist) | Signals | First reply | Settled |
-` + "|---|---|---|---|---|---|";
+  const head = `| Probe | Test | Sent | Reply (gist) | Signals | First reply | Settled |
+` + "|---|---|---|---|---|---|---|";
   const body = rows.map((r) => {
     const id = r.probeId ?? "—";
     const sent = cell(truncate(r.sent, 40));
     const reply = r.timedOut ? "**TIMEOUT**" : cell(gist(r));
     const first = r.firstReplyMs === null ? "—" : `${(r.firstReplyMs / 1000).toFixed(1)}s`;
     const total = `${(r.totalMs / 1000).toFixed(1)}s`;
-    return `| \`${cell(id)}\` | ${sent} | ${reply} | ${cell(signals(r))} | ${first} | ${total} |`;
+    return `| \`${cell(id)}\` | ${verdictCell(r)} | ${sent} | ${reply} | ${cell(signals(r))} | ${first} | ${total} |`;
   }).join(`
 `);
   return `${head}
@@ -569,12 +655,14 @@ function buildReport(records) {
   const latencies = replied.map((r) => r.firstReplyMs ?? 0).filter((n) => n > 0);
   const avg = latencies.length ? latencies.reduce((a, b) => a + b, 0) / latencies.length : 0;
   const max = latencies.length ? Math.max(...latencies) : 0;
+  const graded = records.filter((r) => r.assert);
+  const passed = graded.filter((r) => r.assert?.ok);
   const out = [];
   out.push("# Mira(@mira) probe report (auto-generated)");
   out.push("");
   out.push("> Distilled from the run log by `mira-harness report`.");
   out.push("");
-  out.push(`**${records.length}** probes · **${replied.length}** replied · **${records.length - replied.length}** timed out · ` + `first-reply avg **${(avg / 1000).toFixed(1)}s** / max **${(max / 1000).toFixed(1)}s**.`);
+  out.push(`**${records.length}** probes · **${replied.length}** replied · **${records.length - replied.length}** timed out · ` + (graded.length ? `**${passed.length}/${graded.length}** assertions passed · ` : "") + `first-reply avg **${(avg / 1000).toFixed(1)}s** / max **${(max / 1000).toFixed(1)}s**.`);
   out.push("");
   const builtin = CATEGORIES.filter((c) => records.some((r) => r.category === c));
   const custom = [];
@@ -672,9 +760,9 @@ server.registerTool("mira_send", {
   title: "Send one probe to @mira",
   description: "Send a message to @mira and capture the FULL settled reply (messages, buttons incl. web_app/startapp, links, media, edits, latency).",
   inputSchema: {
-    message: z2.string().describe("the message to send to @mira"),
-    settleMs: z2.number().int().positive().optional().describe("quiet window before concluding"),
-    timeoutMs: z2.number().int().positive().optional().describe("give up if no reply by then")
+    message: z3.string().describe("the message to send to @mira"),
+    settleMs: z3.number().int().positive().optional().describe("quiet window before concluding"),
+    timeoutMs: z3.number().int().positive().optional().describe("give up if no reply by then")
   }
 }, async ({ message, settleMs, timeoutMs }) => {
   if (existsSync(STOP_FILE))
@@ -696,13 +784,13 @@ server.registerTool("mira_loop", {
   title: "Run the experiment catalog (observe-only)",
   description: "Send catalog probes to @mira at a human pace and capture replies. OBSERVE-ONLY: never presses buttons or spends credits (use the CLI `loop --confirm` for that).",
   inputSchema: {
-    category: z2.string().optional().describe(`one of: ${CATEGORIES.join(", ")} (or any category in a custom catalog)`),
-    max: z2.number().int().positive().max(50).optional().describe("max probes (default 6)"),
-    peer: z2.string().optional().describe("'experiment'|'group' for TG_EXPERIMENT_CHAT, or a literal allowlisted peer"),
-    gapMs: z2.number().int().nonnegative().optional().describe("delay between sends (default 15000)"),
-    settleMs: z2.number().int().positive().optional(),
-    timeoutMs: z2.number().int().positive().optional(),
-    catalogFile: z2.string().optional().describe("custom catalog JSON file (instead of the built-in catalog)")
+    category: z3.string().optional().describe(`one of: ${CATEGORIES.join(", ")} (or any category in a custom catalog)`),
+    max: z3.number().int().positive().max(50).optional().describe("max probes (default 6)"),
+    peer: z3.string().optional().describe("'experiment'|'group' for TG_EXPERIMENT_CHAT, or a literal allowlisted peer"),
+    gapMs: z3.number().int().nonnegative().optional().describe("delay between sends (default 15000)"),
+    settleMs: z3.number().int().positive().optional(),
+    timeoutMs: z3.number().int().positive().optional(),
+    catalogFile: z3.string().optional().describe("custom catalog JSON file (instead of the built-in catalog)")
   }
 }, async ({ category, max, peer, gapMs, settleMs, timeoutMs, catalogFile }) => {
   let source;
@@ -757,8 +845,8 @@ server.registerTool("mira_catalog", {
   title: "List the experiment catalog",
   description: "List probes (id, category, hypothesis, flags) without sending anything.",
   inputSchema: {
-    category: z2.string().optional().describe(`one of: ${CATEGORIES.join(", ")} (or any category in a custom catalog)`),
-    catalogFile: z2.string().optional().describe("custom catalog JSON file (instead of the built-in catalog)")
+    category: z3.string().optional().describe(`one of: ${CATEGORIES.join(", ")} (or any category in a custom catalog)`),
+    catalogFile: z3.string().optional().describe("custom catalog JSON file (instead of the built-in catalog)")
   }
 }, async ({ category, catalogFile }) => {
   let source;
@@ -784,8 +872,8 @@ server.registerTool("mira_report", {
   title: "Render the run log as Markdown",
   description: "Distill the JSONL run log into a Markdown report (per-probe gist / signals / latency).",
   inputSchema: {
-    inFile: z2.string().optional().describe("input JSONL path (default: the run log)"),
-    category: z2.string().optional().describe("only include probes from this category")
+    inFile: z3.string().optional().describe("input JSONL path (default: the run log)"),
+    category: z3.string().optional().describe("only include probes from this category")
   }
 }, async ({ inFile, category }) => {
   try {

package/llms.txt

@@ -19,7 +19,7 @@ so the only programmatic path is a userbot (a real user account over MTProto / G
 ## CLI commands
 - `doctor` — env / session / connectivity / @mira resolution check (read-only). Run this first.
 - `send [message...]` — one probe → full reply as JSON. Message via arg or stdin. Flags: `--settle <ms> --timeout <ms> --quiet --no-log`.
-- `loop` — run the catalog, paced. Flags: `--category <core|skills|generation|wallet> --max <n> --peer <experiment> --gap <ms> --settle <ms> --timeout <ms> --list --catalog <file.json> --quiet`. Observe-only unless `--confirm`.
+- `loop` — run the catalog, paced. Flags: `--category <core|skills|generation|wallet> --max <n> --peer <experiment> --gap <ms> --settle <ms> --timeout <ms> --list --catalog <file.json> --no-fail --quiet`. Observe-only unless `--confirm`. Grades probes that declare `expect` (PASS/FAIL) and **exits non-zero** if any graded probe fails (`--no-fail` to report without failing) — drops straight into CI.
 - `catalog` — list the catalog, no network. Flags: `--category --catalog <file.json> --json`.
 - `watch` — live-tail @mira's messages, observe-only. Flag: `--peer`.
 - `report` — distill the JSONL run log into Markdown. Flags: `--in <file> --out <file> --category`.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mira-harness",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "description": "A CLI + MCP dev-tool for communicating with the @mira Telegram bot — capture its full reply (buttons/links/media) and run a self-driving experiment catalog.",
   "type": "module",
   "bin": {