switchroom 0.19.41 → 0.19.43

package/telegram-plugin/tests/reply-quote-wire.test.ts

@@ -0,0 +1,433 @@
+/**
+ * Wire-level outcome test for `quote_text` (surgical quote-reply).
+ *
+ * WHY AT THE FETCH LAYER
+ * ----------------------
+ * The bug this pins was a PAYLOAD-SHAPE bug: `reply_parameters.quote` was built
+ * as `{ text, position: 0 }` when the Bot API declares `quote` a **String**
+ * (`quote_position` is a separate sibling Integer). Telegram answered every
+ * quoted reply with `400 Bad Request: field "quote" must be of type String` —
+ * 378 of 384 total `sendRichMessage` failures across the live fleet
+ * (agent `overlord`, 2026-07-29 → 07-30). A unit test over the options builder
+ * would have asserted the same wrong shape happily; only a test that reads what
+ * actually goes over the wire can catch it, and only that keeps a future
+ * wrapper (a re-render, an opts transform, a new send adapter) from re-breaking
+ * it invisibly.
+ *
+ * So this drives the REAL send path — `sendReply` (the body of the gateway's
+ * `executeReply`, which is a thin wrapper over it) — against a REAL grammy
+ * `Bot` whose HTTP client is an injected `fetch`. Every assertion is made on
+ * the JSON body grammy serialized for `POST /botTOKEN/sendRichMessage`.
+ */
+
+import { describe, it, expect } from 'vitest'
+import { Bot, type Context } from 'grammy'
+import { sendReply, type SendReplyGatewayDeps } from '../gateway/outbound-send-path.js'
+import { handleStreamReply, type StreamReplyDeps } from '../stream-reply-handler.js'
+import type { StreamBotApi } from '../stream-controller.js'
+import type { DraftStreamHandle } from '../draft-stream.js'
+import { OutboundDedupCache } from '../recent-outbound-dedup.js'
+import { FlushedTurnSupersedeRegistry } from '../flushed-turn-supersede.js'
+import { VoiceOnDemandCache } from '../voice-ondemand.js'
+import { PreSynthQueue } from '../voice-presynth.js'
+import type { Access } from '../gateway/gateway.js'
+
+const CHAT = '5550001'
+
+type WireCall = { method: string; payload: Record<string, unknown> }
+
+/** Scripted Telegram error for one wire method. Fires `times` times (default 1),
+ *  then that method succeeds. `times` matters for the body-error cases: a real
+ *  unparseable body fails EVERY time it is re-sent to the same method, so a
+ *  fire-once script would let a wrong fallback branch pass by accident. */
+type ScriptedError = { method: string; description: string; times?: number }
+
+/** A REAL grammy Bot whose HTTP client is an injected fetch — every call is
+ *  captured as the JSON body grammy actually serialized for the Bot API. */
+function makeWireBot(opts: { errors?: ScriptedError[] } = {}) {
+  const calls: WireCall[] = []
+  const errors = [...(opts.errors ?? [])]
+  let nextMessageId = 9000
+
+  const fetchImpl = (async (input: unknown, init?: { body?: unknown }) => {
+    const url = String(input)
+    const method = url.slice(url.lastIndexOf('/') + 1)
+    const body = typeof init?.body === 'string' ? init.body : '{}'
+    const payload = JSON.parse(body) as Record<string, unknown>
+    calls.push({ method, payload })
+    const scripted = errors.findIndex((e) => e.method === method)
+    if (scripted >= 0) {
+      const err = errors[scripted]
+      const remaining = (err.times ?? 1) - 1
+      if (remaining > 0) errors[scripted] = { ...err, times: remaining }
+      else errors.splice(scripted, 1)
+      return new Response(
+        JSON.stringify({ ok: false, error_code: 400, description: err.description }),
+        { status: 400, headers: { 'content-type': 'application/json' } },
+      )
+    }
+    return new Response(
+      JSON.stringify({
+        ok: true,
+        result: {
+          message_id: nextMessageId++,
+          date: Math.floor(Date.now() / 1000),
+          chat: { id: Number(CHAT), type: 'private' },
+        },
+      }),
+      { status: 200, headers: { 'content-type': 'application/json' } },
+    )
+  }) as unknown as typeof fetch
+
+  const bot = new Bot<Context>('123456:TEST-TOKEN', {
+    client: { fetch: fetchImpl },
+    botInfo: {
+      id: 123456,
+      is_bot: true,
+      first_name: 'test',
+      username: 'test_bot',
+      can_join_groups: false,
+      can_read_all_group_messages: false,
+      supports_inline_queries: false,
+      can_connect_to_business_account: false,
+      has_main_web_app: false,
+    },
+  })
+
+  const sends = () => calls.filter((c) => c.method.startsWith('send'))
+  const replyParams = (c: WireCall) => c.payload.reply_parameters as Record<string, unknown> | undefined
+  return { bot, calls, sends, replyParams }
+}
+
+function makeHarness(opts: { errors?: ScriptedError[] } = {}) {
+  const wire = makeWireBot(opts)
+  const bot = wire.bot
+
+  const access: Access = {
+    dmPolicy: 'allowlist',
+    allowFrom: [CHAT],
+    groups: {},
+    pending: {},
+    parseMode: 'html',
+    historyEnabled: false,
+  }
+
+  const noop = () => {}
+  const deps: SendReplyGatewayDeps = {
+    outboundDedup: new OutboundDedupCache(),
+    flushedTurnSupersede: new FlushedTurnSupersedeRegistry(),
+    firstTextReplyLogged: new Set<string>(),
+    suppressPtyPreview: new Set<string>(),
+    activeDraftStreams: new Map(),
+    lastPtyPreviewByChat: new Map(),
+    voiceOnDemandCache: new VoiceOnDemandCache(),
+    voicePreSynthQueue: new PreSynthQueue({ runJob: async () => {} }),
+    pendingProgress: { clearPending: noop, noteOutbound: noop },
+    signalTracker: { noteOutbound: noop, noteSignal: noop },
+    silencePoke: { noteOutbound: noop },
+    getCurrentTurn: () => null,
+    getLastActiveTurnChatId: () => undefined,
+    HISTORY_ENABLED: false,
+    TURN_ORIGIN_ROUTING_ENABLED: false,
+    AUTOCLASSIFY_MIDTURN_SHADOW: false,
+    MAX_ATTACHMENT_BYTES: 50 * 1024 * 1024,
+    MAX_CHUNK_LIMIT: 4000,
+    PHOTO_EXTS: new Set(['.png', '.jpg']),
+    lockedBot: bot,
+    robustApiCall: <T>(fn: () => Promise<T>) => fn(),
+    swallowingApiCall: async (fn: () => Promise<unknown>) => {
+      try {
+        return await fn()
+      } catch {
+        return undefined
+      }
+    },
+    loadAccess: () => access,
+    redactOutboundText: (t: string) => t,
+    assertAllowedChat: noop,
+    assertSendable: noop,
+    statusKey: (chatId: string, threadId?: number | null) => `${chatId}:${threadId ?? ''}`,
+    streamKey: (chatId: string, threadId?: number | null) => `${chatId}:${threadId ?? ''}`,
+    resolveReplyOwnerTurn: () => ({
+      turn: null,
+      tier: 'none',
+      candidates: {
+        liveTurnId: null,
+        originTurnId: null,
+        quotedTurnId: null,
+        latestEndedTurnId: null,
+      },
+    }),
+    findTurnByOriginId: () => null,
+    findTurnByQuotedMessageId: () => null,
+    resolveAnswerThreadWithLog: () => undefined,
+    resolveThreadId: () => undefined,
+    getLatestInboundMessageId: () => null,
+    getLastSubagentHandbackAt: () => null,
+    recordOutbound: noop,
+    emissionAuthorityFor: () =>
+      ({
+        claimOrDowngradePing: () => ({ allowed: true, ping: true }),
+      }) as never,
+    clearActivitySummary: noop,
+    startTypingLoop: noop,
+    stopTypingLoop: noop,
+    logOutbound: noop,
+    closeObligationOnSubstantiveReply: noop,
+    finalizeStatusReaction: noop,
+    releaseTurnBufferGate: noop,
+    reapQueuedStatus: noop,
+    noteAgentOutputAt: noop,
+    rememberAgentButtonMeta: noop,
+    resolveVoiceOutPlan: () => null,
+    synthesizeVoiceOut: async () => null,
+    publishToTelegraph: async () => null,
+    clearSilentEndState: noop,
+    emitRuntimeMetric: noop,
+    shadowEmit: noop,
+    progressDriver: null,
+  }
+
+  return { ...wire, deps }
+}
+
+/** Drive the REAL `stream_reply` handler against the same wire-capturing bot. */
+function makeStreamHarness(opts: { errors?: ScriptedError[] } = {}) {
+  const wire = makeWireBot(opts)
+  const noop = () => {}
+  const deps: StreamReplyDeps = {
+    bot: wire.bot as unknown as { api: StreamBotApi },
+    retry: <T>(fn: () => Promise<T>) => fn(),
+    repairEscapedWhitespace: (t: string) => t,
+    assertAllowedChat: noop,
+    resolveThreadId: () => undefined,
+    disableLinkPreview: true,
+    defaultFormat: 'html',
+    logStreamingEvent: noop,
+    historyEnabled: false,
+    recordOutbound: noop,
+    writeError: noop,
+    throttleMs: 0,
+  }
+  const state = {
+    activeDraftStreams: new Map<string, DraftStreamHandle>(),
+    suppressPtyPreview: new Set<string>(),
+    lastPtyPreviewByChat: new Map<string, string>(),
+  }
+  return { ...wire, deps, state }
+}
+
+describe('quote_text lands on the wire as ReplyParameters.quote (a String)', () => {
+  it('sends reply_parameters.quote as a string and resolves delivered', async () => {
+    const h = makeHarness()
+    const res = await sendReply(h.deps, {
+      args: {
+        chat_id: CHAT,
+        text: 'Yes — that line is the one that breaks.',
+        reply_to: 4242,
+        quote_text: 'the position is a separate field',
+      },
+      turn: null,
+    })
+
+    const sends = h.sends()
+    expect(sends.length).toBeGreaterThan(0)
+    const rp = h.replyParams(sends[0])
+    expect(rp).toBeDefined()
+    expect(rp!.message_id).toBe(4242)
+    // The bug: `quote` was an OBJECT (`{ text, position }`). Telegram requires
+    // a String, and rejected every such send with 400.
+    expect(typeof rp!.quote).toBe('string')
+    expect(rp!.quote).toBe('the position is a separate field')
+    // No fabricated position — Telegram locates the quote itself, and neither
+    // send site has a real position source.
+    expect(rp).not.toHaveProperty('quote_position')
+    expect(rp).not.toHaveProperty('quote_parse_mode')
+    // Delivered: the send resolved and the tool result is not a failure notice.
+    const out = res.content[0]?.text ?? ''
+    expect(out).not.toMatch(/fail/i)
+    expect(out).toMatch(/sent/i)
+  })
+
+  it('is NOT html-escaped — the quote must stay an exact substring of the original', async () => {
+    const h = makeHarness()
+    await sendReply(h.deps, {
+      args: {
+        chat_id: CHAT,
+        text: 'Checked.',
+        reply_to: 77,
+        quote_text: 'tom & jerry <b>not a tag</b>',
+      },
+      turn: null,
+    })
+    const rp = h.replyParams(h.sends()[0])!
+    expect(rp.quote).toBe('tom & jerry <b>not a tag</b>')
+  })
+
+  it('a quote Telegram cannot find degrades to an unquoted reply instead of losing the answer', async () => {
+    const h = makeHarness({
+      errors: [
+        { method: 'sendRichMessage', description: 'Bad Request: QUOTE_TEXT_INVALID' },
+      ],
+    })
+    const res = await sendReply(h.deps, {
+      args: {
+        chat_id: CHAT,
+        text: 'The answer still has to land.',
+        reply_to: 4242,
+        quote_text: 'a paraphrase that is not in the original',
+      },
+      turn: null,
+    })
+
+    const sends = h.sends()
+    expect(sends.length).toBe(2)
+    // First attempt carried the quote and was rejected …
+    expect(h.replyParams(sends[0])!.quote).toBe('a paraphrase that is not in the original')
+    // … the retry keeps the reply target and drops only the highlight.
+    const retry = h.replyParams(sends[1])!
+    expect(retry.message_id).toBe(4242)
+    expect(retry).not.toHaveProperty('quote')
+    expect(res.content[0]?.text ?? '').toMatch(/sent/i)
+  })
+
+  // A `<blockquote>` parse failure is a BODY error whose description happens to
+  // contain the substring "quote". If the quote classifier claims it, the ladder
+  // takes the quote branch, re-sends the SAME unparseable body without the
+  // quote, hits the same 400 with no secondary handling, and throws — so the
+  // pre-existing plain-text rescue never runs and the composed answer is lost.
+  // This is an outcome test on purpose: asserting the classifier returns false
+  // in isolation would not have caught it, because the loss happens in the
+  // ladder, not in the predicate.
+  for (const description of [
+    'Bad Request: can\'t parse entities: Can\'t find end tag corresponding to start tag "blockquote"',
+    'Bad Request: can\'t parse entities: Unsupported start tag "blockquote"',
+  ]) {
+    it(`a blockquote parse error on a QUOTED reply still reaches the plain-text rescue (${description.slice(-24)})`, async () => {
+      const h = makeHarness({
+        // Fails every time it is attempted — a body Telegram cannot parse does
+        // not start parsing on the second try.
+        errors: [{ method: 'sendRichMessage', description, times: 5 }],
+      })
+
+      const res = await sendReply(h.deps, {
+        args: {
+          chat_id: CHAT,
+          text: 'The answer must survive a blockquote parse error.',
+          reply_to: 4242,
+          quote_text: 'a real substring of the original',
+        },
+        turn: null,
+      })
+
+      const sends = h.sends()
+      // The rich attempt failed; the rescue must have gone out over a DIFFERENT
+      // method (plain text, no rich wrapper ⇒ the entity parser never runs).
+      const rescue = sends.filter((c) => c.method !== 'sendRichMessage')
+      expect(rescue.length).toBeGreaterThan(0)
+      // The answer actually landed, with its body intact.
+      expect(String(rescue[0].payload.text ?? '')).toContain(
+        'The answer must survive a blockquote parse error.',
+      )
+      expect(res.content[0]?.text ?? '').toMatch(/sent/i)
+      // And the quote branch was never taken: no second rich send stripped of
+      // the quote (that path is what swallowed the answer).
+      const richSends = sends.filter((c) => c.method === 'sendRichMessage')
+      expect(richSends.length).toBe(1)
+    })
+  }
+
+  it('omits quote entirely when quote_text is absent (plain reply unchanged)', async () => {
+    const h = makeHarness()
+    await sendReply(h.deps, {
+      args: { chat_id: CHAT, text: 'plain reply', reply_to: 12 },
+      turn: null,
+    })
+    const rp = h.replyParams(h.sends()[0])!
+    expect(rp).toEqual({ message_id: 12 })
+  })
+
+  it('stream_reply carries the same String shape to the wire', async () => {
+    const h = makeStreamHarness()
+    await handleStreamReply(
+      {
+        chat_id: CHAT,
+        text: 'streamed answer',
+        done: true,
+        reply_to: '881',
+        quote_text: 'exactly this fragment',
+      },
+      h.state,
+      h.deps,
+    )
+    const sends = h.sends()
+    expect(sends.length).toBeGreaterThan(0)
+    const rp = h.replyParams(sends[0])!
+    expect(rp.message_id).toBe(881)
+    expect(typeof rp.quote).toBe('string')
+    expect(rp.quote).toBe('exactly this fragment')
+    expect(rp).not.toHaveProperty('quote_position')
+  })
+
+  it('stream_reply drops an unfindable quote and still delivers the message', async () => {
+    const h = makeStreamHarness({
+      errors: [
+        {
+          method: 'sendRichMessage',
+          description: 'Bad Request: message quote not found in the original message',
+        },
+      ],
+    })
+    const res = await handleStreamReply(
+      {
+        chat_id: CHAT,
+        text: 'streamed answer',
+        done: true,
+        reply_to: '881',
+        quote_text: 'not present in the original',
+      },
+      h.state,
+      h.deps,
+    )
+    const sends = h.sends()
+    expect(sends.length).toBe(2)
+    expect(h.replyParams(sends[0])!.quote).toBe('not present in the original')
+    expect(h.replyParams(sends[1])!).toEqual({ message_id: 881 })
+    expect(res.messageId).not.toBeNull()
+  })
+
+  it('truncates an over-long quote to the Bot API 1024-char cap (a prefix is still an exact substring)', async () => {
+    const h = makeHarness()
+    const long = 'x'.repeat(1500)
+    await sendReply(h.deps, {
+      args: { chat_id: CHAT, text: 'ok', reply_to: 5, quote_text: long },
+      turn: null,
+    })
+    const rp = h.replyParams(h.sends()[0])!
+    expect((rp.quote as string).length).toBe(1024)
+    expect(long.startsWith(rp.quote as string)).toBe(true)
+  })
+
+  it('never cuts an astral character in half at the 1024 boundary', async () => {
+    const h = makeHarness()
+    // '𝄞' is one astral char = TWO UTF-16 units. 1023 filler units put its high
+    // surrogate at index 1023 and its low surrogate at 1024, so a naive
+    // slice(0, 1024) would emit a LONE HIGH SURROGATE — not valid UTF-8 on the
+    // wire, and no longer a substring of anything.
+    const long = 'x'.repeat(1023) + '𝄞' + 'y'.repeat(200)
+    await sendReply(h.deps, {
+      args: { chat_id: CHAT, text: 'ok', reply_to: 5, quote_text: long },
+      turn: null,
+    })
+    const quote = h.replyParams(h.sends()[0])!.quote as string
+    // Cut back to 1023 rather than splitting the pair …
+    expect(quote.length).toBe(1023)
+    // … so no unpaired surrogate survives …
+    for (const ch of quote) expect(ch.codePointAt(0)! >= 0xd800 && ch.codePointAt(0)! <= 0xdfff).toBe(false)
+    // … and it round-trips through UTF-8 unchanged (a lone surrogate would be
+    // replaced by U+FFFD here).
+    expect(Buffer.from(quote, 'utf8').toString('utf8')).toBe(quote)
+    // … and it is still an exact prefix of the original.
+    expect(long.startsWith(quote)).toBe(true)
+  })
+})