botmux 3.10.0 → 3.23.1

package/dist/im/lark/message-parser.js

@@ -1,1261 +0,0 @@
-import { getMessageDetail } from './client.js';
-import { logger } from '../../utils/logger.js';
-import { REPLY_CARD_FOOTER_ELEMENT_ID, } from './reply-card-footer-signature.js';
-import { hasBotmuxCallbackMarker } from './callback-button-marker.js';
-/**
- * Extract a mention's open_id, tolerating BOTH Lark shapes for `mention.id`:
- *
- *   - WebSocket event (im.message.receive_v1): `id` is an OBJECT
- *       { open_id, union_id, user_id }            ← what botmux has always seen
- *   - Message REST API (im.message.get / list):  `id` is a bare STRING "ou_xxx"
- *       with a sibling `id_type` ("open_id" | "union_id" | "user_id")
- *
- * The two diverged in production: the API already ships the flat-string form
- * while the event still ships the object form. If Lark ever converges the event
- * onto the string form, every `m.id.open_id` read across botmux would silently
- * become `undefined` and @-detection (isBotMentioned) would fail closed with no
- * log. Reading through this helper keeps every caller robust regardless of
- * which shape arrives.
- *
- * For the string form we return the value only when it actually IS an open_id
- * (id_type === 'open_id', or absent — mentions are open_id-keyed by default). A
- * string carrying a non-open_id id_type (union_id / user_id, which Lark may
- * return when the app lacks the open_id scope) yields `undefined` rather than
- * being mis-compared against a botOpenId.
- */
-export function mentionOpenId(m) {
-    const id = m?.id;
-    if (id == null)
-        return undefined;
-    if (typeof id === 'object')
-        return id.open_id || undefined;
-    if (typeof id === 'string') {
-        if (m?.id_type && m.id_type !== 'open_id')
-            return undefined;
-        return id || undefined;
-    }
-    return undefined;
-}
-/** Extract all stable ids Lark provides for @mentions, across WS and REST shapes. */
-export function mentionIdentity(m) {
-    const id = m?.id;
-    const out = {
-        key: m?.key,
-        name: m?.name,
-        idType: m?.id_type,
-    };
-    if (id && typeof id === 'object') {
-        out.openId = id.open_id || undefined;
-        out.userId = id.user_id || undefined;
-        out.unionId = id.union_id || undefined;
-        out.appId = id.app_id || undefined;
-        return out;
-    }
-    if (typeof id === 'string' && id) {
-        if (!m?.id_type || m.id_type === 'open_id')
-            out.openId = id;
-        else if (m.id_type === 'user_id')
-            out.userId = id;
-        else if (m.id_type === 'union_id')
-            out.unionId = id;
-        else if (m.id_type === 'app_id')
-            out.appId = id;
-    }
-    return out;
-}
-/** id_type across both shapes Lark uses (snake_case REST, camelCase legacy). */
-export function mentionIdType(m) {
-    if (!m || typeof m !== 'object')
-        return undefined;
-    if (typeof m.id_type === 'string')
-        return m.id_type;
-    if (typeof m.idType === 'string')
-        return m.idType;
-    return undefined;
-}
-/**
- * Extract a bot mention's app_id across ALL shapes Lark has been observed to
- * use. app_id-form bot mentions must never flow through mentionOpenId() (which
- * is persisted/used as an open_id), so they are matched separately:
- *   1. top-level camelCase `m.appId`
- *   2. top-level snake_case `m.app_id`
- *   3. string `m.id` with id_type === 'app_id' (REST bare-string OR legacy)
- *   4. object `m.id.app_id`
- * Keep this exhaustive: the realtime @-gate (isBotMentioned) depends on it, so
- * dropping a shape silently un-@'s a bot in that shape.
- */
-export function mentionAppId(m) {
-    if (!m || typeof m !== 'object')
-        return undefined;
-    if (typeof m.appId === 'string')
-        return m.appId;
-    if (typeof m.app_id === 'string')
-        return m.app_id;
-    const idType = mentionIdType(m);
-    if (idType === 'app_id' && typeof m.id === 'string')
-        return m.id;
-    if (m.id && typeof m.id === 'object' && typeof m.id.app_id === 'string')
-        return m.id.app_id;
-    return undefined;
-}
-/**
- * Whether a message explicitly @mentions the given bot. SHARED single source of
- * truth for the @-gate across every consumer (realtime routing's isBotMentioned,
- * the 30s poll backfill, and the dashboard preview/run-preview collector) so the
- * "explicit @ hands off to normal routing, not the listener" rule can never
- * diverge between legs. Matches by open_id OR by app_id (via mentionAppId, which
- * covers every observed app_id shape), and also scans post-content inline `at`
- * tags (post messages may not populate `mentions`).
- */
-export function messageMentionsBot(message, larkAppId, botOpenId) {
-    if (!botOpenId && !larkAppId)
-        return false;
-    const mentions = message?.mentions ?? [];
-    for (const m of mentions) {
-        if (botOpenId && mentionOpenId(m) === botOpenId)
-            return true;
-        if (larkAppId && mentionAppId(m) === larkAppId)
-            return true;
-    }
-    // Post-content inline `at` tags (bot-sent post messages may omit `mentions`).
-    // Realtime events carry the body in `content`; the REST message-list API
-    // (poll + dashboard preview/run-preview) carries it in `body.content` — read
-    // both so the @-gate is identical across all three legs.
-    if (botOpenId) {
-        const rawContent = message?.content ?? message?.body?.content;
-        try {
-            const content = JSON.parse(rawContent ?? '{}');
-            const inner = content.zh_cn ?? content.en_us ?? content;
-            if (Array.isArray(inner?.content)) {
-                for (const paragraph of inner.content) {
-                    if (!Array.isArray(paragraph))
-                        continue;
-                    for (const node of paragraph) {
-                        if (node?.tag === 'at' && node.user_id === botOpenId)
-                            return true;
-                    }
-                }
-            }
-        }
-        catch { /* ignore parse errors */ }
-    }
-    return false;
-}
-export function extractMentionIdentities(message) {
-    const out = (message?.mentions ?? []).map(mentionIdentity);
-    try {
-        const content = JSON.parse(message?.content ?? '{}');
-        const inner = content.zh_cn ?? content.en_us ?? content;
-        if (Array.isArray(inner?.content)) {
-            for (const paragraph of inner.content) {
-                if (!Array.isArray(paragraph))
-                    continue;
-                for (const node of paragraph) {
-                    if (node?.tag !== 'at')
-                        continue;
-                    // In post/rich-text content Lark carries the mentionee's OPEN_ID in the
-                    // at-node's `user_id` field (cf. isBotMentioned, which compares
-                    // node.user_id against botOpenId), NOT a tenant user_id. Map it to
-                    // openId only — mislabeling it as userId would both miss a userId-only
-                    // target and pollute the userId leg with an open_id value.
-                    out.push({
-                        name: node.user_name,
-                        openId: node.user_id,
-                    });
-                }
-            }
-        }
-    }
-    catch { /* ignore non-JSON content */ }
-    return out;
-}
-export function mentionUnionId(m) {
-    const id = m?.id;
-    if (id == null)
-        return undefined;
-    if (typeof id === 'object')
-        return id.union_id || undefined;
-    if (typeof id === 'string') {
-        if (m?.id_type !== 'union_id')
-            return undefined;
-        return id || undefined;
-    }
-    return undefined;
-}
-/**
- * When the WebSocket event delivers message_type "nonsupport", call the REST API
- * to fetch the real message content and patch the event data in-place.
- *
- * Also handles `interactive`: the WebSocket event only carries a simplified
- * fallback view of cards (often literally "请升级至最新版本客户端，以查看内容"),
- * so we fetch the real card JSON (including v2 `body.elements`) via REST.
- */
-export async function resolveNonsupportMessage(data, larkAppId) {
-    const type = data.message.message_type;
-    if (type !== 'nonsupport' && type !== 'interactive')
-        return;
-    // Interactive cards: resolve to the COMPLETE merged text (union of both
-    // im.message.get representations) so forwarded cards reach the model fully
-    // parsed — same resolver `botmux history` uses. resolveEventCard falls back
-    // to local user_dsl unwrap if REST is unavailable (e.g. cross-tenant).
-    if (type === 'interactive') {
-        await resolveEventCard(data, larkAppId);
-        return;
-    }
-    try {
-        const detail = await getMessageDetail(larkAppId, data.message.message_id);
-        const msg = detail?.items?.[0];
-        if (!msg)
-            return;
-        const realType = msg.msg_type;
-        const realContent = msg.body?.content;
-        if (realType && realContent) {
-            logger.info(`[parser] Resolved ${type} → ${realType} for ${data.message.message_id}`);
-            data.message.message_type = realType;
-            data.message.content = realContent;
-        }
-    }
-    catch (err) {
-        logger.debug(`[parser] Failed to resolve ${type} message ${data.message.message_id}: ${err}`);
-    }
-}
-/**
- * Lark bundles the real v2 card JSON inside a `user_dsl` string on the
- * simplified interactive payload. When present, return the unwrapped v2
- * body so downstream extractors see schema/body.elements directly.
- */
-export function unwrapUserDslContent(rawContent) {
-    try {
-        const outer = JSON.parse(rawContent);
-        if (typeof outer?.user_dsl !== 'string')
-            return null;
-        const inner = JSON.parse(outer.user_dsl);
-        if (!inner || typeof inner !== 'object')
-            return null;
-        if (!inner.body && !inner.elements && !inner.header)
-            return null;
-        return JSON.stringify(inner);
-    }
-    catch {
-        return null;
-    }
-}
-function unwrapUserDsl(data) {
-    const unwrapped = unwrapUserDslContent(data.message.content);
-    if (unwrapped === null)
-        return false;
-    data.message.content = unwrapped;
-    logger.info(`[parser] Unwrapped user_dsl for ${data.message.message_id}`);
-    return true;
-}
-/**
- * Lark's simplified "upgrade your client" card fallback marker. When this text
- * shows up in a card's resolved content, the real body was stripped and must be
- * recovered via `im.message.get` (with `card_msg_content_type=user_card_content`).
- */
-export const CARD_UPGRADE_FALLBACK = '请升级至最新版本客户端';
-/**
- * Broad check — content carries the upgrade notice *somewhere*. Used to decide
- * whether a card needs REST re-resolution. Deliberately a substring match: a
- * false positive only costs one extra `im.message.get`, and this is the only
- * way to catch **embedded** fallbacks — complex cards (e.g. Argos alarm cards
- * with nested sub-cards) render fine at the top level but bury one or more
- * `请升级…` placeholders mid-body where an anchored check would miss them.
- */
-export function cardContentHasUpgradeFallback(content) {
-    return content.includes(CARD_UPGRADE_FALLBACK);
-}
-/**
- * Narrow check — content *is* essentially just Lark's upgrade notice, not a
- * body that merely mentions it. Anchored at the start after stripping leading
- * `[图片]` / `[文件 N]` placeholders (the bare fallback renders as
- * `[图片]请升级至最新版本客户端，以查看内容`). Used as the replace gate when
- * re-resolving via REST: it keeps a card that legitimately quotes the phrase
- * mid-text — e.g. a message discussing this very fallback — from being
- * discarded, while still rejecting a REST view that came back as a bare
- * fallback. Makes no structural assumptions, so it's safe for the
- * simplified-but-real Format A shape (no schema/body/header) message.list
- * returns.
- */
-export function isPureCardUpgradeFallback(content) {
-    const stripped = content.replace(/^(?:\s*\[(?:图片|文件)[^\]]*\])+/, '').trimStart();
-    return stripped.startsWith(CARD_UPGRADE_FALLBACK);
-}
-export function createImgNumberer() {
-    const map = new Map();
-    let imgCounter = 0;
-    let fileCounter = 0;
-    return {
-        assign(key) {
-            const existing = map.get(key);
-            if (existing !== undefined)
-                return { num: existing, isNew: false };
-            // Key prefix selects the counter so image/file numbering stays
-            // independent (mirrors formatAttachmentsHint's per-type imgN/fileN).
-            // Unknown prefixes share the image counter as a safe default.
-            const num = key.startsWith('file:') ? ++fileCounter : ++imgCounter;
-            map.set(key, num);
-            return { num, isNew: true };
-        },
-    };
-}
-export function extractResources(msgType, rawContent, numberer) {
-    const content = normalizeApiMessageContent(msgType, rawContent);
-    const nb = numberer ?? createImgNumberer();
-    const pushIfNew = (resources, r) => {
-        if (nb.assign(`${r.type}:${r.key}`).isNew)
-            resources.push(r);
-    };
-    try {
-        const parsed = JSON.parse(content);
-        if (msgType === 'image') {
-            const resources = [];
-            const imageKey = parsed.image_key;
-            if (imageKey)
-                pushIfNew(resources, { type: 'image', key: imageKey, name: `${imageKey}.jpg` });
-            return resources;
-        }
-        if (msgType === 'file') {
-            const resources = [];
-            const fileKey = parsed.file_key;
-            if (fileKey)
-                pushIfNew(resources, { type: 'file', key: fileKey, name: parsed.file_name ?? fileKey });
-            return resources;
-        }
-        if (msgType === 'post') {
-            const resources = [];
-            const { content: contentBlocks } = resolvePostBody(parsed);
-            for (const block of contentBlocks) {
-                const nodes = Array.isArray(block) ? block : [block];
-                for (const node of nodes) {
-                    if ((node.tag === 'img' || node.tag === 'media') && node.image_key) {
-                        pushIfNew(resources, { type: 'image', key: node.image_key, name: `${node.image_key}.jpg` });
-                    }
-                    if (node.tag === 'file' && node.file_key) {
-                        pushIfNew(resources, { type: 'file', key: node.file_key, name: node.file_name ?? node.file_key });
-                    }
-                }
-            }
-            return resources;
-        }
-        if (msgType === 'interactive') {
-            const resources = [];
-            // v2 cards nest elements under `body`; fall back to legacy top-level.
-            const rootElements = Array.isArray(parsed.body?.elements)
-                ? parsed.body.elements
-                : Array.isArray(parsed.elements) ? parsed.elements : null;
-            if (rootElements) {
-                const isApiFormat = rootElements.length > 0 && Array.isArray(rootElements[0]);
-                if (isApiFormat) {
-                    // Format A: [[{tag:"img",image_key:"..."}, ...], ...]
-                    for (const block of rootElements) {
-                        if (!Array.isArray(block))
-                            continue;
-                        for (const node of block) {
-                            const key = node.image_key ?? node.img_key;
-                            if ((node.tag === 'img' || node.tag === 'image') && key) {
-                                pushIfNew(resources, { type: 'image', key, name: `${key}.jpg` });
-                            }
-                        }
-                    }
-                }
-                else {
-                    for (const el of rootElements) {
-                        extractElementImages(el, resources, pushIfNew);
-                    }
-                }
-            }
-            return resources;
-        }
-    }
-    catch {
-        // ignore parse errors
-    }
-    return [];
-}
-export function parseEventMessage(data, numberer = createImgNumberer()) {
-    const { sender, message } = data;
-    // Trace non-text messages at debug only (DEBUG=1 gated). The raw card/post
-    // content can be many KB and may include attachment metadata that's
-    // noisy or sensitive — truncate to ~500 chars so DEBUG logs stay scannable.
-    if (message.message_type !== 'text' && logger.isDebug()) {
-        const raw = message.content ?? '';
-        const trimmed = raw.length > 500 ? raw.slice(0, 500) + `…(+${raw.length - 500}b)` : raw;
-        logger.debug(`[parser] type=${message.message_type} content=${trimmed} keys=${Object.keys(message).join(',')}`);
-    }
-    // Share numberer so in-body [图片 N] placeholders use the same numbers as
-    // the attachment list. Resources first → numbers assigned; text second →
-    // reuses them.
-    const resources = extractResources(message.message_type, message.content, numberer);
-    // Extract structured mentions
-    const mentions = message.mentions && message.mentions.length > 0
-        ? message.mentions.map(m => ({
-            key: m.key,
-            name: m.name,
-            openId: mentionOpenId(m),
-            userId: mentionIdentity(m).userId,
-            unionId: mentionUnionId(m),
-            idType: m.id_type,
-        }))
-        : undefined;
-    const parsed = {
-        messageId: message.message_id,
-        rootId: message.root_id ?? '',
-        threadId: message.thread_id || undefined,
-        parentId: message.parent_id || undefined,
-        senderId: sender.sender_id?.open_id ?? '',
-        senderUnionId: sender.sender_id?.union_id,
-        senderType: sender.sender_type,
-        msgType: message.message_type,
-        content: extractTextContent(message.message_type, message.content, message.mentions, numberer),
-        createTime: message.create_time,
-        mentions,
-    };
-    return { parsed, resources };
-}
-export function parseApiMessage(msg, numberer) {
-    const msgType = msg.msg_type ?? 'text';
-    const rawContent = msg.body?.content ?? '';
-    // sender_name is only present when the fetch opted in via with_sender_name=true
-    // (all client.ts message read paths do). Covers bot senders too — the one case
-    // local rosters can't resolve for third-party bots.
-    const senderName = typeof msg.sender?.sender_name === 'string' && msg.sender.sender_name.trim()
-        ? msg.sender.sender_name.trim()
-        : undefined;
-    return {
-        messageId: msg.message_id ?? '',
-        rootId: msg.root_id ?? msg.thread_id ?? '',
-        senderId: msg.sender?.id ?? '',
-        senderType: msg.sender?.sender_type ?? 'unknown',
-        ...(senderName ? { senderName } : {}),
-        msgType,
-        content: extractTextContent(msgType, normalizeApiMessageContent(msgType, rawContent), undefined, numberer),
-        createTime: msg.create_time ?? '',
-    };
-}
-function normalizeApiMessageContent(msgType, rawContent) {
-    if (msgType !== 'interactive')
-        return rawContent;
-    return unwrapUserDslContent(rawContent) ?? rawContent;
-}
-/** Resolve post body from either wrapped {"zh_cn":{title,content}} or unwrapped {title,content} format */
-function resolvePostBody(parsed) {
-    // Unwrapped: has content array directly
-    if (Array.isArray(parsed.content)) {
-        return { title: parsed.title ?? '', content: parsed.content };
-    }
-    // Wrapped in language key: {"zh_cn": {title, content}}
-    for (const key of Object.keys(parsed)) {
-        const val = parsed[key];
-        if (val && typeof val === 'object' && Array.isArray(val.content)) {
-            return { title: val.title ?? '', content: val.content };
-        }
-    }
-    return { title: '', content: [] };
-}
-/**
- * Strip leading `@<name>` mentions from a resolved-content string so callers
- * can detect daemon `/commands` even when the user @-mentioned the bot first.
- *
- * Uses the structured mentions list when available (handles names with spaces);
- * falls back to a `@\S+` regex for cases where Lark didn't populate mentions
- * (e.g. some post messages where the at-tag becomes a plain `@<user_name>`
- * string in the rendered text).
- */
-export function stripLeadingMentions(content, mentions) {
-    let s = content.trimStart();
-    if (mentions && mentions.length > 0) {
-        // Sort by name length desc so "@Claude分身" wins over "@Claude" when both
-        // could startsWith — otherwise the short name eats "@Claude" and leaves
-        // "分身 @CoCo /close" stranded, breaking slash-command detection in
-        // multi-bot @ chains like "@Claude @Claude分身 @CoCo /close".
-        const sortedMentions = [...mentions].sort((a, b) => b.name.length - a.name.length);
-        let changed = true;
-        while (changed) {
-            changed = false;
-            for (const m of sortedMentions) {
-                const tag = `@${m.name}`;
-                if (s.startsWith(tag)) {
-                    s = s.slice(tag.length).trimStart();
-                    changed = true;
-                    break;
-                }
-            }
-        }
-        return s;
-    }
-    // No mentions list (e.g. some post messages) — best-effort strip leading
-    // single-word @<word> patterns. Multi-word names without a mentions list
-    // can't be reliably detected and will be left in place.
-    let changed = true;
-    while (changed) {
-        changed = false;
-        const m = s.match(/^@\S+/);
-        if (m) {
-            s = s.slice(m[0].length).trimStart();
-            changed = true;
-        }
-    }
-    return s;
-}
-function resolveMentions(text, mentions) {
-    if (!mentions || mentions.length === 0) {
-        // No mention info available — strip placeholders
-        return text.replace(/@_user_\d+/g, '').replace(/[^\S\r\n]{2,}/g, ' ').trim();
-    }
-    let result = text;
-    for (const m of mentions) {
-        result = result.replace(m.key, `@${m.name}`);
-    }
-    return result.trim();
-}
-function normalizeFenceLanguage(lang) {
-    return typeof lang === 'string' ? lang.trim().replace(/\s+/g, '_') : '';
-}
-function renderPostCodeBlock(node) {
-    const raw = typeof node.text === 'string'
-        ? node.text
-        : typeof node.content === 'string'
-            ? node.content
-            : typeof node.code === 'string'
-                ? node.code
-                : '';
-    const code = raw.replace(/\n+$/, '');
-    const lang = normalizeFenceLanguage(node.language ?? node.lang);
-    const longestFence = Math.max(2, ...[...code.matchAll(/`+/g)].map(m => m[0].length));
-    const fence = '`'.repeat(longestFence + 1);
-    return `\n${fence}${lang}\n${code}\n${fence}\n`;
-}
-function renderPostNode(node, numberer) {
-    if (node.tag === 'text')
-        return node.text ?? '';
-    if (node.tag === 'a')
-        return node.text ?? node.href ?? '';
-    if (node.tag === 'at')
-        return `@${node.user_name ?? 'unknown'}`;
-    if (node.tag === 'code_block')
-        return renderPostCodeBlock(node);
-    if (node.tag === 'img' || node.tag === 'media') {
-        const key = node.image_key ?? node.file_key;
-        if (key && numberer)
-            return `[图片 ${numberer.assign(`image:${key}`).num}]`;
-        return '[图片]';
-    }
-    if (node.tag === 'file') {
-        const key = node.file_key;
-        const name = node.file_name ?? '';
-        if (key && numberer) {
-            const n = numberer.assign(`file:${key}`).num;
-            return name ? `[文件 ${n}: ${name}]` : `[文件 ${n}]`;
-        }
-        return name ? `[文件: ${name}]` : '[文件]';
-    }
-    return '';
-}
-function joinPostNodeText(parts) {
-    return parts.join('').replace(/\n{3,}/g, '\n\n').trim();
-}
-function extractTextContent(msgType, rawContent, mentions, numberer) {
-    try {
-        if (msgType === 'text') {
-            const parsed = JSON.parse(rawContent);
-            return resolveMentions(parsed.text ?? rawContent, mentions);
-        }
-        if (msgType === 'post') {
-            const parsed = JSON.parse(rawContent);
-            const { title, content } = resolvePostBody(parsed);
-            const body = content
-                .map((paragraph) => {
-                const nodes = Array.isArray(paragraph) ? paragraph : [paragraph];
-                return joinPostNodeText(nodes.map((node) => renderPostNode(node, numberer)));
-            })
-                .filter(Boolean)
-                .join('\n');
-            return title ? `${title}\n${body}` : body;
-        }
-        if (msgType === 'image') {
-            try {
-                const p = JSON.parse(rawContent);
-                if (p.image_key && numberer)
-                    return `[图片 ${numberer.assign(`image:${p.image_key}`).num}]`;
-            }
-            catch { /* fall through */ }
-            return '[图片]';
-        }
-        if (msgType === 'file') {
-            try {
-                const p = JSON.parse(rawContent);
-                if (p.file_key && numberer) {
-                    const n = numberer.assign(`file:${p.file_key}`).num;
-                    return p.file_name ? `[文件 ${n}: ${p.file_name}]` : `[文件 ${n}]`;
-                }
-                return `[文件: ${p.file_name ?? 'unknown'}]`;
-            }
-            catch {
-                return '[文件]';
-            }
-        }
-        if (msgType === 'interactive') {
-            return extractCardContent(rawContent, numberer);
-        }
-        if (msgType === 'merge_forward') {
-            return '[合并转发消息]';
-        }
-        return rawContent;
-    }
-    catch {
-        return rawContent;
-    }
-}
-/**
- * botmux-generated reply-card footer signature. New cards carry both a visible
- * versioned link marker and an exact element id. The canonical repository URL
- * is NOT a signature: it is valid body content. For pre-signature cards we
- * recognize only the exact old default-brand + recipient chrome shape; a
- * default-brand-only old card is intentionally preserved because it is
- * indistinguishable from an ordinary repository link.
- */
-const LEGACY_DEFAULT_FOOTER_URL = 'https://github.com/deepcoldy/botmux';
-const FOOTER_MARKER_HASHES = new Set(['#reply-card-footer', '#reply-card-footer-v1']);
-function isBotmuxFooterMarkerUrl(value) {
-    if (typeof value !== 'string')
-        return false;
-    try {
-        const url = new URL(value);
-        return url.protocol === 'https:'
-            && url.hostname === 'github.com'
-            && url.port === ''
-            && url.username === ''
-            && url.password === ''
-            && url.search === ''
-            && decodeURIComponent(url.pathname) === '/deepcoldy/botmux'
-            && FOOTER_MARKER_HASHES.has(url.hash);
-    }
-    catch {
-        return false;
-    }
-}
-function isBotmuxFooterMarkerText(value) {
-    return value === '·' || value === '\u200B';
-}
-function isBotmuxFooterMarkerAnchor(href, text) {
-    return isBotmuxFooterMarkerText(text) && isBotmuxFooterMarkerUrl(href);
-}
-function greyFontInner(line) {
-    const font = /^\s*<font\b([^>]*)>([\s\S]*?)<\/font>\s*$/i.exec(line);
-    if (!font || !/\bcolor\s*=\s*(?:"grey"|'grey'|grey)(?:\s|$)/i.test(font[1]))
-        return null;
-    return font[2];
-}
-function hasExactMarkerMarkdown(value) {
-    // Match only the protocol's reserved anchor texts. A generic Markdown-link
-    // regex can start at an unmatched `[` inside a custom brand and swallow the
-    // later marker before it gets a chance to match.
-    for (const match of value.matchAll(/\[(·|\u200B)\]\(([^)\s]+)\)/gu)) {
-        if (isBotmuxFooterMarkerAnchor(match[2], match[1]))
-            return true;
-    }
-    return false;
-}
-function isSignedFormatBFooterLine(line) {
-    const inner = greyFontInner(line);
-    return inner !== null && hasExactMarkerMarkdown(inner);
-}
-/** Strict compatibility recognizer for Format A cards emitted before the
- * versioned marker existed. Recipient chrome is required; without it, the old
- * default-brand-only footer is indistinguishable from a real repository link. */
-function isMeaningfulFormatANode(node) {
-    if (!node || typeof node !== 'object')
-        return false;
-    const value = node;
-    if (value.tag === 'text')
-        return typeof value.text === 'string' && value.text.trim().length > 0;
-    return true;
-}
-function isLegacyFormatAFooterParagraph(paragraph) {
-    const nodes = paragraph.filter(isMeaningfulFormatANode);
-    const first = nodes[0];
-    if (first?.tag !== 'a'
-        || first.href !== LEGACY_DEFAULT_FOOTER_URL
-        || typeof first.text !== 'string'
-        || first.text.trim().toLowerCase() !== 'botmux') {
-        return false;
-    }
-    let label = '';
-    let sawAt = false;
-    for (const node of nodes.slice(1)) {
-        if (node.tag === 'at') {
-            sawAt = true;
-            continue;
-        }
-        if (node.tag !== 'text' || typeof node.text !== 'string')
-            return false;
-        if (sawAt && node.text.trim() !== '')
-            return false;
-        label += node.text;
-    }
-    return sawAt && /^\s*·\s*(?:发送给：|Sent to:)\s*$/i.test(label);
-}
-/** Strict compatibility recognizer for an old original-format footer line.
- * Grey styling plus exact default brand, locale label, and one or more native
- * mentions are all required; styling or the canonical URL alone proves
- * nothing. */
-function isLegacyFormatBFooterLine(line) {
-    const inner = greyFontInner(line);
-    if (inner === null)
-        return false;
-    return /^\s*\[botmux\]\(https:\/\/github\.com\/deepcoldy\/botmux\)\s*·\s*(?:发送给：|Sent to:)\s*(?:<at\s+id=(?:"[^"]+"|'[^']+'|[^\s>]+)\s*><\/at>\s*)+$/i
-        .test(inner);
-}
-function extractRenderedCardContent(rawContent) {
-    const match = rawContent.match(/^\s*<card(?:\s+title="([^"]*)")?\s*>([\s\S]*)<\/card>\s*$/);
-    if (!match)
-        return undefined;
-    const title = match[1]?.trim();
-    const body = match[2]
-        .split('\n')
-        .map(line => line.trimEnd())
-        .filter(line => !isSignedFormatBFooterLine(line))
-        .join('\n')
-        .trim();
-    return [
-        title ? `[卡片: ${title}]` : '',
-        body,
-    ].filter(Boolean).join('\n') || '[卡片]';
-}
-/**
- * Extract human-readable text from an interactive card.
- *
- * Lark API returns card content in a **simplified format** (not the original card JSON):
- *   { title: "...", elements: [[{tag:"text",text:"..."}, ...], ...] }
- * This is similar to post message body.  We also handle the original card JSON
- * (header/config/elements with tag objects) for locally-cached cards.
- */
-export function extractCardContent(rawContent, numberer) {
-    const rendered = extractRenderedCardContent(rawContent);
-    if (rendered !== undefined)
-        return rendered;
-    try {
-        const card = JSON.parse(rawContent);
-        // Pre-resolved merged text injected by resolveMergedCardContent (the
-        // A+B union for complex cards). Return it verbatim so the merge flows
-        // through both parseEventMessage (live) and parseApiMessage (history).
-        if (typeof card[RESOLVED_TEXT_KEY] === 'string')
-            return card[RESOLVED_TEXT_KEY];
-        // Template-based card — no inline content to extract
-        if (card.type === 'template') {
-            return '[卡片 (模板)]';
-        }
-        const parts = [];
-        // --- Format A: Lark API simplified format ---
-        // { title: "...", elements: [[{tag,text}, ...], ...] }
-        // 只有 title 存在时才 push 标识行；没 title 的卡片（例如 markdown 自动转
-        // 卡片的 outgoing 消息）让 elements 内容自己说话，避免在正文前堆一行
-        // 多余的 `[卡片]`。整张卡片真的没内容时下方 `parts.join('\n') || '[卡片]'`
-        // 会兜底返回占位。
-        const title = card.title ?? card.header?.title?.content;
-        if (title)
-            parts.push(`[卡片: ${title}]`);
-        // v2 cards nest elements under `body`; fall back to legacy top-level.
-        const rootElements = Array.isArray(card.body?.elements)
-            ? card.body.elements
-            : Array.isArray(card.elements) ? card.elements : null;
-        const imgLabel = (key) => numberer ? `[图片 ${numberer.assign(`image:${key}`).num}]` : '[图片]';
-        if (rootElements) {
-            const isApiFormat = rootElements.length > 0 && Array.isArray(rootElements[0]);
-            if (isApiFormat) {
-                // Format A: [[{tag:"text",text:"..."}, {tag:"img",...}, {tag:"button",...}], ...]
-                let lastMeaningfulParagraphIndex = -1;
-                for (let i = rootElements.length - 1; i >= 0; i--) {
-                    const candidate = rootElements[i];
-                    if (Array.isArray(candidate) && candidate.some(isMeaningfulFormatANode)) {
-                        lastMeaningfulParagraphIndex = i;
-                        break;
-                    }
-                }
-                for (let paragraphIndex = 0; paragraphIndex < rootElements.length; paragraphIndex++) {
-                    const paragraph = rootElements[paragraphIndex];
-                    if (!Array.isArray(paragraph))
-                        continue;
-                    if (paragraphIndex === lastMeaningfulParagraphIndex
-                        && isLegacyFormatAFooterParagraph(paragraph)) {
-                        continue;
-                    }
-                    const textNodes = [];
-                    const buttons = [];
-                    let inSignedFooter = false;
-                    for (const node of paragraph) {
-                        if (inSignedFooter)
-                            continue;
-                        if (node.tag === 'text') {
-                            if (node.text)
-                                textNodes.push(node.text);
-                        }
-                        else if (node.tag === 'a') {
-                            if (isBotmuxFooterMarkerAnchor(node.href, node.text)) {
-                                inSignedFooter = true;
-                                continue;
-                            }
-                            // Keep the href so links survive — Format A separates text/href,
-                            // and dropping href loses real content (规则配置/详情/Trace 链接).
-                            // Bare long URLs (e.g. Argos Kibana 处理建议) come back with
-                            // text === href, both cut at the same offset, and the remainder
-                            // spilling into the FOLLOWING text nodes of the same paragraph.
-                            // Emitting `text(href)` would double the truncated URL and
-                            // corrupt the line; a single copy lets join('') below reconnect
-                            // the full URL from those remainder nodes.
-                            const t = node.text ?? '';
-                            if (t && t === node.href)
-                                textNodes.push(t);
-                            else
-                                textNodes.push(node.href && t ? `${t}(${node.href})` : (t || node.href || ''));
-                        }
-                        else if (node.tag === 'at')
-                            textNodes.push(`@${node.user_name ?? 'unknown'}`);
-                        else if (node.tag === 'img' || node.tag === 'image') {
-                            const k = node.image_key ?? node.img_key;
-                            if (k)
-                                textNodes.push(imgLabel(k));
-                        }
-                        else if (node.tag === 'button') {
-                            // Same jump-URL policy as Format B: simple cards reach history
-                            // via the Format A list view WITHOUT a resolve pass, so dropping
-                            // the URL here would lose button links on that main path.
-                            if (isBotmuxCallbackButton(node))
-                                continue;
-                            const btnText = typeof node.text === 'string' ? node.text : node.text?.content;
-                            if (btnText) {
-                                const url = buttonOpenUrl(node);
-                                buttons.push(url ? `[${btnText}](${url})` : `[${btnText}]`);
-                            }
-                        }
-                        else if (node.tag === 'input') {
-                            const ph = typeof node.placeholder === 'string' ? node.placeholder : node.placeholder?.content;
-                            if (ph)
-                                buttons.push(`[输入框: ${ph}]`);
-                        }
-                        else if (node.tag === 'select_static' || node.tag === 'multi_select_static' || node.tag === 'overflow') {
-                            const ph = typeof node.placeholder === 'string' ? node.placeholder : node.placeholder?.content;
-                            const opts = Array.isArray(node.options)
-                                ? node.options.map((o) => (typeof o.text === 'string' ? o.text : o.text?.content)).filter(Boolean)
-                                : [];
-                            const head = ph ? `[下拉: ${ph}` : '[下拉';
-                            buttons.push(opts.length ? `${head} | 选项: ${opts.join(' / ')}]` : `${head}]`);
-                        }
-                    }
-                    const line = textNodes.join('').trim();
-                    if (line) {
-                        if (inSignedFooter) {
-                            const lastBreak = line.lastIndexOf('\n');
-                            if (lastBreak >= 0) {
-                                const beforeFooter = line.slice(0, lastBreak).trim();
-                                if (beforeFooter)
-                                    parts.push(beforeFooter);
-                            }
-                            // No newline before the signed marker means the whole paragraph is
-                            // footer chrome (brand/usage/recipient). Keep nothing.
-                        }
-                        else {
-                            parts.push(line);
-                        }
-                    }
-                    if (buttons.length)
-                        parts.push(buttons.join(' '));
-                }
-            }
-            else {
-                for (const el of rootElements) {
-                    extractElementText(el, parts, imgLabel);
-                }
-            }
-        }
-        // Drop the botmux footer chrome so a receiving bot's prompt isn't polluted
-        // by the grey `botmux` badge / `发送给：@owner` line. Line-level (not
-        // part-level) so a footer never takes adjacent real content with it.
-        // Stable legacy markers may occur inside a multiline element, so remove
-        // only their exact line. Marker-less footer-shaped text is deliberately
-        // preserved: an arbitrary custom brand is indistinguishable from user data.
-        let visibleParts = parts
-            .map(part => part
-            .split('\n')
-            .filter(line => !isSignedFormatBFooterLine(line))
-            .join('\n'))
-            .filter(part => !!part.trim());
-        // Marker-less legacy compatibility is intentionally tail-only. Old
-        // botmux footers were always last; applying this heuristic in the middle
-        // of a foreign card would turn a footer-shaped body note into data loss.
-        if (visibleParts.length > 0) {
-            const lastIndex = visibleParts.length - 1;
-            const lines = visibleParts[lastIndex].split('\n');
-            let lastLine = lines.length - 1;
-            while (lastLine >= 0 && !lines[lastLine].trim())
-                lastLine--;
-            if (lastLine >= 0 && isLegacyFormatBFooterLine(lines[lastLine])) {
-                lines.splice(lastLine, 1);
-                visibleParts[lastIndex] = lines.join('\n');
-                visibleParts = visibleParts.filter(part => !!part.trim());
-            }
-        }
-        const cleaned = visibleParts.join('\n');
-        return cleaned || '[卡片]';
-    }
-    catch {
-        return '[卡片]';
-    }
-}
-// ─── Complete card resolution: union of both Lark representations ──────────
-/** Sentinel key carrying pre-merged card text through extractCardContent. */
-const RESOLVED_TEXT_KEY = '__botmux_card_text__';
-/**
- * Marker for sub-cards Lark renders only client-side (collapsible panels,
- * lazy "展开" sections). Neither `im.message.get` representation returns their
- * body — Format A shows the upgrade fallback, Format B omits them — so we
- * surface an honest placeholder instead of a misleading blank or raw fallback.
- */
-export const CARD_EMBEDDED_PLACEHOLDER = '[卡片内嵌组件，需在飞书客户端展开查看]';
-/** Wrap merged text so extractCardContent returns it verbatim downstream. */
-export function wrapResolvedCardText(text) {
-    return JSON.stringify({ [RESOLVED_TEXT_KEY]: text });
-}
-/** Strip inline markup (font/text_tag tags, bold markers) but keep text,
- *  brackets and links so labels and values stay readable. */
-function stripInlineMarkup(s) {
-    return s.replace(/<\/?[a-z_]+[^>]*>/gi, '').replace(/\*\*/g, '');
-}
-/** Normalize for content-presence checks: drop whitespace, punctuation, inline
- *  markup and link URLs so the same content rendered by the two API formats
- *  compares equal regardless of cosmetic differences. */
-function normalizeForDedup(s) {
-    return stripInlineMarkup(s)
-        .replace(/\((https?:[^)]+)\)/g, '') // (https://…) link targets
-        .replace(/\s+/g, '')
-        .replace(/[，,：:。.\[\]()（）|/、]/g, '');
-}
-/** Find the value Format A rendered for a `label:` field — the text after the
- *  first occurrence of the label up to the line end. Returns '' if not found. */
-function findLabelValue(strippedA, label) {
-    const idx = strippedA.indexOf(label);
-    if (idx < 0)
-        return '';
-    const rest = strippedA.slice(idx + label.length);
-    return rest.split('\n')[0].trim();
-}
-/**
- * Merge the two Lark card renderings into one complete text. Format B (full
- * structured) is the base — it preserves links, sub-card bodies and select
- * options. From Format A (server-rendered) we recover only the field VALUES B
- * left blank (e.g. 值班人 names) via targeted label-fill: a B label line whose
- * value is empty is filled from A only when A's value isn't already present in
- * B (so fields B already renders aren't duplicated). Sub-cards Lark serves
- * client-side only — which A shows as upgrade holes and B omits — get one
- * honest placeholder rather than a silent drop or raw "请升级" text.
- */
-export function mergeCardText(textA, textB) {
-    const markHoles = (t) => t.split('\n').map(l => isPureCardUpgradeFallback(l) ? CARD_EMBEDDED_PLACEHOLDER : l).join('\n');
-    const a = (textA || '').trim();
-    const b = (textB || '').trim();
-    if (!b || isPureCardUpgradeFallback(b))
-        return markHoles(a) || a;
-    if (!a || isPureCardUpgradeFallback(a))
-        return markHoles(b) || b;
-    // Drop A's holes from the base, count them so we can flag client-only content.
-    const aHoleCount = a.split('\n').filter(isPureCardUpgradeFallback).length;
-    const strippedA = stripInlineMarkup(a);
-    const baseLines = b.split('\n').filter(l => !isPureCardUpgradeFallback(l));
-    const bAll = baseLines.map(normalizeForDedup).join('');
-    const filled = baseLines.map(line => {
-        const sm = stripInlineMarkup(line).trimEnd();
-        // empty-value field label, e.g. "值班人:" — but not bracketed section
-        // headers like "[ 检测结果 ]:" whose values live on following lines.
-        const m = sm.match(/^([^[\]\n]{1,16}?[:：])$/);
-        if (!m)
-            return line;
-        const value = findLabelValue(strippedA, m[1]);
-        if (!value)
-            return line;
-        // Only fill when B genuinely lacks this value (avoids duplicating fields
-        // whose values B already renders on adjacent lines, e.g. Tags/检测结果).
-        if (normalizeForDedup(value) && bAll.includes(normalizeForDedup(value)))
-            return line;
-        return `${line.replace(/\s*$/, '')} ${value}`;
-    });
-    // One honest marker if A carried sub-cards Lark only renders client-side.
-    if (aHoleCount > 0)
-        filled.push(CARD_EMBEDDED_PLACEHOLDER);
-    return filled.join('\n');
-}
-/**
- * Resolve a card to its most complete text by unioning both `im.message.get`
- * representations (server-rendered Format A + full structured Format B). Used
- * by BOTH the live event path and `botmux history` so a single message_id
- * resolves identically everywhere. Returns null when neither representation
- * could be fetched (caller keeps whatever it already had).
- */
-export async function resolveMergedCardContent(larkAppId, messageId, numberer) {
-    const [aRes, bRes] = await Promise.all([
-        getMessageDetail(larkAppId, messageId, { userCardContent: false }).catch(() => null),
-        getMessageDetail(larkAppId, messageId, { userCardContent: true }).catch(() => null),
-    ]);
-    const aContent = aRes?.items?.[0]?.body?.content;
-    const bContent = bRes?.items?.[0]?.body?.content;
-    if (!aContent && !bContent)
-        return null;
-    const structuredContent = (bContent ?? aContent);
-    // Resources BEFORE text so the shared numberer assigns [图片 N] in attachment
-    // order and the text extraction below reuses those numbers (same ordering
-    // contract as parseEventMessage / buildForwardedTree). Callers that don't
-    // pass a numberer are unaffected (fresh internal one, unnumbered text).
-    const resources = extractResources('interactive', structuredContent, numberer);
-    const textA = aContent ? extractCardContent(normalizeApiMessageContent('interactive', aContent), numberer) : '';
-    const textB = bContent ? extractCardContent(normalizeApiMessageContent('interactive', bContent), numberer) : '';
-    const merged = mergeCardText(textA, textB);
-    if (!merged)
-        return null;
-    // Carry the structured card JSON (B preferred) alongside the merged text so
-    // resource extraction (image_key/file_key) keeps working — extractResources
-    // walks elements/body, extractCardContent short-circuits on the text key.
-    return { text: merged, structuredContent, resources };
-}
-/**
- * Resolve an interactive event's card to its complete merged text in place.
- * Stores a sentinel that carries BOTH the merged text (for extractCardContent)
- * and the structured card JSON (for extractResources image/file extraction).
- * Falls back to local user_dsl unwrap if the REST merge yields nothing. Shared
- * by the live daemon path so forwarded cards reach the model fully parsed.
- */
-export async function resolveEventCard(data, larkAppId) {
-    let resolved = null;
-    try {
-        resolved = await resolveMergedCardContent(larkAppId, data.message.message_id);
-    }
-    catch { /* fall through to local unwrap */ }
-    if (resolved) {
-        const structured = (() => { try {
-            return JSON.parse(resolved.structuredContent);
-        }
-        catch {
-            return {};
-        } })();
-        data.message.content = JSON.stringify({ ...structured, [RESOLVED_TEXT_KEY]: resolved.text });
-        return;
-    }
-    unwrapUserDsl(data);
-}
-/** First non-empty string among the candidates. Deliberately NOT `??` — Lark
- *  payloads carry empty-string placeholders (e.g. multi_url: {url: '', pc_url:
- *  'https://…'}) which nullish coalescing would return, swallowing the valid
- *  platform URL behind it. */
-function firstNonEmptyString(...candidates) {
-    for (const c of candidates) {
-        if (typeof c === 'string' && c)
-            return c;
-    }
-    return undefined;
-}
-/** botmux's own card control buttons (🔊 语音总结 / 关闭会话 / 重启 / 配置 …) are
- *  pure callbacks back into the daemon: when another bot reads the card
- *  (history / cross-bot relay / quote), rendering them as `[🔊 语音总结]`
- *  pollutes the receiving prompt with affordances it can never click, so
- *  drop them structurally — same rationale as the botmux grey-footer strip
- *  in extractElementText.
- *
- *  Primary detection is the egress-stamped ownership marker
- *  (callback-button-marker.ts). This wordlist is the LEGACY fallback for
- *  cards sent by pre-marker builds (chat history is long-lived), covering
- *  every action dispatched by card-handler as of the fix — it is frozen;
- *  new botmux buttons rely on the marker, not on growing this list.
- *  Third-party cards' buttons (Argos 确认/创建群组…) stay: they carry no
- *  marker and their `value` never uses botmux's action vocabulary. */
-const BOTMUX_INTERNAL_CARD_ACTIONS = new Set([
-    // session / reply card controls (card-handler's actionType dispatch)
-    'close', 'disconnect', 'export_text', 'get_write_link', 'open_local_cli',
-    'open_local_terminal', 'refresh_screenshot', 'restart', 'resume',
-    'retry_last_task', 'takeover', 'term_action', 'toggle_display',
-    'toggle_stream', 'tui_keys', 'tui_text_input', 'voice_summary',
-    // pendingRepo gates
-    'repo_manual_submit', 'repo_worktree_submit', 'skip_repo',
-    'worktree_toggle_mode',
-    // config / grant / relay cards
-    'config_toggle', 'config_set', 'config_quota', 'config_text_open',
-    'config_text_save', 'grant_chat', 'grant_global', 'grant_deny',
-    'relay_search', 'relay_page', 'relay_select', 'relay_confirm',
-    // host-overload alert + codex notifier cards
-    'overload_noop', 'overload_clean_stopped', 'overload_suspend_idle',
-    'codex_notifier_continue', 'codex_notifier_open_app',
-]);
-/** True when the button is one of botmux's own callback buttons. Primary
- *  signal: the egress-stamped ownership marker (see
- *  callback-button-marker.ts) — future-proof against any new botmux action
- *  name. Legacy fallback: cards sent by pre-marker builds are still
- *  recognized via the internal action wordlist, in EITHER of the two real
- *  payload shapes — legacy top-level `value.action` (session/config cards)
- *  or v2 `behaviors:[{type:'callback', value.action}]` (reply-card voice
- *  button inside column_set). A button that somehow carries BOTH a botmux
- *  signal AND an open_url jump target is kept, since a real link is always
- *  worth surfacing. */
-function isBotmuxCallbackButton(el) {
-    if (buttonOpenUrl(el))
-        return false;
-    if (hasBotmuxCallbackMarker(el))
-        return true;
-    let action = el?.value?.action;
-    if (typeof action !== 'string' && Array.isArray(el?.behaviors)) {
-        action = el.behaviors.find((b) => b?.type === 'callback')?.value?.action;
-    }
-    return typeof action === 'string' && BOTMUX_INTERNAL_CARD_ACTIONS.has(action);
-}
-/** Resolve a card button's jump target across schema generations: v1 `url` /
- *  `multi_url`, v2 `behaviors: [{type:'open_url', default_url, pc_url, …}]`.
- *  Returns undefined for callback-only buttons (they have no followable URL). */
-function buttonOpenUrl(el) {
-    const direct = firstNonEmptyString(el.url, el.multi_url?.url, el.multi_url?.pc_url, el.multi_url?.android_url, el.multi_url?.ios_url);
-    if (direct)
-        return direct;
-    if (Array.isArray(el.behaviors)) {
-        for (const b of el.behaviors) {
-            if (b?.type !== 'open_url')
-                continue;
-            const u = firstNonEmptyString(b.default_url, b.pc_url, b.url, b.android_url, b.ios_url);
-            if (u)
-                return u;
-        }
-    }
-    return undefined;
-}
-/** Fold an image's alt text into its placeholder: `[图片 2]` + alt →
- *  `[图片 2: 报警前30分钟今(红)昨(蓝)同比]`. Consumers that pattern-match the
- *  placeholder tolerate the suffix (they match `[图片…]` loosely). */
-function withImgAlt(label, alt) {
-    return label.endsWith(']') ? `${label.slice(0, -1)}: ${alt}]` : `${label} (${alt})`;
-}
-/** Recursively extract image resources from an original-format card element. */
-function extractElementImages(el, resources, pushIfNew) {
-    if (!el || typeof el !== 'object')
-        return;
-    const tag = el.tag;
-    const key = el.image_key ?? el.img_key;
-    if ((tag === 'img' || tag === 'image') && key) {
-        pushIfNew(resources, { type: 'image', key, name: `${key}.jpg` });
-    }
-    // div.extra can contain an image
-    if (el.extra)
-        extractElementImages(el.extra, resources, pushIfNew);
-    // column_set / column — recurse into nested elements
-    if (Array.isArray(el.columns)) {
-        for (const col of el.columns) {
-            if (Array.isArray(col.elements)) {
-                for (const child of col.elements)
-                    extractElementImages(child, resources, pushIfNew);
-            }
-        }
-    }
-    if (Array.isArray(el.elements)) {
-        for (const child of el.elements)
-            extractElementImages(child, resources, pushIfNew);
-    }
-}
-/** Recursively extract readable text from an original-format card element. */
-function extractElementText(el, parts, imgLabel) {
-    if (!el || typeof el !== 'object')
-        return;
-    const tag = el.tag;
-    // The public element id alone is not ownership proof: third-party cards may
-    // collide with it. New botmux footers carry the id plus the exact reserved
-    // marker; unexpected large text stays visible rather than being stripped.
-    const elementText = el.text?.content ?? el.content;
-    if (el.element_id === REPLY_CARD_FOOTER_ELEMENT_ID
-        && tag === 'markdown'
-        && (el.text_size === undefined || el.text_size === 'notation_small_v2')
-        && typeof elementText === 'string'
-        && hasExactMarkerMarkdown(elementText)) {
-        return;
-    }
-    // div / markdown / plain_text blocks
-    if (tag === 'div' || tag === 'markdown' || tag === 'plain_text') {
-        const text = el.text?.content ?? el.content;
-        if (text)
-            parts.push(text);
-    }
-    // div.fields[] — v2 cards put most body text in a fields array of lark_md
-    // cells (规则/报警时间/Tags/检测结果/详情链接…), NOT in el.text. Without this
-    // the entire detail section of field-based cards (e.g. Argos alarm cards)
-    // silently vanishes. A div can carry both el.text and el.fields, so this is
-    // additive rather than an else-branch.
-    if (Array.isArray(el.fields)) {
-        for (const f of el.fields) {
-            const t = f?.text?.content ?? f?.content;
-            if (t)
-                parts.push(t);
-        }
-    }
-    // button — text may be a plain_text object (v2) or a string (v1 simplified).
-    // Jump buttons carry their target as v1 `url`/`multi_url` or v2 `behaviors`
-    // open_url; keep it so the reader can actually follow the link (e.g. Argos
-    // [分析报告] → the report URL). Callback buttons have no URL and stay bare.
-    // botmux's own callback buttons (🔊 语音总结 …) are dropped entirely.
-    if (tag === 'button') {
-        if (isBotmuxCallbackButton(el))
-            return;
-        const btnText = typeof el.text === 'string' ? el.text : el.text?.content;
-        if (btnText) {
-            const url = buttonOpenUrl(el);
-            parts.push(url ? `[${btnText}](${url})` : `[${btnText}]`);
-        }
-    }
-    // input — surface the placeholder so the reader knows the field is there.
-    if (tag === 'input') {
-        const ph = el.placeholder?.content;
-        if (ph)
-            parts.push(`[输入框: ${ph}]`);
-    }
-    // select / multi-select / overflow — surface placeholder + selectable
-    // options so the choices the card offers aren't lost.
-    if (tag === 'select_static' || tag === 'multi_select_static' || tag === 'overflow') {
-        const ph = el.placeholder?.content;
-        const opts = Array.isArray(el.options)
-            ? el.options.map((o) => o.text?.content).filter(Boolean)
-            : [];
-        const head = ph ? `[下拉: ${ph}` : '[下拉';
-        parts.push(opts.length ? `${head} | 选项: ${opts.join(' / ')}]` : `${head}]`);
-    }
-    // image — emit a numbered placeholder matching the attachment list order.
-    // Carry the alt text when present: for chart images (Argos 同比曲线图) the
-    // alt is often the only machine-readable description of what the image shows.
-    if (tag === 'img' || tag === 'image') {
-        const k = el.image_key ?? el.img_key;
-        const alt = (typeof el.alt === 'string' ? el.alt : el.alt?.content)?.trim();
-        if (k)
-            parts.push(alt ? withImgAlt(imgLabel(k), alt) : imgLabel(k));
-    }
-    // note blocks (v1 only — v2 removed the tag but we still parse v1 cards)
-    if (tag === 'note' && Array.isArray(el.elements)) {
-        const noteTexts = el.elements
-            .map((n) => n.content ?? n.text?.content ?? '')
-            .filter(Boolean);
-        if (noteTexts.length)
-            parts.push(noteTexts.join(' '));
-    }
-    // div.extra can host an image
-    if (el.extra)
-        extractElementText(el.extra, parts, imgLabel);
-    // column_set / column — recurse into nested elements
-    if (Array.isArray(el.columns)) {
-        for (const col of el.columns) {
-            if (Array.isArray(col.elements)) {
-                for (const child of col.elements)
-                    extractElementText(child, parts, imgLabel);
-            }
-        }
-    }
-    // action blocks hold their children in `actions`, not `elements` — recurse so
-    // their buttons / inputs / selects (确认/创建群组/驾驶舱…) get surfaced.
-    if (Array.isArray(el.actions)) {
-        for (const child of el.actions)
-            extractElementText(child, parts, imgLabel);
-    }
-    if (Array.isArray(el.elements) && tag !== 'note') {
-        for (const child of el.elements)
-            extractElementText(child, parts, imgLabel);
-    }
-}
-//# sourceMappingURL=message-parser.js.map