picoclaw/pkg/channels/feishu/feishu_64.go

1267 lines
36 KiB
Go
Raw Normal View History

//go:build amd64 || arm64 || riscv64 || mips64 || ppc64
package feishu
import (
"context"
"encoding/json"
"fmt"
"io"
"math/rand"
"net/http"
"os"
"path/filepath"
"strings"
"sync"
"sync/atomic"
"time"
lark "github.com/larksuite/oapi-sdk-go/v3"
larkcore "github.com/larksuite/oapi-sdk-go/v3/core"
larkdispatcher "github.com/larksuite/oapi-sdk-go/v3/event/dispatcher"
larkim "github.com/larksuite/oapi-sdk-go/v3/service/im/v1"
larkws "github.com/larksuite/oapi-sdk-go/v3/ws"
"github.com/sipeed/picoclaw/pkg/bus"
"github.com/sipeed/picoclaw/pkg/channels"
"github.com/sipeed/picoclaw/pkg/config"
"github.com/sipeed/picoclaw/pkg/identity"
"github.com/sipeed/picoclaw/pkg/logger"
"github.com/sipeed/picoclaw/pkg/media"
"github.com/sipeed/picoclaw/pkg/utils"
)
// errCodeTenantTokenInvalid is the Feishu API error code for an expired/revoked
// tenant_access_token. The Lark SDK's built-in retry does not clear its cache
// on this error, so we do it ourselves.
const errCodeTenantTokenInvalid = 99991663
type FeishuChannel struct {
*channels.BaseChannel
bc *config.Channel
config *config.FeishuSettings
client *lark.Client
wsClient *larkws.Client
tokenCache *tokenCache // custom cache that supports invalidation
botOpenID atomic.Value // stores string; populated lazily for @mention detection
messageCache sync.Map // caches fetched messages (messageID -> *larkim.Message)
mu sync.Mutex
cancel context.CancelFunc
progress *channels.ToolFeedbackAnimator
deleteMessageFn func(context.Context, string, string) error
sendMediaPartFn func(context.Context, string, bus.MediaPart, media.MediaStore) error
sendTextFn func(context.Context, string, string) (string, error)
}
type cachedMessage struct {
msg *larkim.Message
expiry time.Time
}
func NewFeishuChannel(bc *config.Channel, cfg *config.FeishuSettings, bus *bus.MessageBus) (*FeishuChannel, error) {
base := channels.NewBaseChannel("feishu", cfg, bus, bc.AllowFrom,
channels.WithGroupTrigger(bc.GroupTrigger),
channels.WithReasoningChannelID(bc.ReasoningChannelID),
)
tc := newTokenCache()
opts := []lark.ClientOptionFunc{lark.WithTokenCache(tc)}
if cfg.IsLark {
opts = append(opts, lark.WithOpenBaseUrl(lark.LarkBaseUrl))
}
ch := &FeishuChannel{
BaseChannel: base,
bc: bc,
config: cfg,
tokenCache: tc,
client: lark.NewClient(cfg.AppID, cfg.AppSecret.String(), opts...),
}
ch.deleteMessageFn = ch.deleteMessageAPI
ch.sendMediaPartFn = ch.sendMediaPart
ch.sendTextFn = ch.sendText
ch.progress = channels.NewToolFeedbackAnimator(ch.EditMessage)
ch.SetOwner(ch)
return ch, nil
}
func (c *FeishuChannel) Start(ctx context.Context) error {
if c.config.AppID == "" || c.config.AppSecret.String() == "" {
return fmt.Errorf("feishu app_id or app_secret is empty")
}
// Fetch bot open_id via API for reliable @mention detection.
if err := c.fetchBotOpenID(ctx); err != nil {
logger.ErrorCF("feishu", "Failed to fetch bot open_id, @mention detection may not work", map[string]any{
"error": err.Error(),
})
}
dispatcher := larkdispatcher.NewEventDispatcher(c.config.VerificationToken.String(), c.config.EncryptKey.String()).
OnP2MessageReceiveV1(c.handleMessageReceive)
runCtx, cancel := context.WithCancel(ctx)
c.mu.Lock()
c.cancel = cancel
domain := lark.FeishuBaseUrl
if c.config.IsLark {
domain = lark.LarkBaseUrl
}
c.wsClient = larkws.NewClient(
c.config.AppID,
c.config.AppSecret.String(),
larkws.WithEventHandler(dispatcher),
larkws.WithDomain(domain),
)
wsClient := c.wsClient
c.mu.Unlock()
c.SetRunning(true)
logger.InfoC("feishu", "Feishu channel started (websocket mode)")
go func() {
if err := wsClient.Start(runCtx); err != nil {
2026-02-21 08:35:56 +00:00
logger.ErrorCF("feishu", "Feishu websocket stopped with error", map[string]any{
"error": err.Error(),
})
}
}()
return nil
}
func (c *FeishuChannel) Stop(ctx context.Context) error {
c.mu.Lock()
if c.cancel != nil {
c.cancel()
c.cancel = nil
}
c.wsClient = nil
c.mu.Unlock()
if c.progress != nil {
c.progress.StopAll()
}
c.SetRunning(false)
logger.InfoC("feishu", "Feishu channel stopped")
return nil
}
// Send sends a message using Interactive Card format for markdown rendering.
// Falls back to plain text message if card sending fails (e.g., table limit exceeded).
func (c *FeishuChannel) Send(ctx context.Context, msg bus.OutboundMessage) ([]string, error) {
if !c.IsRunning() {
return nil, channels.ErrNotRunning
}
if msg.ChatID == "" {
return nil, fmt.Errorf("chat ID is empty: %w", channels.ErrSendFailed)
}
isToolFeedback := outboundMessageIsToolFeedback(msg)
if isToolFeedback {
if msgID, handled, err := c.progress.Update(ctx, msg.ChatID, msg.Content); handled {
if err != nil {
// Feishu can fall back to plain text for a previous progress
// message, and those messages cannot be patched through the card
// edit API. Drop the stale tracker and recreate the progress
// message so later tool feedback is not blocked.
c.resetTrackedToolFeedbackAfterEditFailure(ctx, msg.ChatID)
} else {
return []string{msgID}, nil
}
}
} else {
if msgIDs, handled := c.FinalizeToolFeedbackMessage(ctx, msg); handled {
return msgIDs, nil
}
}
trackedMsgID, hasTrackedMsg := c.currentToolFeedbackMessage(msg.ChatID)
// Build interactive card with markdown content
sendContent := msg.Content
if isToolFeedback {
sendContent = channels.InitialAnimatedToolFeedbackContent(msg.Content)
}
cardContent, err := buildMarkdownCard(sendContent)
if err != nil {
// If card build fails, fall back to plain text
msgID, sendErr := c.sendText(ctx, msg.ChatID, sendContent)
if sendErr != nil {
return nil, sendErr
}
if isToolFeedback {
c.RecordToolFeedbackMessage(msg.ChatID, msgID, msg.Content)
} else if hasTrackedMsg {
c.dismissTrackedToolFeedbackMessage(ctx, msg.ChatID, trackedMsgID)
}
return []string{msgID}, nil
}
// First attempt: try sending as interactive card
msgID, err := c.sendCard(ctx, msg.ChatID, cardContent)
if err == nil {
if isToolFeedback {
c.RecordToolFeedbackMessage(msg.ChatID, msgID, msg.Content)
} else if hasTrackedMsg {
c.dismissTrackedToolFeedbackMessage(ctx, msg.ChatID, trackedMsgID)
}
return []string{msgID}, nil
}
// Check if error is due to card table limit (error code 11310)
// See: https://open.feishu.cn/document/server-docs/im-api/message-content-description/create_json
errMsg := err.Error()
isCardLimitError := strings.Contains(errMsg, "11310")
if isCardLimitError {
logger.WarnCF("feishu", "Card send failed (table limit), falling back to text message", map[string]any{
"chat_id": msg.ChatID,
"error": errMsg,
})
// Second attempt: fall back to plain text message
msgID, textErr := c.sendText(ctx, msg.ChatID, sendContent)
if textErr == nil {
if isToolFeedback {
c.RecordToolFeedbackMessage(msg.ChatID, msgID, msg.Content)
} else if hasTrackedMsg {
c.dismissTrackedToolFeedbackMessage(ctx, msg.ChatID, trackedMsgID)
}
return []string{msgID}, nil
}
// If text also fails, return the text error
return nil, textErr
}
// For other errors, return the original card error
return nil, err
}
// EditMessage implements channels.MessageEditor.
// Uses Message.Patch to update an interactive card message.
func (c *FeishuChannel) EditMessage(ctx context.Context, chatID, messageID, content string) error {
cardContent, err := buildMarkdownCard(content)
if err != nil {
return fmt.Errorf("feishu edit: card build failed: %w", err)
}
req := larkim.NewPatchMessageReqBuilder().
MessageId(messageID).
Body(larkim.NewPatchMessageReqBodyBuilder().Content(cardContent).Build()).
Build()
resp, err := c.client.Im.V1.Message.Patch(ctx, req)
if err != nil {
return fmt.Errorf("feishu edit: %w", err)
}
if !resp.Success() {
c.invalidateTokenOnAuthError(resp.Code)
return fmt.Errorf("feishu edit api error (code=%d msg=%s)", resp.Code, resp.Msg)
}
return nil
}
// DeleteMessage implements channels.MessageDeleter.
func (c *FeishuChannel) DeleteMessage(ctx context.Context, chatID, messageID string) error {
deleteFn := c.deleteMessageFn
if deleteFn == nil {
deleteFn = c.deleteMessageAPI
}
return deleteFn(ctx, chatID, messageID)
}
func (c *FeishuChannel) deleteMessageAPI(ctx context.Context, chatID, messageID string) error {
req := larkim.NewDeleteMessageReqBuilder().
MessageId(messageID).
Build()
resp, err := c.client.Im.V1.Message.Delete(ctx, req)
if err != nil {
return fmt.Errorf("feishu delete: %w", err)
}
if !resp.Success() {
c.invalidateTokenOnAuthError(resp.Code)
return fmt.Errorf("feishu delete api error (code=%d msg=%s)", resp.Code, resp.Msg)
}
return nil
}
// SendPlaceholder implements channels.PlaceholderCapable.
// Sends an interactive card with placeholder text and returns its message ID.
func (c *FeishuChannel) SendPlaceholder(ctx context.Context, chatID string) (string, error) {
if !c.bc.Placeholder.Enabled {
logger.DebugCF("feishu", "Placeholder disabled, skipping", map[string]any{
"chat_id": chatID,
})
return "", nil
}
text := c.bc.Placeholder.GetRandomText()
cardContent, err := buildMarkdownCard(text)
if err != nil {
return "", fmt.Errorf("feishu placeholder: card build failed: %w", err)
}
req := larkim.NewCreateMessageReqBuilder().
ReceiveIdType(larkim.CreateMessageV1ReceiveIDTypeChatId).
Body(larkim.NewCreateMessageReqBodyBuilder().
ReceiveId(chatID).
MsgType(larkim.MsgTypeInteractive).
Content(cardContent).
Build()).
Build()
resp, err := c.client.Im.V1.Message.Create(ctx, req)
if err != nil {
return "", fmt.Errorf("feishu placeholder send: %w", err)
}
if !resp.Success() {
c.invalidateTokenOnAuthError(resp.Code)
return "", fmt.Errorf("feishu placeholder api error (code=%d msg=%s)", resp.Code, resp.Msg)
}
if resp.Data != nil && resp.Data.MessageId != nil {
return *resp.Data.MessageId, nil
}
return "", nil
}
func outboundMessageIsToolFeedback(msg bus.OutboundMessage) bool {
if len(msg.Context.Raw) == 0 {
return false
}
return strings.EqualFold(strings.TrimSpace(msg.Context.Raw["message_kind"]), "tool_feedback")
}
func (c *FeishuChannel) currentToolFeedbackMessage(chatID string) (string, bool) {
if c.progress == nil {
return "", false
}
return c.progress.Current(chatID)
}
func (c *FeishuChannel) takeToolFeedbackMessage(chatID string) (string, string, bool) {
if c.progress == nil {
return "", "", false
}
return c.progress.Take(chatID)
}
func (c *FeishuChannel) RecordToolFeedbackMessage(chatID, messageID, content string) {
if c.progress == nil {
return
}
c.progress.Record(chatID, messageID, content)
}
func (c *FeishuChannel) ClearToolFeedbackMessage(chatID string) {
if c.progress == nil {
return
}
c.progress.Clear(chatID)
}
func (c *FeishuChannel) DismissToolFeedbackMessage(ctx context.Context, chatID string) {
msgID, ok := c.currentToolFeedbackMessage(chatID)
if !ok {
return
}
c.dismissTrackedToolFeedbackMessage(ctx, chatID, msgID)
}
func (c *FeishuChannel) resetTrackedToolFeedbackAfterEditFailure(ctx context.Context, chatID string) {
msgID, ok := c.currentToolFeedbackMessage(chatID)
if !ok {
return
}
c.dismissTrackedToolFeedbackMessage(ctx, chatID, msgID)
}
func (c *FeishuChannel) dismissTrackedToolFeedbackMessage(ctx context.Context, chatID, messageID string) {
if strings.TrimSpace(chatID) == "" || strings.TrimSpace(messageID) == "" {
return
}
c.ClearToolFeedbackMessage(chatID)
deleteFn := c.deleteMessageFn
if deleteFn == nil {
deleteFn = c.deleteMessageAPI
}
_ = deleteFn(ctx, chatID, messageID)
}
func (c *FeishuChannel) finalizeTrackedToolFeedbackMessage(
ctx context.Context,
chatID string,
content string,
editFn func(context.Context, string, string, string) error,
) ([]string, bool) {
msgID, baseContent, ok := c.takeToolFeedbackMessage(chatID)
if !ok || editFn == nil {
return nil, false
}
if err := editFn(ctx, chatID, msgID, content); err != nil {
c.RecordToolFeedbackMessage(chatID, msgID, baseContent)
return nil, false
}
return []string{msgID}, true
}
func (c *FeishuChannel) FinalizeToolFeedbackMessage(ctx context.Context, msg bus.OutboundMessage) ([]string, bool) {
if outboundMessageIsToolFeedback(msg) {
return nil, false
}
return c.finalizeTrackedToolFeedbackMessage(ctx, msg.ChatID, msg.Content, c.EditMessage)
}
// ReactToMessage implements channels.ReactionCapable.
// Adds a reaction (randomly chosen from config) and returns an undo function to remove it.
func (c *FeishuChannel) ReactToMessage(ctx context.Context, chatID, messageID string) (func(), error) {
// Get emoji list from config (Feishu emoji_type keys, e.g. Pin, THUMBSUP).
// Ignore empty entries so a list like ["", "Pin"] does not randomly pick "" (API 231001).
var candidates []string
for _, e := range c.config.RandomReactionEmoji {
e = strings.TrimSpace(e)
if e != "" {
candidates = append(candidates, e)
}
}
chosenEmoji := "Pin"
if len(candidates) > 0 {
chosenEmoji = candidates[rand.Intn(len(candidates))]
}
req := larkim.NewCreateMessageReactionReqBuilder().
MessageId(messageID).
Body(larkim.NewCreateMessageReactionReqBodyBuilder().
ReactionType(larkim.NewEmojiBuilder().EmojiType(chosenEmoji).Build()).
Build()).
Build()
resp, err := c.client.Im.V1.MessageReaction.Create(ctx, req)
if err != nil {
logger.ErrorCF("feishu", "Failed to add reaction", map[string]any{
"emoji": chosenEmoji,
"message_id": messageID,
"error": err.Error(),
})
return func() {}, fmt.Errorf("feishu react: %w", err)
}
if !resp.Success() {
c.invalidateTokenOnAuthError(resp.Code)
logger.ErrorCF("feishu", "Reaction API error", map[string]any{
"emoji": chosenEmoji,
"message_id": messageID,
"code": resp.Code,
"msg": resp.Msg,
})
return func() {}, fmt.Errorf("feishu react api error (code=%d msg=%s)", resp.Code, resp.Msg)
}
var reactionID string
if resp.Data != nil && resp.Data.ReactionId != nil {
reactionID = *resp.Data.ReactionId
}
if reactionID == "" {
return func() {}, nil
}
var undone atomic.Bool
undo := func() {
if !undone.CompareAndSwap(false, true) {
return
}
delReq := larkim.NewDeleteMessageReactionReqBuilder().
MessageId(messageID).
ReactionId(reactionID).
Build()
_, _ = c.client.Im.V1.MessageReaction.Delete(context.Background(), delReq)
}
return undo, nil
}
// SendMedia implements channels.MediaSender.
// Uploads images/files via Feishu API then sends as messages.
func (c *FeishuChannel) SendMedia(ctx context.Context, msg bus.OutboundMediaMessage) ([]string, error) {
if !c.IsRunning() {
return nil, channels.ErrNotRunning
}
trackedMsgID, hasTrackedMsg := c.currentToolFeedbackMessage(msg.ChatID)
if msg.ChatID == "" {
return nil, fmt.Errorf("chat ID is empty: %w", channels.ErrSendFailed)
}
store := c.GetMediaStore()
if store == nil {
return nil, fmt.Errorf("no media store available: %w", channels.ErrSendFailed)
}
caption := firstMediaCaption(msg.Parts)
sentAny := false
for _, part := range msg.Parts {
if err := c.sendMediaPartFn(ctx, msg.ChatID, part, store); err != nil {
return nil, err
}
sentAny = true
}
if sentAny && caption != "" {
if _, err := c.sendTextFn(ctx, msg.ChatID, caption); err != nil {
return nil, err
}
}
if hasTrackedMsg {
c.dismissTrackedToolFeedbackMessage(ctx, msg.ChatID, trackedMsgID)
}
return nil, nil
}
// sendMediaPart resolves and sends a single media part.
func (c *FeishuChannel) sendMediaPart(
ctx context.Context,
chatID string,
part bus.MediaPart,
store media.MediaStore,
) error {
localPath, err := store.Resolve(part.Ref)
if err != nil {
logger.ErrorCF("feishu", "Failed to resolve media ref", map[string]any{
"ref": part.Ref,
"error": err.Error(),
})
return nil // skip this part
}
file, err := os.Open(localPath)
if err != nil {
logger.ErrorCF("feishu", "Failed to open media file", map[string]any{
"path": localPath,
"error": err.Error(),
})
return nil // skip this part
}
defer file.Close()
switch part.Type {
case "image":
err = c.sendImage(ctx, chatID, file)
default:
filename := part.Filename
if filename == "" {
filename = "file"
}
err = c.sendFile(ctx, chatID, file, filename, part.Type)
}
if err != nil {
logger.ErrorCF("feishu", "Failed to send media", map[string]any{
"type": part.Type,
"error": err.Error(),
})
return fmt.Errorf("feishu send media: %w", channels.ErrTemporary)
}
return nil
}
func firstMediaCaption(parts []bus.MediaPart) string {
for _, part := range parts {
if caption := strings.TrimSpace(part.Caption); caption != "" {
return caption
}
}
return ""
}
// --- Inbound message handling ---
func (c *FeishuChannel) handleMessageReceive(ctx context.Context, event *larkim.P2MessageReceiveV1) error {
if event == nil || event.Event == nil || event.Event.Message == nil {
return nil
}
message := event.Event.Message
sender := event.Event.Sender
chatID := stringValue(message.ChatId)
if chatID == "" {
return nil
}
senderID := extractFeishuSenderID(sender)
if senderID == "" {
senderID = "unknown"
}
messageType := stringValue(message.MessageType)
messageID := stringValue(message.MessageId)
rawContent := stringValue(message.Content)
// Check allowlist early to avoid downloading media for rejected senders.
// BaseChannel.HandleMessage will check again, but this avoids wasted network I/O.
senderInfo := bus.SenderInfo{
Platform: "feishu",
PlatformID: senderID,
CanonicalID: identity.BuildCanonicalID("feishu", senderID),
}
if !c.IsAllowedSender(senderInfo) {
return nil
}
// Extract content based on message type
content := extractContent(messageType, rawContent)
// Handle media messages (download and store)
var mediaRefs []string
if store := c.GetMediaStore(); store != nil && messageID != "" {
mediaRefs = c.downloadInboundMedia(ctx, chatID, messageID, messageType, rawContent, store)
}
Feat/feishu card parsing (#1534) * feat(feishu): add interactive card message parsing Add support for parsing inbound Feishu interactive card messages. When a user sends a card message, the text content is now extracted and passed to the LLM for processing. - Add extractCardText() to recursively extract text from card JSON - Support both JSON 1.0 (legacy) and JSON 2.0 schema formats - Handle nested elements: header, body, actions, columns - Extract text from markdown, lark_md, and plain_text elements - Add comprehensive unit tests for card parsing Fixes #<issue_number> 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * feat(feishu): extract and download images from interactive cards When receiving interactive card messages, extract embedded images (img_key, src, icon_key) and download them for LLM processing. - Add extractCardImageKeys() to recursively extract image keys from card JSON - Support img elements (img_key, src) and icon elements (icon_key) - Update downloadInboundMedia() to handle MsgTypeInteractive - Add comprehensive unit tests for image extraction Images are downloaded and stored via MediaStore, then appended to the message content as [image: photo] tags for LLM visibility. 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): simplify card parsing - pass raw JSON, only extract images Address review feedback: text extraction cannot exhaustively handle all card formats (i18n_elements, div.fields, etc.). Pass raw JSON to LLM instead - same approach as MsgTypePost. Only image extraction remains as images must be downloaded for LLM to process. - Remove extractCardText() and helper functions - extractContent() now returns raw JSON for MsgTypeInteractive - Keep extractCardImageKeys() for downloading embedded images - Update tests to expect raw JSON for interactive cards * fix(feishu): don't append media tags to interactive card JSON Appending media tags like "[attachment]" to raw JSON content produces invalid JSON format. For interactive cards, the JSON already contains image information and media refs are downloaded separately. - Skip appendMediaTags for MsgTypeInteractive to preserve valid JSON - Add test case for interactive card with images * fix(feishu): filter out external URLs from card image extraction Only Feishu-hosted image keys (img_xxx, icon_xxx) can be downloaded via the Feishu API. External URLs in src field (https://...) should be filtered out to avoid download failures. - Add isFeishuImageKey() to detect Feishu-hosted keys vs external URLs - Update extractImageKeysRecursive to skip external URLs in src field - Add tests for external URL filtering and mixed scenarios * feat(feishu): support downloading external images from interactive cards Previously only Feishu-hosted images (img_key, icon_key) could be downloaded. Now external URLs in src field are also downloaded via HTTP and made available to the LLM. - extractCardImageKeys now returns two slices: Feishu keys and external URLs - Add downloadExternalImage to download images from HTTP URLs - Update downloadInboundMedia to handle both Feishu API and HTTP downloads - Update tests for new function signature * fix(feishu): use HTTP client with timeout for external image downloads Replaced http.DefaultClient with a client that has a 30-second timeout to prevent hanging on unresponsive external URLs. Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): resolve lint errors for shadow and formatting - Rename err variables to avoid shadowing in downloadExternalImage - Fix struct field alignment in TestExtractCardImageKeys Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * refactor(feishu): pass external image URLs to LLM instead of downloading Instead of downloading external images from interactive cards, pass the URLs directly to LLM. This reduces network overhead and lets vision-capable models fetch images as needed. - Remove downloadExternalImage function - Append external URLs to card content for LLM processing - Only download Feishu-hosted images via API 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): add blank line between functions for gci formatting * fix(feishu): keep interactive card content as valid JSON
2026-03-20 04:59:43 +00:00
// For interactive cards, pass external image URLs via media refs.
// Keep content as valid raw JSON for downstream parsing.
if messageType == larkim.MsgTypeInteractive {
_, externalURLs := extractCardImageKeys(rawContent)
if len(externalURLs) > 0 {
mediaRefs = append(mediaRefs, externalURLs...)
}
}
// Append media tags to content (like Telegram does)
content = appendMediaTags(content, messageType, mediaRefs)
if content == "" {
content = "[empty message]"
}
chatType := stringValue(message.ChatType)
metadata := buildInboundMetadata(message, sender)
var (
inboundChatType string
isMentioned bool
)
if chatType == "p2p" {
inboundChatType = "direct"
} else {
inboundChatType = "group"
// Check if bot was mentioned
isMentioned = c.isBotMentioned(message)
// Strip mention placeholders from content before group trigger check
if len(message.Mentions) > 0 {
content = stripMentionPlaceholders(content, message.Mentions)
}
// In group chats, apply unified group trigger filtering
respond, cleaned := c.ShouldRespondInGroup(isMentioned, content)
if !respond {
return nil
}
content = cleaned
}
if replyTargetID(message) != "" || stringValue(message.ThreadId) != "" {
content, mediaRefs = c.prependReplyContext(ctx, message, chatID, content, mediaRefs)
}
if content == "" {
content = "[empty message]"
}
2026-02-21 08:35:56 +00:00
logger.InfoCF("feishu", "Feishu message received", map[string]any{
"sender_id": senderID,
"chat_id": chatID,
"message_id": messageID,
"preview": utils.Truncate(content, 80),
})
logger.InfoCF("feishu", "Feishu reply linkage", map[string]any{
"message_id": messageID,
"parent_id": stringValue(message.ParentId),
"root_id": stringValue(message.RootId),
"thread_id": stringValue(message.ThreadId),
})
inboundCtx := bus.InboundContext{
Channel: "feishu",
ChatID: chatID,
ChatType: inboundChatType,
SenderID: senderID,
MessageID: messageID,
Mentioned: isMentioned,
Raw: metadata,
}
if sender != nil && sender.TenantKey != nil && *sender.TenantKey != "" {
inboundCtx.SpaceType = "tenant"
inboundCtx.SpaceID = *sender.TenantKey
}
c.HandleInboundContext(ctx, chatID, content, mediaRefs, inboundCtx, senderInfo)
return nil
}
// --- Internal helpers ---
// fetchBotOpenID calls the Feishu bot info API to retrieve and store the bot's open_id.
func (c *FeishuChannel) fetchBotOpenID(ctx context.Context) error {
resp, err := c.client.Do(ctx, &larkcore.ApiReq{
HttpMethod: http.MethodGet,
ApiPath: "/open-apis/bot/v3/info",
SupportedAccessTokenTypes: []larkcore.AccessTokenType{larkcore.AccessTokenTypeTenant},
})
if err != nil {
return fmt.Errorf("bot info request: %w", err)
}
var result struct {
Code int `json:"code"`
Bot struct {
OpenID string `json:"open_id"`
} `json:"bot"`
}
if err := json.Unmarshal(resp.RawBody, &result); err != nil {
return fmt.Errorf("bot info parse: %w", err)
}
if result.Code != 0 {
c.invalidateTokenOnAuthError(result.Code)
return fmt.Errorf("bot info api error (code=%d)", result.Code)
}
if result.Bot.OpenID == "" {
return fmt.Errorf("bot info: empty open_id")
}
c.botOpenID.Store(result.Bot.OpenID)
logger.InfoCF("feishu", "Fetched bot open_id from API", map[string]any{
"open_id": result.Bot.OpenID,
})
return nil
}
// isBotMentioned checks if the bot was @mentioned in the message.
func (c *FeishuChannel) isBotMentioned(message *larkim.EventMessage) bool {
if message.Mentions == nil {
return false
}
knownID, ok := c.botOpenID.Load().(string)
if !ok || knownID == "" {
logger.DebugCF("feishu", "Bot open_id unknown, cannot detect @mention", nil)
return false
}
for _, m := range message.Mentions {
if m.Id == nil {
continue
}
if m.Id.OpenId != nil && *m.Id.OpenId == knownID {
return true
}
}
return false
}
// extractContent extracts text content from different message types.
func extractContent(messageType, rawContent string) string {
if rawContent == "" {
return ""
}
switch messageType {
case larkim.MsgTypeText:
var textPayload struct {
Text string `json:"text"`
}
if err := json.Unmarshal([]byte(rawContent), &textPayload); err == nil {
return textPayload.Text
}
return rawContent
case larkim.MsgTypePost:
// Pass raw JSON to LLM — structured rich text is more informative than flattened plain text
return rawContent
Feat/feishu card parsing (#1534) * feat(feishu): add interactive card message parsing Add support for parsing inbound Feishu interactive card messages. When a user sends a card message, the text content is now extracted and passed to the LLM for processing. - Add extractCardText() to recursively extract text from card JSON - Support both JSON 1.0 (legacy) and JSON 2.0 schema formats - Handle nested elements: header, body, actions, columns - Extract text from markdown, lark_md, and plain_text elements - Add comprehensive unit tests for card parsing Fixes #<issue_number> 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * feat(feishu): extract and download images from interactive cards When receiving interactive card messages, extract embedded images (img_key, src, icon_key) and download them for LLM processing. - Add extractCardImageKeys() to recursively extract image keys from card JSON - Support img elements (img_key, src) and icon elements (icon_key) - Update downloadInboundMedia() to handle MsgTypeInteractive - Add comprehensive unit tests for image extraction Images are downloaded and stored via MediaStore, then appended to the message content as [image: photo] tags for LLM visibility. 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): simplify card parsing - pass raw JSON, only extract images Address review feedback: text extraction cannot exhaustively handle all card formats (i18n_elements, div.fields, etc.). Pass raw JSON to LLM instead - same approach as MsgTypePost. Only image extraction remains as images must be downloaded for LLM to process. - Remove extractCardText() and helper functions - extractContent() now returns raw JSON for MsgTypeInteractive - Keep extractCardImageKeys() for downloading embedded images - Update tests to expect raw JSON for interactive cards * fix(feishu): don't append media tags to interactive card JSON Appending media tags like "[attachment]" to raw JSON content produces invalid JSON format. For interactive cards, the JSON already contains image information and media refs are downloaded separately. - Skip appendMediaTags for MsgTypeInteractive to preserve valid JSON - Add test case for interactive card with images * fix(feishu): filter out external URLs from card image extraction Only Feishu-hosted image keys (img_xxx, icon_xxx) can be downloaded via the Feishu API. External URLs in src field (https://...) should be filtered out to avoid download failures. - Add isFeishuImageKey() to detect Feishu-hosted keys vs external URLs - Update extractImageKeysRecursive to skip external URLs in src field - Add tests for external URL filtering and mixed scenarios * feat(feishu): support downloading external images from interactive cards Previously only Feishu-hosted images (img_key, icon_key) could be downloaded. Now external URLs in src field are also downloaded via HTTP and made available to the LLM. - extractCardImageKeys now returns two slices: Feishu keys and external URLs - Add downloadExternalImage to download images from HTTP URLs - Update downloadInboundMedia to handle both Feishu API and HTTP downloads - Update tests for new function signature * fix(feishu): use HTTP client with timeout for external image downloads Replaced http.DefaultClient with a client that has a 30-second timeout to prevent hanging on unresponsive external URLs. Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): resolve lint errors for shadow and formatting - Rename err variables to avoid shadowing in downloadExternalImage - Fix struct field alignment in TestExtractCardImageKeys Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * refactor(feishu): pass external image URLs to LLM instead of downloading Instead of downloading external images from interactive cards, pass the URLs directly to LLM. This reduces network overhead and lets vision-capable models fetch images as needed. - Remove downloadExternalImage function - Append external URLs to card content for LLM processing - Only download Feishu-hosted images via API 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): add blank line between functions for gci formatting * fix(feishu): keep interactive card content as valid JSON
2026-03-20 04:59:43 +00:00
case larkim.MsgTypeInteractive:
// Pass raw JSON to LLM — structured card is more informative than flattened text
return rawContent
case larkim.MsgTypeImage:
// Image messages don't have text content
return ""
case larkim.MsgTypeFile, larkim.MsgTypeAudio, larkim.MsgTypeMedia:
// File/audio/video messages may have a filename
name := extractFileName(rawContent)
if name != "" {
return name
}
return ""
default:
return rawContent
}
}
// downloadInboundMedia downloads media from inbound messages and stores in MediaStore.
func (c *FeishuChannel) downloadInboundMedia(
ctx context.Context,
chatID, messageID, messageType, rawContent string,
store media.MediaStore,
) []string {
var refs []string
scope := channels.BuildMediaScope("feishu", chatID, messageID)
switch messageType {
case larkim.MsgTypeImage:
imageKey := extractImageKey(rawContent)
if imageKey == "" {
return nil
}
ref := c.downloadResource(ctx, messageID, imageKey, "image", ".jpg", store, scope)
if ref != "" {
refs = append(refs, ref)
}
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
case larkim.MsgTypePost:
for _, imageKey := range extractPostImageKeys(rawContent) {
ref := c.downloadResource(ctx, messageID, imageKey, "image", ".jpg", store, scope)
if ref != "" {
refs = append(refs, ref)
}
}
Feat/feishu card parsing (#1534) * feat(feishu): add interactive card message parsing Add support for parsing inbound Feishu interactive card messages. When a user sends a card message, the text content is now extracted and passed to the LLM for processing. - Add extractCardText() to recursively extract text from card JSON - Support both JSON 1.0 (legacy) and JSON 2.0 schema formats - Handle nested elements: header, body, actions, columns - Extract text from markdown, lark_md, and plain_text elements - Add comprehensive unit tests for card parsing Fixes #<issue_number> 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * feat(feishu): extract and download images from interactive cards When receiving interactive card messages, extract embedded images (img_key, src, icon_key) and download them for LLM processing. - Add extractCardImageKeys() to recursively extract image keys from card JSON - Support img elements (img_key, src) and icon elements (icon_key) - Update downloadInboundMedia() to handle MsgTypeInteractive - Add comprehensive unit tests for image extraction Images are downloaded and stored via MediaStore, then appended to the message content as [image: photo] tags for LLM visibility. 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): simplify card parsing - pass raw JSON, only extract images Address review feedback: text extraction cannot exhaustively handle all card formats (i18n_elements, div.fields, etc.). Pass raw JSON to LLM instead - same approach as MsgTypePost. Only image extraction remains as images must be downloaded for LLM to process. - Remove extractCardText() and helper functions - extractContent() now returns raw JSON for MsgTypeInteractive - Keep extractCardImageKeys() for downloading embedded images - Update tests to expect raw JSON for interactive cards * fix(feishu): don't append media tags to interactive card JSON Appending media tags like "[attachment]" to raw JSON content produces invalid JSON format. For interactive cards, the JSON already contains image information and media refs are downloaded separately. - Skip appendMediaTags for MsgTypeInteractive to preserve valid JSON - Add test case for interactive card with images * fix(feishu): filter out external URLs from card image extraction Only Feishu-hosted image keys (img_xxx, icon_xxx) can be downloaded via the Feishu API. External URLs in src field (https://...) should be filtered out to avoid download failures. - Add isFeishuImageKey() to detect Feishu-hosted keys vs external URLs - Update extractImageKeysRecursive to skip external URLs in src field - Add tests for external URL filtering and mixed scenarios * feat(feishu): support downloading external images from interactive cards Previously only Feishu-hosted images (img_key, icon_key) could be downloaded. Now external URLs in src field are also downloaded via HTTP and made available to the LLM. - extractCardImageKeys now returns two slices: Feishu keys and external URLs - Add downloadExternalImage to download images from HTTP URLs - Update downloadInboundMedia to handle both Feishu API and HTTP downloads - Update tests for new function signature * fix(feishu): use HTTP client with timeout for external image downloads Replaced http.DefaultClient with a client that has a 30-second timeout to prevent hanging on unresponsive external URLs. Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): resolve lint errors for shadow and formatting - Rename err variables to avoid shadowing in downloadExternalImage - Fix struct field alignment in TestExtractCardImageKeys Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * refactor(feishu): pass external image URLs to LLM instead of downloading Instead of downloading external images from interactive cards, pass the URLs directly to LLM. This reduces network overhead and lets vision-capable models fetch images as needed. - Remove downloadExternalImage function - Append external URLs to card content for LLM processing - Only download Feishu-hosted images via API 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): add blank line between functions for gci formatting * fix(feishu): keep interactive card content as valid JSON
2026-03-20 04:59:43 +00:00
case larkim.MsgTypeInteractive:
// Extract and download images embedded in interactive cards
feishuKeys, _ := extractCardImageKeys(rawContent)
// Download Feishu-hosted images via API
for _, imageKey := range feishuKeys {
ref := c.downloadResource(ctx, messageID, imageKey, "image", ".jpg", store, scope)
if ref != "" {
refs = append(refs, ref)
}
}
// External URLs are passed directly to LLM, not downloaded
case larkim.MsgTypeFile, larkim.MsgTypeAudio, larkim.MsgTypeMedia:
fileKey := extractFileKey(rawContent)
if fileKey == "" {
return nil
}
// Derive a fallback extension from the message type.
var ext string
switch messageType {
case larkim.MsgTypeAudio:
ext = ".ogg"
case larkim.MsgTypeMedia:
ext = ".mp4"
default:
ext = "" // generic file — rely on resp.FileName
}
ref := c.downloadResource(ctx, messageID, fileKey, "file", ext, store, scope)
if ref != "" {
refs = append(refs, ref)
}
}
return refs
}
// downloadResource downloads a message resource (image/file) from Feishu,
// writes it to the project media directory, and stores the reference in MediaStore.
// fallbackExt (e.g. ".jpg") is appended when the resolved filename has no extension.
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
//
// For image resources, if the primary MessageResource.Get API fails (which
// requires im:message or im:message:readonly scope), a fallback to the
// Image.Get API (which requires im:resource scope) is attempted. This ensures
// image downloads succeed regardless of which permission the user has granted.
func (c *FeishuChannel) downloadResource(
ctx context.Context,
messageID, fileKey, resourceType, fallbackExt string,
store media.MediaStore,
scope string,
) string {
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
file, filename := c.fetchResourceData(ctx, messageID, fileKey, resourceType)
if file == nil {
return ""
}
if closer, ok := file.(io.Closer); ok {
defer closer.Close()
}
if filename == "" {
filename = fileKey
}
if filepath.Ext(filename) == "" && fallbackExt != "" {
filename += fallbackExt
}
return c.storeResourceFile(ctx, messageID, fileKey, filename, file, store, scope)
}
// fetchResourceData tries to download a resource from Feishu, first via
// MessageResource.Get, then falling back to Image.Get for image resources.
func (c *FeishuChannel) fetchResourceData(
ctx context.Context,
messageID, fileKey, resourceType string,
) (io.Reader, string) {
req := larkim.NewGetMessageResourceReqBuilder().
MessageId(messageID).
FileKey(fileKey).
Type(resourceType).
Build()
resp, err := c.client.Im.V1.MessageResource.Get(ctx, req)
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
if err == nil && resp.Success() && resp.File != nil {
return resp.File, resp.FileName
}
if err != nil {
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
logger.WarnCF("feishu", "MessageResource.Get failed", map[string]any{
"message_id": messageID,
"file_key": fileKey,
"error": err.Error(),
})
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
} else if !resp.Success() {
c.invalidateTokenOnAuthError(resp.Code)
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
logger.WarnCF("feishu", "MessageResource.Get api error", map[string]any{
"message_id": messageID,
"file_key": fileKey,
"code": resp.Code,
"msg": resp.Msg,
})
} else {
logger.WarnCF("feishu", "MessageResource.Get returned empty file body", map[string]any{
"message_id": messageID,
"file_key": fileKey,
})
}
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
if resourceType != "image" {
return nil, ""
}
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
return c.fetchImageDirect(ctx, fileKey)
}
// fetchImageDirect downloads an image using the Image.Get API
// (/open-apis/im/v1/images/:image_key), which requires the im:resource scope.
func (c *FeishuChannel) fetchImageDirect(ctx context.Context, imageKey string) (io.Reader, string) {
req := larkim.NewGetImageReqBuilder().
ImageKey(imageKey).
Build()
resp, err := c.client.Im.V1.Image.Get(ctx, req)
if err != nil {
logger.ErrorCF("feishu", "Image.Get fallback failed", map[string]any{
"image_key": imageKey,
"error": err.Error(),
})
return nil, ""
}
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
if !resp.Success() {
c.invalidateTokenOnAuthError(resp.Code)
logger.ErrorCF("feishu", "Image.Get fallback api error", map[string]any{
"image_key": imageKey,
"code": resp.Code,
"msg": resp.Msg,
})
return nil, ""
}
if resp.File == nil {
return nil, ""
}
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
logger.DebugCF("feishu", "Image downloaded via Image.Get fallback", map[string]any{
"image_key": imageKey,
})
return resp.File, resp.FileName
}
// storeResourceFile writes downloaded resource data to disk and registers it in the MediaStore.
func (c *FeishuChannel) storeResourceFile(
ctx context.Context,
messageID, fileKey, filename string,
file io.Reader,
store media.MediaStore,
scope string,
) string {
mediaDir := media.TempDir()
2026-03-02 17:04:06 +00:00
if mkdirErr := os.MkdirAll(mediaDir, 0o700); mkdirErr != nil {
logger.ErrorCF("feishu", "Failed to create media directory", map[string]any{
2026-03-02 17:04:06 +00:00
"error": mkdirErr.Error(),
})
return ""
}
ext := filepath.Ext(filename)
localPath := filepath.Join(mediaDir, utils.SanitizeFilename(messageID+"-"+fileKey+ext))
out, err := os.Create(localPath)
if err != nil {
logger.ErrorCF("feishu", "Failed to create local file for resource", map[string]any{
"error": err.Error(),
})
return ""
}
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
if _, copyErr := io.Copy(out, file); copyErr != nil {
out.Close()
os.Remove(localPath)
logger.ErrorCF("feishu", "Failed to write resource to file", map[string]any{
2026-03-02 17:04:06 +00:00
"error": copyErr.Error(),
})
return ""
}
if closeErr := out.Close(); closeErr != nil {
logger.ErrorCF("feishu", "Failed to close downloaded resource file", map[string]any{
"error": closeErr.Error(),
})
os.Remove(localPath)
return ""
}
ref, err := store.Store(localPath, media.MediaMeta{
Filename: filename,
Source: "feishu",
CleanupPolicy: media.CleanupPolicyDeleteOnCleanup,
}, scope)
if err != nil {
logger.ErrorCF("feishu", "Failed to store downloaded resource", map[string]any{
"file_key": fileKey,
"error": err.Error(),
})
os.Remove(localPath)
return ""
}
return ref
}
// appendMediaTags appends media type tags to content (like Telegram's "[image: photo]").
Feat/feishu card parsing (#1534) * feat(feishu): add interactive card message parsing Add support for parsing inbound Feishu interactive card messages. When a user sends a card message, the text content is now extracted and passed to the LLM for processing. - Add extractCardText() to recursively extract text from card JSON - Support both JSON 1.0 (legacy) and JSON 2.0 schema formats - Handle nested elements: header, body, actions, columns - Extract text from markdown, lark_md, and plain_text elements - Add comprehensive unit tests for card parsing Fixes #<issue_number> 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * feat(feishu): extract and download images from interactive cards When receiving interactive card messages, extract embedded images (img_key, src, icon_key) and download them for LLM processing. - Add extractCardImageKeys() to recursively extract image keys from card JSON - Support img elements (img_key, src) and icon elements (icon_key) - Update downloadInboundMedia() to handle MsgTypeInteractive - Add comprehensive unit tests for image extraction Images are downloaded and stored via MediaStore, then appended to the message content as [image: photo] tags for LLM visibility. 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): simplify card parsing - pass raw JSON, only extract images Address review feedback: text extraction cannot exhaustively handle all card formats (i18n_elements, div.fields, etc.). Pass raw JSON to LLM instead - same approach as MsgTypePost. Only image extraction remains as images must be downloaded for LLM to process. - Remove extractCardText() and helper functions - extractContent() now returns raw JSON for MsgTypeInteractive - Keep extractCardImageKeys() for downloading embedded images - Update tests to expect raw JSON for interactive cards * fix(feishu): don't append media tags to interactive card JSON Appending media tags like "[attachment]" to raw JSON content produces invalid JSON format. For interactive cards, the JSON already contains image information and media refs are downloaded separately. - Skip appendMediaTags for MsgTypeInteractive to preserve valid JSON - Add test case for interactive card with images * fix(feishu): filter out external URLs from card image extraction Only Feishu-hosted image keys (img_xxx, icon_xxx) can be downloaded via the Feishu API. External URLs in src field (https://...) should be filtered out to avoid download failures. - Add isFeishuImageKey() to detect Feishu-hosted keys vs external URLs - Update extractImageKeysRecursive to skip external URLs in src field - Add tests for external URL filtering and mixed scenarios * feat(feishu): support downloading external images from interactive cards Previously only Feishu-hosted images (img_key, icon_key) could be downloaded. Now external URLs in src field are also downloaded via HTTP and made available to the LLM. - extractCardImageKeys now returns two slices: Feishu keys and external URLs - Add downloadExternalImage to download images from HTTP URLs - Update downloadInboundMedia to handle both Feishu API and HTTP downloads - Update tests for new function signature * fix(feishu): use HTTP client with timeout for external image downloads Replaced http.DefaultClient with a client that has a 30-second timeout to prevent hanging on unresponsive external URLs. Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): resolve lint errors for shadow and formatting - Rename err variables to avoid shadowing in downloadExternalImage - Fix struct field alignment in TestExtractCardImageKeys Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * refactor(feishu): pass external image URLs to LLM instead of downloading Instead of downloading external images from interactive cards, pass the URLs directly to LLM. This reduces network overhead and lets vision-capable models fetch images as needed. - Remove downloadExternalImage function - Append external URLs to card content for LLM processing - Only download Feishu-hosted images via API 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): add blank line between functions for gci formatting * fix(feishu): keep interactive card content as valid JSON
2026-03-20 04:59:43 +00:00
// For interactive cards, media tags are not appended because content is raw JSON
// and appending would produce invalid JSON format.
func appendMediaTags(content, messageType string, mediaRefs []string) string {
if len(mediaRefs) == 0 {
return content
}
fix(feishu): fix image download with API fallback and post image support (#2708) * fix(feishu): fix image download with API fallback and post image support - Add Image.Get API fallback when MessageResource.Get fails (different permission scope: im:resource vs im:message:readonly) - Extract and download images from post (rich text) messages - Extract images from interactive card messages - Deduplicate post image keys across locales - Add comprehensive tests for new helpers Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(media): add image path tags alongside base64 for LLM file access Images are still base64-encoded into msg.Media for multimodal LLMs, but now also get [image:path] tags injected into message content so the LLM knows the local file path for save/forward operations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * refactor(media): only auto-inject images for tool results, not user messages Channel-received images (role=user) now get path tags only, letting the LLM decide whether to view via load_image or just operate on the file. Tool result images (role=tool, e.g. load_image) are base64-encoded into a synthetic user message appended after the tool message, since many LLM APIs don't support image_url in tool messages. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): preserve tool-message ordering for multi-tool-call scenarios Move synthetic user message (carrying base64 tool images) to after the entire contiguous tool-message block instead of immediately after each tool message. This preserves the assistant→tool→tool ordering required by OpenAI-compatible APIs. Also fix load_image to use generic [image: photo] placeholder so injectPathTags can properly replace it with the actual path. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(test): update load_image test for [image: photo] placeholder The test was checking ForLLM for the media:// ref, but load_image now emits the generic [image: photo] placeholder instead. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): match all channel image placeholders in injectPathTags Different channels emit different placeholder formats — Telegram/Feishu use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use [image: <filename>]. The previous string-match code only handled [image: photo], so for the other channels the path tag was appended as a duplicate, producing content like "[image] [image:/path]". Switch to per-type regex that matches all generic placeholder shapes while leaving path tags ([image:/path]) untouched. Also fixes the same issue for [audio], [video], [file] tags. Added test coverage for the various placeholder shapes. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(media): skip path tag append for JSON content (Feishu cards/posts) When content is structured JSON (interactive cards, post messages), injectPathTags now skips the fallback append — only placeholder replacement is attempted. This prevents corrupting JSON payloads like {"schema":"2.0",...} with appended [image:/path] tags. Adds looksLikeJSON() helper and three test cases covering JSON objects, arrays, and an end-to-end resolveMediaRefs scenario with Feishu card content. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(media): prepend path tags for JSON content, narrow looksLikeJSON Two fixes from code review: 1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding false positives on regular text like "[update] see attached". 2. For JSON content (Feishu cards/posts), path tags are prepended before the JSON instead of being silently dropped. This ensures the LLM can discover attached images via the path tag while the JSON payload stays valid for downstream parsing. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-30 03:08:00 +00:00
// Don't append tags to JSON content - would produce invalid JSON
if messageType == larkim.MsgTypeInteractive || messageType == larkim.MsgTypePost {
Feat/feishu card parsing (#1534) * feat(feishu): add interactive card message parsing Add support for parsing inbound Feishu interactive card messages. When a user sends a card message, the text content is now extracted and passed to the LLM for processing. - Add extractCardText() to recursively extract text from card JSON - Support both JSON 1.0 (legacy) and JSON 2.0 schema formats - Handle nested elements: header, body, actions, columns - Extract text from markdown, lark_md, and plain_text elements - Add comprehensive unit tests for card parsing Fixes #<issue_number> 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * feat(feishu): extract and download images from interactive cards When receiving interactive card messages, extract embedded images (img_key, src, icon_key) and download them for LLM processing. - Add extractCardImageKeys() to recursively extract image keys from card JSON - Support img elements (img_key, src) and icon elements (icon_key) - Update downloadInboundMedia() to handle MsgTypeInteractive - Add comprehensive unit tests for image extraction Images are downloaded and stored via MediaStore, then appended to the message content as [image: photo] tags for LLM visibility. 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): simplify card parsing - pass raw JSON, only extract images Address review feedback: text extraction cannot exhaustively handle all card formats (i18n_elements, div.fields, etc.). Pass raw JSON to LLM instead - same approach as MsgTypePost. Only image extraction remains as images must be downloaded for LLM to process. - Remove extractCardText() and helper functions - extractContent() now returns raw JSON for MsgTypeInteractive - Keep extractCardImageKeys() for downloading embedded images - Update tests to expect raw JSON for interactive cards * fix(feishu): don't append media tags to interactive card JSON Appending media tags like "[attachment]" to raw JSON content produces invalid JSON format. For interactive cards, the JSON already contains image information and media refs are downloaded separately. - Skip appendMediaTags for MsgTypeInteractive to preserve valid JSON - Add test case for interactive card with images * fix(feishu): filter out external URLs from card image extraction Only Feishu-hosted image keys (img_xxx, icon_xxx) can be downloaded via the Feishu API. External URLs in src field (https://...) should be filtered out to avoid download failures. - Add isFeishuImageKey() to detect Feishu-hosted keys vs external URLs - Update extractImageKeysRecursive to skip external URLs in src field - Add tests for external URL filtering and mixed scenarios * feat(feishu): support downloading external images from interactive cards Previously only Feishu-hosted images (img_key, icon_key) could be downloaded. Now external URLs in src field are also downloaded via HTTP and made available to the LLM. - extractCardImageKeys now returns two slices: Feishu keys and external URLs - Add downloadExternalImage to download images from HTTP URLs - Update downloadInboundMedia to handle both Feishu API and HTTP downloads - Update tests for new function signature * fix(feishu): use HTTP client with timeout for external image downloads Replaced http.DefaultClient with a client that has a 30-second timeout to prevent hanging on unresponsive external URLs. Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): resolve lint errors for shadow and formatting - Rename err variables to avoid shadowing in downloadExternalImage - Fix struct field alignment in TestExtractCardImageKeys Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * refactor(feishu): pass external image URLs to LLM instead of downloading Instead of downloading external images from interactive cards, pass the URLs directly to LLM. This reduces network overhead and lets vision-capable models fetch images as needed. - Remove downloadExternalImage function - Append external URLs to card content for LLM processing - Only download Feishu-hosted images via API 💘 Generated with Crush Assisted-by: GLM-5 via Crush <crush@charm.land> * fix(feishu): add blank line between functions for gci formatting * fix(feishu): keep interactive card content as valid JSON
2026-03-20 04:59:43 +00:00
return content
}
var tag string
switch messageType {
case larkim.MsgTypeImage:
tag = "[image: photo]"
case larkim.MsgTypeAudio:
tag = "[audio]"
case larkim.MsgTypeMedia:
tag = "[video]"
case larkim.MsgTypeFile:
tag = "[file]"
default:
tag = "[attachment]"
}
if content == "" {
return tag
}
return content + " " + tag
}
// sendCard sends an interactive card message to a chat.
func (c *FeishuChannel) sendCard(ctx context.Context, chatID, cardContent string) (string, error) {
req := larkim.NewCreateMessageReqBuilder().
ReceiveIdType(larkim.CreateMessageV1ReceiveIDTypeChatId).
Body(larkim.NewCreateMessageReqBodyBuilder().
ReceiveId(chatID).
MsgType(larkim.MsgTypeInteractive).
Content(cardContent).
Build()).
Build()
resp, err := c.client.Im.V1.Message.Create(ctx, req)
if err != nil {
return "", fmt.Errorf("feishu send card: %w", channels.ErrTemporary)
}
if !resp.Success() {
c.invalidateTokenOnAuthError(resp.Code)
return "", fmt.Errorf("feishu api error (code=%d msg=%s): %w", resp.Code, resp.Msg, channels.ErrTemporary)
}
logger.DebugCF("feishu", "Feishu card message sent", map[string]any{
"chat_id": chatID,
})
if resp.Data != nil && resp.Data.MessageId != nil {
return *resp.Data.MessageId, nil
}
return "", nil
}
// sendText sends a plain text message to a chat (fallback when card fails).
func (c *FeishuChannel) sendText(ctx context.Context, chatID, text string) (string, error) {
content, _ := json.Marshal(map[string]string{"text": text})
req := larkim.NewCreateMessageReqBuilder().
ReceiveIdType(larkim.CreateMessageV1ReceiveIDTypeChatId).
Body(larkim.NewCreateMessageReqBodyBuilder().
ReceiveId(chatID).
MsgType(larkim.MsgTypeText).
Content(string(content)).
Build()).
Build()
resp, err := c.client.Im.V1.Message.Create(ctx, req)
if err != nil {
return "", fmt.Errorf("feishu send text: %w", channels.ErrTemporary)
}
if !resp.Success() {
return "", fmt.Errorf("feishu text api error (code=%d msg=%s): %w", resp.Code, resp.Msg, channels.ErrTemporary)
}
logger.DebugCF("feishu", "Feishu text message sent (fallback)", map[string]any{
"chat_id": chatID,
})
if resp.Data != nil && resp.Data.MessageId != nil {
return *resp.Data.MessageId, nil
}
return "", nil
}
// sendImage uploads an image and sends it as a message.
func (c *FeishuChannel) sendImage(ctx context.Context, chatID string, file *os.File) error {
// Upload image to get image_key
uploadReq := larkim.NewCreateImageReqBuilder().
Body(larkim.NewCreateImageReqBodyBuilder().
ImageType("message").
Image(file).
Build()).
Build()
uploadResp, err := c.client.Im.V1.Image.Create(ctx, uploadReq)
if err != nil {
return fmt.Errorf("feishu image upload: %w", err)
}
if !uploadResp.Success() {
c.invalidateTokenOnAuthError(uploadResp.Code)
return fmt.Errorf("feishu image upload api error (code=%d msg=%s)", uploadResp.Code, uploadResp.Msg)
}
if uploadResp.Data == nil || uploadResp.Data.ImageKey == nil {
return fmt.Errorf("feishu image upload: no image_key returned")
}
imageKey := *uploadResp.Data.ImageKey
// Send image message
content, _ := json.Marshal(map[string]string{"image_key": imageKey})
req := larkim.NewCreateMessageReqBuilder().
ReceiveIdType(larkim.CreateMessageV1ReceiveIDTypeChatId).
Body(larkim.NewCreateMessageReqBodyBuilder().
ReceiveId(chatID).
MsgType(larkim.MsgTypeImage).
Content(string(content)).
Build()).
Build()
resp, err := c.client.Im.V1.Message.Create(ctx, req)
if err != nil {
return fmt.Errorf("feishu image send: %w", err)
}
if !resp.Success() {
c.invalidateTokenOnAuthError(resp.Code)
return fmt.Errorf("feishu image send api error (code=%d msg=%s)", resp.Code, resp.Msg)
}
return nil
}
// sendFile uploads a file and sends it as a message.
func (c *FeishuChannel) sendFile(ctx context.Context, chatID string, file *os.File, filename, fileType string) error {
// Map part type to Feishu file type
feishuFileType := "stream"
switch fileType {
case "audio":
feishuFileType = "opus"
case "video":
feishuFileType = "mp4"
}
// Upload file to get file_key
uploadReq := larkim.NewCreateFileReqBuilder().
Body(larkim.NewCreateFileReqBodyBuilder().
FileType(feishuFileType).
FileName(filename).
File(file).
Build()).
Build()
uploadResp, err := c.client.Im.V1.File.Create(ctx, uploadReq)
if err != nil {
return fmt.Errorf("feishu file upload: %w", err)
}
if !uploadResp.Success() {
c.invalidateTokenOnAuthError(uploadResp.Code)
return fmt.Errorf("feishu file upload api error (code=%d msg=%s)", uploadResp.Code, uploadResp.Msg)
}
if uploadResp.Data == nil || uploadResp.Data.FileKey == nil {
return fmt.Errorf("feishu file upload: no file_key returned")
}
fileKey := *uploadResp.Data.FileKey
// Send file message
content, _ := json.Marshal(map[string]string{"file_key": fileKey})
req := larkim.NewCreateMessageReqBuilder().
ReceiveIdType(larkim.CreateMessageV1ReceiveIDTypeChatId).
Body(larkim.NewCreateMessageReqBodyBuilder().
ReceiveId(chatID).
MsgType(larkim.MsgTypeFile).
Content(string(content)).
Build()).
Build()
resp, err := c.client.Im.V1.Message.Create(ctx, req)
if err != nil {
return fmt.Errorf("feishu file send: %w", err)
}
if !resp.Success() {
c.invalidateTokenOnAuthError(resp.Code)
return fmt.Errorf("feishu file send api error (code=%d msg=%s)", resp.Code, resp.Msg)
}
return nil
}
func extractFeishuSenderID(sender *larkim.EventSender) string {
if sender == nil || sender.SenderId == nil {
return ""
}
if sender.SenderId.UserId != nil && *sender.SenderId.UserId != "" {
return *sender.SenderId.UserId
}
if sender.SenderId.OpenId != nil && *sender.SenderId.OpenId != "" {
return *sender.SenderId.OpenId
}
if sender.SenderId.UnionId != nil && *sender.SenderId.UnionId != "" {
return *sender.SenderId.UnionId
}
return ""
}
// invalidateTokenOnAuthError clears the cached tenant_access_token when the
// Feishu API reports it as invalid (99991663), so the next request fetches a
// fresh one. The Lark SDK's built-in retry does not clear the cache, causing
// all API calls to fail until the token naturally expires (~2 hours).
func (c *FeishuChannel) invalidateTokenOnAuthError(code int) {
if code == errCodeTenantTokenInvalid {
c.tokenCache.InvalidateAll()
logger.WarnCF("feishu", "Invalidated cached token due to auth error", nil)
}
}