Skip to content

Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.

fix(im): handle unresolved Feishu resource images (#2761)

Jones5c717ca045ab
diff --git a/internal/im/feishu/adapter.go b/internal/im/feishu/adapter.go
index eaa92c518..32b3b4617 100644
--- a/internal/im/feishu/adapter.go
+++ b/internal/im/feishu/adapter.go
@@ -1097,8 +1097,11 @@ func (a *Adapter) cardkitUpdateElement(ctx context.Context, accessToken, cardID,
// reaches the card we must download each referenced image and re-upload it to
// Feishu to obtain a usable image_key.
-// feishuMarkdownImageRe matches a markdown image whose target is an http(s) URL.
-var feishuMarkdownImageRe = regexp.MustCompile(`!\[([^\]]*)\]\((https?://[^)\s]+)\)`)
+// feishuMarkdownImageRe matches every markdown image. Only http(s) targets can
+// be uploaded to an image_key; unresolved internal handles (resource://,
+// local://, minio://, …) are not valid Feishu image keys, so
+// resolveMarkdownImages degrades them instead of leaving ![]() in the card.
+var feishuMarkdownImageRe = regexp.MustCompile(`!\[([^\]]*)\]\(([^)\s]+)\)`)
// feishuMarkdownLinkRe matches a markdown link whose target is an http(s) URL.
// Applied after image stripping so signed image URLs are not left as links.
@@ -1128,28 +1131,49 @@ func imageCacheKey(rawURL string) string {
}
// resolveMarkdownImages replaces the URL inside every ![alt](httpURL) with a
-// Feishu image_key. On failure it degrades the image to a plain text link so the
-// rest of the card still renders instead of failing the whole update.
+// Feishu image_key. On failure, or when the storage resolver left an internal
+// handle unresolved, it degrades the image so the rest of the card still
+// renders instead of failing the whole update. HTTP(S) targets become a
+// markdown link (Feishu accepts only http/https hrefs); other schemes become
+// plain text, because [text](resource://…) is also rejected as an illegal link.
func (a *Adapter) resolveMarkdownImages(ctx context.Context, accessToken, content string) string {
if !strings.Contains(content, "![") {
return content
}
+ fallback := func(alt, rawURL string) string {
+ if alt == "" {
+ alt = a.region.ImageFallbackLabel
+ }
+ if isHTTPImageURL(rawURL) {
+ return fmt.Sprintf("[%s](%s)", alt, rawURL)
+ }
+ return alt
+ }
return feishuMarkdownImageRe.ReplaceAllStringFunc(content, func(match string) string {
sub := feishuMarkdownImageRe.FindStringSubmatch(match)
+ if len(sub) < 3 {
+ return match
+ }
alt, rawURL := sub[1], sub[2]
+ if !isHTTPImageURL(rawURL) {
+ return fallback(alt, rawURL)
+ }
imgKey, err := a.imageKeyForURL(ctx, accessToken, rawURL)
if err != nil || imgKey == "" {
logger.Warnf(ctx, "[%s] image upload failed, degrading to link: url=%s err=%v", a.region.Label, rawURL, err)
- label := alt
- if label == "" {
- label = a.region.ImageFallbackLabel
- }
- return fmt.Sprintf("[%s](%s)", label, rawURL)
+ return fallback(alt, rawURL)
}
return fmt.Sprintf("![%s](%s)", alt, imgKey)
})
}
+// isHTTPImageURL reports whether rawURL is an http(s) URL. Scheme match is
+// case-insensitive: IM rewrite may emit HTTPS:// from a configured proxy domain.
+func isHTTPImageURL(rawURL string) bool {
+ lower := strings.ToLower(rawURL)
+ return strings.HasPrefix(lower, "http://") || strings.HasPrefix(lower, "https://")
+}
+
// imageKeyForURL returns a Feishu image_key for the given URL, uploading it if
// not already cached.
func (a *Adapter) imageKeyForURL(ctx context.Context, accessToken, rawURL string) (string, error) {
diff --git a/internal/im/feishu/adapter_test.go b/internal/im/feishu/adapter_test.go
index dabe4d75e..eb4484426 100644
--- a/internal/im/feishu/adapter_test.go
+++ b/internal/im/feishu/adapter_test.go
@@ -93,6 +93,61 @@ func TestResolveMarkdownImages_FallbackToLinkOnFailure(t *testing.T) {
}
}
+func TestResolveMarkdownImages_InternalHandleFallsBackToPlainText(t *testing.T) {
+ a := &Adapter{region: RegionFeishu}
+ tests := []struct {
+ name string
+ in string
+ want string
+ }{
+ {
+ name: "resource handle",
+ in: "see ![diagram](resource://Zb2IrqfTdlCFkbUqFtmS5A) here",
+ want: "see diagram here",
+ },
+ {
+ name: "local scheme",
+ in: "see ![shot](local://10000/exports/a.png) here",
+ want: "see shot here",
+ },
+ {
+ name: "empty alt uses region label",
+ in: "![](resource://Zb2IrqfTdlCFkbUqFtmS5A)",
+ want: "图片",
+ },
+ {
+ name: "mixed internal and failed http",
+ in: "![diagram](resource://Zb2IrqfTdlCFkbUqFtmS5A) and ![x](http://127.0.0.1/x.png)",
+ want: "diagram and [x](http://127.0.0.1/x.png)",
+ },
+ }
+ for _, tt := range tests {
+ t.Run(tt.name, func(t *testing.T) {
+ got := a.resolveMarkdownImages(context.Background(), "tok", tt.in)
+ if got != tt.want {
+ t.Errorf("fallback = %q, want %q", got, tt.want)
+ }
+ if strings.Contains(got, "![") {
+ t.Errorf("failed image should not remain as image markdown: %q", got)
+ }
+ if strings.Contains(got, "resource://") || strings.Contains(got, "local://") {
+ t.Errorf("internal scheme must not remain in card markdown: %q", got)
+ }
+ })
+ }
+}
+
+func TestResolveMarkdownImages_UppercaseHTTPSchemeStillUploads(t *testing.T) {
+ a := &Adapter{region: RegionFeishu}
+ // IM rewrite may emit HTTPS://; loopback still fails SSRF and must degrade
+ // to a link rather than being treated as a non-HTTP handle.
+ got := a.resolveMarkdownImages(context.Background(), "tok", "![](HTTPS://127.0.0.1/x.png)")
+ want := "[图片](HTTPS://127.0.0.1/x.png)"
+ if got != want {
+ t.Errorf("uppercase scheme fallback = %q, want %q", got, want)
+ }
+}
+
// An image with no alt text falls back to the region's own label, so Lark users
// do not get a Chinese link label.
func TestResolveMarkdownImages_FallbackLabelFollowsRegion(t *testing.T) {
@@ -387,6 +442,11 @@ func TestCardSummaryPreview(t *testing.T) {
}{
{name: "collapse whitespace", in: " 最终\n回答\t✅ ", want: "最终 回答 ✅"},
{name: "image keeps alt only", in: "![架构图](https://cdn.example/a.png?sig=secret) 说明", want: "架构图 说明"},
+ {
+ name: "resource image keeps alt only",
+ in: "![diagram](resource://Zb2IrqfTdlCFkbUqFtmS5A) 说明",
+ want: "diagram 说明",
+ },
{name: "link keeps text only", in: "见 [文档](https://cdn.example/a.png?sig=secret) 说明", want: "见 文档 说明"},
{name: "unicode limit", in: strings.Repeat("界", 121), want: strings.Repeat("界", 120)},
{name: "empty", in: " \n\t ", want: ""},