actions/claude-code-action
1
0
Fork 0
mirror of https://github.com/anthropics/claude-code-action.git synced 2026-08-22 03:18:54 +08:00
Code Issues Packages Projects Releases Wiki Activity

Match downloaded images to their source URLs by asset identifier (#1588)

Browse Source 
* Match downloaded images to source URLs by asset identifier

* Derive the signed URL asset identifier from the parsed path
This commit is contained in:
Ashwin Bhat
2026-08-04 10:04:47 -07:00
committed by GitHub
parent 6fb6bb6858
commit b80a0f042f
2 changed files with 328 additions and 110 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/github/utils/image-downloader.ts
+56 -10
View File
@@ -14,6 +14,39 @@ const HTML_IMG_REGEX = new RegExp(
"gi",
);
const SIGNED_URL_REGEX =
/https:\/\/private-user-images\.githubusercontent\.com\/[^"]+\?jwt=[^"]+/g;
// GitHub identifies an uploaded asset by a GUID that appears both in the
// user-attachment URL and in the signed download URL rendered in body_html.
const ASSET_GUID_REGEX =
/[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}/i;
function extractAssetGuid(url: string): string | undefined {
return url.match(ASSET_GUID_REGEX)?.[0]?.toLowerCase();
}
const SIGNED_URL_HOST = "private-user-images.githubusercontent.com";
// Signed download URLs have the shape /<owner-id>/<asset-id>-<guid>.<ext>.
// The GUID must come from the resolved filename, not from anywhere in the raw
// string, so text that merely embeds a GUID cannot claim another asset.
const SIGNED_URL_PATH_REGEX =
/^\/[^/]+\/[^/]*-([0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12})(?:\.[a-z0-9]+)?$/i;
function extractSignedUrlAssetGuid(signedUrl: string): string | undefined {
let parsed: URL;
try {
parsed = new URL(signedUrl);
} catch {
return undefined;
}
if (parsed.host !== SIGNED_URL_HOST) {
return undefined;
}
return parsed.pathname.match(SIGNED_URL_PATH_REGEX)?.[1]?.toLowerCase();
}
type IssueComment = {
type: "issue_comment";
id: string;
@@ -174,21 +207,34 @@ export async function downloadCommentImages(
}
// Extract signed URLs from HTML
const signedUrlRegex =
/https:\/\/private-user-images\.githubusercontent\.com\/[^"]+\?jwt=[^"]+/g;
const signedUrls = bodyHtml.match(signedUrlRegex) || [];
const signedUrls = bodyHtml.match(SIGNED_URL_REGEX) || [];
// Index the signed URLs by the asset GUID they reference. The signed
// URLs come from a separate render of the body, so their order and
// count are not guaranteed to line up with the URLs extracted from the
// markdown; pairing by asset identifier keeps each download tied to the
// URL it actually belongs to.
const signedUrlByGuid = new Map<string, string>();
for (const signedUrl of signedUrls) {
const guid = extractSignedUrlAssetGuid(signedUrl);
if (guid && !signedUrlByGuid.has(guid)) {
signedUrlByGuid.set(guid, signedUrl);
}
}
// Download each image
for (let i = 0; i < Math.min(signedUrls.length, urls.length); i++) {
const signedUrl = signedUrls[i];
const originalUrl = urls[i];
if (!signedUrl || !originalUrl) {
for (const [i, originalUrl] of urls.entries()) {
// Check if we've already downloaded this URL
if (urlToPathMap.has(originalUrl)) {
continue;
}
// Check if we've already downloaded this URL
if (urlToPathMap.has(originalUrl)) {
const guid = extractAssetGuid(originalUrl);
const signedUrl = guid ? signedUrlByGuid.get(guid) : undefined;
if (!signedUrl) {
console.warn(
`No matching signed URL found for ${originalUrl}, skipping`,
);
continue;
}