| 제목 | continuedev continue 2.0.0-vscode Server-Side Request Forgery |
|---|
| 설명 | A workspace can ship a .continue/docs/*.yaml block whose startUrl points at a loopback or internal address and which sets useLocalCrawling:true. useLocalCrawling is a per-doc field in the config schema (packages/config-yaml/src/schemas/index.ts, docs type, approx line 33), so the workspace controls it. Workspace-local .continue docs blocks are auto-merged into the active config (core/config/yaml/loadYaml.ts merges workspace definition files with includeWorkspace:true). DocsService.init() registers configHandler.onConfigUpdate(...) which calls syncDocs(...) (DocsService.ts approx lines 235 and 385); syncDocs indexes added docs via indexAndAdd(doc) with no user prompt (approx line 1008). In DocsCrawler (core/indexing/docs/crawlers/DocsCrawler.ts approx lines 52-82) the default/hosted crawler cannot reach internal URLs, so the code falls back to CheerioCrawler, whose fetch (CheerioCrawler.ts approx line 94) applies NO SSRF protection - no private-IP, loopback, or scheme filtering - and follows in-page links recursively. The retrieved internal content is embedded/indexed and becomes retrievable by the model through @docs.
Impact: the victim host issues GET requests to attacker-chosen loopback/internal URLs (including cloud metadata such as http://x.x.x.x/); the response body is read, crawled recursively, and indexed, then retrievable via @docs.
CVSS 3.1: AV:L/AC:H/PR:N/UI:R/S:U/C:H/I:N/A:N = 4.7.
Preconditions: an embeddings provider must be configured (otherwise indexAndAdd returns before the crawl); the attacker must know or guess the internal URL; exfiltration via @docs requires the model to be asked or injected to use @docs.
Proof of concept:
Attacker ships this file as .continue/docs/internal.yaml:
name: Internal docs
version: 0.0.1
schema: v1
docs:
- name: Internal service
startUrl: http://127.0.0.1:8080/
useLocalCrawling: true
(startUrl can be http://x.x.x.x/latest/meta-data/ for cloud IAM credential theft.) Once an embeddings provider is configured, Continue auto-indexes the added doc; CheerioCrawler fetches the internal URL from the victim host and follows links. Reproduced dynamically with the real CheerioCrawler and cheerio against a loopback server: the crawler fetches an internal page containing a secret and follows an in-page link to a second internal page (1 test passes). The fetched content is then retrievable via @docs.
Affected: continuedev/continue VS Code extension 2.0.0 (tag v2.0.0-vscode, 03b05ef); crawler code byte-identical in v2.1.0-vscode and main 5522c6f. Continue acquired by Cursor June 2026, repository read-only, no vendor fix will be issued.
Vendor status: [email protected] contacted 2026-10-15; no response.
Ship .continue/docs/internal.yaml with startUrl http://x.x.x.x/ (or a loopback port) and useLocalCrawling: true; Continue auto-indexes it and CheerioCrawler fetches the internal URL from the victim host with no SSRF filtering, exposing the response via @docs. Full PoC on request. |
|---|
| 사용자 | Yujin (UID 102324) |
|---|
| 제출 | 2026. 10. 22. AM 06:40 (29 날 ago) |
|---|
| 모더레이션 | 2026. 11. 20. PM 12:52 (29 days later) |
|---|
| 상태 | 수락 |
|---|
| VulDB 항목 | 441421 [continuedev continue 2.0.0/2.1.0 Docs Crawler DocsCrawler.ts DocsCrawler startUrl/useLocalCrawling 권한 상승] |
|---|
| 포인트들 | 17 |
|---|