| 제목 | https://github.com/MemTensor memmy-agent 1.1.0 Local sensitive logic exposure lead to the user's usage leakage |
|---|
| 설명 | Project: Memmy Agent
Analyzed version: v1.1.0
Git commit: b4c37f78557b1d9a2ed9addf148c53d406ce954a
Affected component: Memory HTTP service
Affected endpoint: GET /api/v1/memory/logs
Default listen address: 127.0.0.1:18960
Vulnerability class: unauthenticated sensitive-log disclosure combined with permissive CORS
Attack surface: a web page rendered in a browser on the same machine as the victim
Impact: disclosure of stored memory-operation logs, including raw queries, Agent answers, recall candidates, memory summaries, tags, metadata, session identifiers, and other diagnostic data
Please note the specific video for the poc demonstration is here : https://drive.google.com/drive/folders/1u_JVxGUTjYXe-URL4EJT2aiOyiTC1f6k?usp=drive_link
If you want to manually find the vulnerability ,please gitclone as this command : git clone --branch v1.1.0 --depth 1 https://github.com/MemTensor/memmy-agent.git
The Memory HTTP service exposes GET /api/v1/memory/logs without authentication in its default local configuration. The service also applies a global wildcard CORS policy:
Access-Control-Allow-Origin: *
Access-Control-Allow-Methods: GET,POST,DELETE,OPTIONS
Access-Control-Allow-Headers: content-type,authorization,x-api-key,...
A browser web page can therefore issue a cross-origin GET request to http://127.0.0.1:18960/api/v1/memory/logs and read the JSON response. The default request in the supplied POC is a simple GET with no custom header, so it does not require a preflight. If the optional x-memmy-user-id header is supplied, the service also allows that header in an OPTIONS preflight response.
The endpoint returns the most recent memory_add and memory_search operation logs. Each log record contains:
tool name: memory_add, memory_search, skill_generate, or skill_evolve;
source Agent identifier;
inputJson;
outputJson;
duration;
success flag;
call time.
The inputJson and outputJson fields are the primary sensitive payload. Source code shows that they store complete JSON-serialized request and result objects. For example:
a memory_add log can contain the user query, Agent answer, source Agent, trace ID, episode ID, and memory summary;
a memory_search log can contain the search query, session and episode IDs, requested layers, candidate memories, filtered hits, dropped candidates, relevance statistics, and recall status.
Consequently, passwords, API keys, cookies, private keys, internal project names, source-code fragments, private conversation content, or other sensitive data previously processed by Memmy may be readable from the log endpoint.
The route supports limit and offset. The service clamps limit to at most 500 and returns nextOffset when more records exist. An attacker can therefore enumerate the log table in pages. Log retention is best-effort and limited to 10,000 rows, but that is still a substantial disclosure window.
The service listens only on 127.0.0.1 by default, which prevents direct access from other network hosts. This does not prevent an arbitrary web page in the victim's browser from making a localhost request. The wildcard CORS response removes the browser's normal cross-origin read barrier. |
|---|
| 원천 | ⚠️ https://github.com/MemTensor/memmy-agent/issues/462 |
|---|
| 사용자 | Anonymous User |
|---|
| 제출 | 2026. 10. 20. AM 10:22 (29 날 ago) |
|---|
| 모더레이션 | 2026. 11. 18. PM 06:27 (29 days later) |
|---|
| 상태 | 수락 |
|---|
| VulDB 항목 | 439658 [MemTensor memmy-agent 까지 1.1.1 Memory HTTP Service /api/v1/memory/logs 정보 공개] |
|---|
| 포인트들 | 20 |
|---|