This skill contains three exploitation scripts targeting LangFlow instances, sharing the same core RCE primitive— injecting Python code into a custom component in a public flow—but differing in how the JWT token is obtained.
All scripts are located in the scripts/ directory and depend on requests(pip install requests)。
Quick decision tree
- The target has CVE-2026-33017 (
auto_loginunpatched) → cve-2026-33017.py auto_loginpatched, default password unchanged → langflowLogin.py- Both uncertain → langflowWeakRCE.py(covers all paths)
Quick reference table
| Scenario | Script | Authentication method | Minimal command |
|---|---|---|---|
| Unauthorized (CVE-2026-33017) | scripts/cve-2026-33017.py |
auto_login bypass |
python3 scripts/cve-2026-33017.py IP -c "id" |
| Default password | scripts/langflowLogin.py |
Default credentials | python3 scripts/langflowLogin.py IP -c "id" |
| Weak password brute force | scripts/langflowWeakRCE.py |
7-password dictionary | python3 scripts/langflowWeakRCE.py IP -c "id" |
Exploitation chain summary
All scripts follow: obtain JWT → create public flow → inject code execution → echo output.
Key endpoints:/api/v1/auto_login、/api/v1/login、/api/v1/flows/、
/api/v1/build_public_tmp/{id}/flow。