telegram-mini-apps-react
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEDATA_EXFILTRATION
Full Analysis
- [Indirect Prompt Injection] (LOW): The skill demonstrates processing untrusted input from Telegram deep links and user profile data which could be exploited if interpolated into agent prompts. • Ingestion points: initData.startParam() in 'references/deeplink.md' and initData.user() in 'examples/user-data.tsx'. • Boundary markers: Absent; data is decoded and used directly for navigation or display. • Capability inventory: Subprocess-equivalent navigation via 'useNavigate' and network operations via 'fetch' in the ApiClient. • Sanitization: Absent; the code performs Base64 decoding but lacks semantic validation of the resulting parameters.
- [Data Exposure & Exfiltration] (LOW): The 'ApiClient' class in 'examples/user-data.tsx' is configured to send 'initDataRaw' (sensitive Telegram user and session data) to 'https://api.example.com'. While used as a placeholder, this pattern involves transmitting authentication tokens to a non-whitelisted external domain.
Audit Metadata