telegram-reader

Warn

Audited by Gen Agent Trust Hub on Apr 11, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses the local encrypted database and key files of the Telegram macOS app (Keepcoder) located at ~/Library/Group Containers/6N38VWS5BX.ru.keepcoder.Telegram/stable/. This provides the agent with full read access to the user's private message history, contact names, and peer IDs. Although the script operates locally and is read-only, the access to sensitive personal data constitutes significant exposure.
  • [COMMAND_EXECUTION]: The scripts/tg-reader utility uses the subprocess module to create a Python virtual environment and install external libraries (sqlcipher3, pycryptodomex, mmh3) when the --setup flag is utilized.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection risks because it ingests untrusted text content from Telegram messages into the agent's context. Ingestion points: Message data is read from the db_sqlite file in the account directories. Boundary markers: The output provided to the agent does not use delimiters or instructions to ignore embedded commands in the messages. Capability inventory: The skill allows the agent to search history, list peers, and read recent conversations. Sanitization: The extraction process retrieves printable strings from binary blobs but does not perform sanitization to filter out malicious instructions or escape content.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 11, 2026, 03:03 AM
Security Audit — agent-trust-hub — telegram-reader