The Agent Skills Directory

Prompt Injection (LOW): The skill is susceptible to indirect prompt injection because it processes untrusted data from Telegram messages.
Ingestion points: The script telegram_read.py reads message content from a local Telegram database.
Boundary markers: Messages are delimited using Markdown headers (e.g., `### 10:30 AM
@alice`), but there are no explicit system instructions for the agent to ignore potentially malicious commands embedded within those messages.
Capability inventory: The skill displays message content to the agent, which may include instructions that attempt to hijack the conversation or extract information.
Sanitization: No sanitization or content filtering is described in the documentation for the message data before it is presented to the model.
Command Execution (SAFE): The skill executes local Python scripts (telegram_read.py) to perform its functions. This is the primary intended behavior and does not involve arbitrary or high-risk command execution outside the scope of the skill's utility.

telegram-read