The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs shell operations to verify the environment, check for the existence of the bot script, and execute the bot using the bun runtime from a local directory.
[DATA_EXFILTRATION]: The bot is designed to retrieve private data from a user's Gmail account (reading, searching, and listing emails) and transmit it to an external Telegram chat. While restricted to an authorized chat ID, this creates a channel for sensitive data to leave the primary environment.
[PROMPT_INJECTION]:
Self-Modification Instructions: The skill contains explicit 'Self-Evolving' instructions that direct the agent to modify the SKILL.md file immediately upon failure or observed drift. This feedback loop allows for the permanent alteration of the skill's operational logic.
Indirect Prompt Injection Surface:
Ingestion points: Untrusted data enters the agent's context through incoming Telegram messages and Gmail email content retrieved by tools.
Boundary markers: The instructions do not specify any boundary markers, delimiters, or safety warnings to prevent the agent from obeying instructions embedded within the emails it reads.
Capability inventory: The agent is granted capabilities to list, search, read, and draft emails in Gmail, which are high-privilege operations.
Sanitization: No sanitization or validation of external content (emails or messages) is documented before the content is passed to the AI routing tier.

interactive-bot