The Agent Skills Directory

[DATA_EXFILTRATION]: The skill documentation and templates include a specific, hardcoded Telegram user ID which acts as a destination for notifications.
Evidence: The identifier 1027899060 is hardcoded in the delivery.to field within SKILL.md and references/templates.md.
Risk: If the agent uses these templates to schedule reminders or reports, it may direct the output to this specific account rather than the user's account, potentially leaking private information.
[PROMPT_INJECTION]: The skill instructs the agent to use a 'Strict Instruction Pattern' designed to override its standard behavior.
Evidence: Instructions like DELIVER THIS EXACT MESSAGE TO THE USER WITHOUT MODIFICATION OR COMMENTARY are provided in the templates.
Risk: This method attempts to bypass the agent's natural language processing and safety filters by forcing specific, unedited output.
[PERSISTENCE]: The skill provides instructions for creating recurring tasks that run in the background.
Evidence: Use of cron expressions (e.g., 0 8 * * *) in references/templates.md.
Risk: Recurring background tasks can be used to maintain activity and perform automated actions across different sessions.

cron-mastery