The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to analyze the current session history and distill it into a self-contained prompt for future autonomous runs. This creates a significant surface for indirect prompt injection where an attacker could place malicious instructions in the session history (e.g., via a downloaded file or web content) that the agent then incorporates into a persistent, scheduled task.
Ingestion points: Review of session history in Step 1 of the 'Creating a new task' workflow.
Boundary markers: None; the agent is explicitly told to make the prompt self-contained and second-person imperative, which could facilitate the adoption of injected instructions.
Capability inventory: The skill uses the create_scheduled_task and update_scheduled_task tools to schedule future execution.
Sanitization: The skill lacks instructions to sanitize or validate the content gathered from the session history before generating the new task prompt.
[PERSISTENCE]: The skill provides a functional interface for creating recurring tasks using cronExpression. While this is the intended purpose, it establishes a persistence mechanism that allows prompts to be executed repeatedly without direct user supervision or initiation, which can be abused if the task generation logic is compromised.

schedule