The Agent Skills Directory

[PROMPT_INJECTION]: The skill allows for the creation of scheduled tasks that deliver text messages (prompts) to agents at a future time. This creates a surface for indirect prompt injection where malicious instructions can be persisted and executed with a delay.
Ingestion points: Untrusted data enters the system via the --message argument in scripts/schedule.sh or the message field in the /api/schedules REST endpoint.
Boundary markers: The skill does not implement delimiters or instructions to ignore embedded commands within the scheduled messages.
Capability inventory: Scheduled messages are delivered to agents who may have high-privilege capabilities such as file system access or shell execution.
Sanitization: While scripts/schedule.sh performs basic character escaping for JSON compatibility, it does not sanitize the content of the message for malicious instructions.
[COMMAND_EXECUTION]: The scripts/schedule.sh utility manually constructs JSON payloads for curl commands using string concatenation. Several variables, including --agent, --channel, and --sender, are interpolated into the JSON string without escaping, which could lead to malformed JSON or payload injection if these arguments contain double quotes or other special characters.

schedule