The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a task monitoring workflow that reads the content of messages from other agents via sessions_list to determine state transitions. This creates an indirect prompt injection surface where a monitored agent could influence the supervising agent's actions.
Ingestion points: The sessions_list tool is called in 'Step 2' and 'Batch Check' in SKILL.md to retrieve session status and message content.
Boundary markers: Absent; the instructions do not specify delimiters or provide warnings to the agent to ignore instructions embedded within the monitored messages.
Capability inventory: The sessions_send tool is used across multiple steps to dispatch new commands, status updates, and prompts based on the ingested data.
Sanitization: Absent; there is no logic described for validating or filtering the message content retrieved from sessions before it is used to trigger subsequent logic.
[DATA_EXFILTRATION]: The skill instructions contain a hardcoded Telegram group identifier (-1003890797239) within the sessionKey parameter. This exposes specific internal infrastructure details in the skill's logic.

agent-task-confirm