The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a reference implementation for an alert management script (alert.sh) that executes several standard Unix utilities including grep, sed, shasum, stat, date, and tail. It also triggers a local notification script (notify.sh) for urgent alerts. These commands are used to manage local state and process alert logic as described in the documentation.
[DATA_EXFILTRATION]: The alert.sh script performs file operations, writing alert data to local paths such as alerts.log, .chat-queue, .pending-alerts, and outbox.txt. No unauthorized network operations or exfiltration to external domains were detected; all data movement is local and aligns with the stated purpose of queuing alerts for agent triage.
[PROMPT_INJECTION]: The skill establishes a surface for indirect prompt injection by accepting arbitrary message content from external sources and storing it for agent triaging.
Ingestion points: The alert.sh script ingests untrusted text via the $MSG positional parameter.
Boundary markers: The implementation appends raw message content directly to triage files without using delimiters (like XML tags or triple backticks) or providing instructions to the triaging agent to ignore embedded commands.
Capability inventory: The script performs file system writes and executes shell commands. The intended workflow involves an agent reading these messages to perform automated triage and resolution.
Sanitization: There is no sanitization or filtering of the message content before it is stored in the queue, potentially allowing malicious instructions in an alert to influence the triaging agent's behavior.

alert-routing