The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The execution block concatenates user-provided input (<ARGS>) directly into a shell command: --content "<ARGS>". This allows an attacker to execute arbitrary commands on the host system by using shell metacharacters such as backticks, dollar-parenthesis, or by closing the double quotes and appending a new command (e.g., "; rm -rf /; #).
[PROMPT_INJECTION] (LOW): The skill is a vector for indirect prompt injection. Since it stores 'learnings' in a database for future recall, a malicious user or untrusted data source could inject instructions that the agent later retrieves and treats as authoritative guidelines or patterns.
Ingestion points: User input via the /remember command.
Boundary markers: Absent; the content is wrapped in quotes but not sanitized for instruction markers.
Capability inventory: Execution of local Python scripts via shell subprocess.
Sanitization: None detected in the provided skill definition.

remember