The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input via the $ARGUMENTS variable to define measurement targets, which can contain malicious instructions.
Ingestion points: The $ARGUMENTS placeholder in SKILL.md receives input from the calling orchestrator.
Boundary markers: No delimiters or protective instructions are used to separate the input data from the skill's logic.
Capability inventory: The skill requires Bash and Read tools for file system interaction.
Sanitization: The instructions do not specify any validation or escaping for the file paths or content received via input.
[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands such as wc -c <path> using arguments provided externally. This patterns represents a command injection risk if the input contains shell metacharacters that are not properly handled by the agent's execution environment.

impact-measurement