The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that passes raw user input directly into a sub-agent's execution prompt.
Ingestion points: The $ARGUMENTS variable in SKILL.md captures arbitrary user input which is then used to construct the prompt for the Agent tool.
Boundary markers: The skill lacks delimiters (such as XML tags or triple backticks) or explicit instructions to the sub-agent to treat the interpolated input as untrusted data rather than instructions.
Capability inventory: The skill uses the Agent tool to spawn sub-agents which have access to the Bash tool, granting them full shell execution capabilities.
Sanitization: There is no evidence of escaping, validation, or filtering of the $ARGUMENTS content before it is passed to the downstream agent.

parallel