The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external content from a scenario file and terms provided by the user, incorporating them directly into prompts for multiple models without sanitization.
Ingestion points: The content of the <scenario-file> and the <term> arguments are read and processed in the probe.sh script (lines 19-25).
Boundary markers: There are no protective delimiters or "ignore" instructions used to prevent the AI model from potentially following commands embedded within the scenario text.
Capability inventory: The skill executes shell commands to invoke the claude CLI, which naturally has access to the user context and configuration files like ~/.claude/CLAUDE.md.
Sanitization: No input validation or filtering is performed on the user-controlled data before it is interpolated into the prompt.
[COMMAND_EXECUTION]: The probe.sh script executes the claude CLI using sub-shells to run tests in parallel. It utilizes the sed utility for string manipulation on user-provided inputs, which could lead to unintended script behavior if inputs contain unexpected control characters.

term-check