The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local tools and scripts as part of its operational workflow.
Evidence: SKILL.md requires scripts/preflight.sh to be executed as a prerequisite.
Evidence: Step 2 in both the English and Korean skill definitions involves calling a distill command-line utility with arguments derived from user input.
[PROMPT_INJECTION]: The skill processes untrusted user input, which presents a surface for indirect prompt injection attacks.
Ingestion points: User input is captured via the {input} and {audience} variables in SKILL.md.
Boundary markers: The instructions lack explicit delimiters to encapsulate or isolate the user-provided data from the command structure.
Capability inventory: The skill utilizes the Write tool to create and modify files within the docs/ directory.
Sanitization: There is no mention of sanitization or validation logic for the content provided by the user.
Mitigation: The risk of malicious automated file operations is significantly reduced by a mandatory human-in-the-loop (HITL) step in Step 4, which requires the user to explicitly confirm any file-saving actions.

pitch