The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the processing of external JSON content through an LLM-based copywriting skill, creating a potential surface for indirect prompt injection. The implementation includes significant mitigations.
Ingestion points: Data is ingested from the user-provided FILE_PATH (JSON file) via the Read tool.
Boundary markers: Extracted fields are isolated using specific HTML comment delimiters () in temporary markdown files to maintain data structure.
Capability inventory: The skill is configured with Read, Write, Edit, Bash, and Skill tool access to perform file operations and delegate copywriting tasks.
Sanitization: The skill implements defensive validation by rejecting polished text that contains markdown syntax (like headers or bolding), enforcing length constraints (2x-2.5x the original size), and verifying character sets against language-specific standards. This prevents the polished content from being used to inject unintended formatting or structure into the target JSON file.

copy-json