The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted text and use it to modify local files, creating a potential vector for indirect prompt injection. 1. Ingestion points: The skill reads content from inline user inputs and external files via the read_file tool. 2. Boundary markers: The prompt instructions do not provide explicit delimiters or instructions to treat the input as pure data to prevent the execution of embedded instructions. 3. Capability inventory: The skill utilizes the patch and write_file tools to apply changes to the workspace. 4. Sanitization: There are no specific instructions to sanitize or validate the input text to prevent malicious instruction injection into the file-writing process.

humanizer