The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download a ZIP archive containing an executable binary from an untrusted GitHub repository.
Evidence: Invoke-WebRequest -Uri "https://github.com/SteveTheKiller/KillerPDF/releases/latest/download/KillerPDF.zip" -OutFile "KillerPDF.zip" in SKILL.md.
[REMOTE_CODE_EXECUTION]: The skill explicitly directs the agent or user to execute a downloaded binary from an unverified source, which can lead to arbitrary code execution on the host machine.
Evidence: .\KillerPDF\KillerPDF.exe in SKILL.md.
[COMMAND_EXECUTION]: The skill includes shell commands for cloning repositories and building software from source, which involves executing code from external repositories.
Evidence: git clone https://github.com/SteveTheKiller/KillerPDF.git and dotnet publish -c Release in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external PDF data, which may contain hidden instructions that could manipulate the agent's behavior.
Ingestion points: The Open method in PdfDocument.cs reads PDF files from user-specified paths.
Boundary markers: There are no explicit delimiters or warnings to ignore instructions embedded within the PDF content during rendering or text extraction.
Capability inventory: The skill has the ability to read and write files, as well as execute shell commands as described in the setup instructions.
Sanitization: The GetPageText method uses PDFium to extract raw text, which is returned without any sanitization or escaping.

killerpdf-portable-editor