The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a vendor-specific CLI tool (pdfcraft-cli) and a local Python helper script to manage PDF watermarking tasks.
[DYNAMIC_EXECUTION]: The scripts/write_script.py utility dynamically generates a sequence of tool calls in JSON format based on a recipe file. This script includes safety checks, such as validating document slot indices and using atomic file creation mode (x) to avoid accidental data loss or unauthorized path access.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied PDF documents, which creates a potential surface for indirect instructions.
Ingestion points: PDF content and metadata are accessed through the doc_open and doc_info tools as defined in the skill instructions.
Boundary markers: The instructions explicitly direct the agent to distinguish between template samples and actual user data, and provide clear warnings that watermarks are not suitable for security redaction.
Capability inventory: The skill possesses file-writing capabilities through the doc_save tool and can execute local scripts.
Sanitization: The write_script.py component performs integer validation for document identifiers to ensure operations remain within the scope of the opened documents.
[SAFE]: The skill references the well-known py-pdf/pypdf GitHub repository to document workflow facts. This is a neutral reference to an established library for educational purposes and does not involve runtime code execution or downloading from untrusted sources.

pdfcraft-review-watermark