The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted PDF documents to distill writing patterns, creating an attack surface where malicious instructions in a paper could influence the agent's behavior.
Ingestion points: PDF files are processed by scripts/convert.py and the resulting text is analyzed by the agent in SKILL.md (Workflow Step 3 and 4).
Boundary markers: The workflow does not specify the use of clear delimiters or instructions to the agent to disregard embedded commands when analyzing the extracted text.
Capability inventory: The skill has the ability to write files to the project directory and modify library index files via scripts/convert.py and the instructions in SKILL.md.
Sanitization: While the extraction script cleans formatting (ligatures and newlines), it does not sanitize the text for potentially malicious natural language instructions.
[DYNAMIC_EXECUTION]: The script scripts/convert.py uses the __import__ function to dynamically load the pymupdf library at runtime.
Evidence: __import__('pymupdf') is used in the _check_pymupdf() function to verify dependency availability before proceeding with extraction.
[COMMAND_EXECUTION]: The skill requires the execution of a Python script that interacts with the local file system.
Evidence: The write_current function in scripts/convert.py uses NamedTemporaryFile and os.replace to atomically write markdown cards to the disk, and the SKILL.md workflow instructs the agent to run this script.

ccf-paper-to-exemplar