The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions require the installation of the pyHanko library from the Python Package Index (PyPI) to facilitate PDF validation.
[METADATA_POISONING]: The LICENSE file contains inconsistent metadata that does not match the skill's primary documentation. It refers to the project as "doc-intel-llm-tier-pl" and lists "contextgem" as a runtime dependency, whereas SKILL.md identifies the skill as "waliduj-podpis-pdf-pl" and utilizes pyHanko.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external PDF files, creating a surface for indirect prompt injection if the agent interprets extracted metadata as instructions.
Ingestion points: The waliduj_podpis.py script reads user-provided PDF files from the local file system.
Boundary markers: The validation results are returned as structured JSON, which helps delineate data fields for the agent.
Capability inventory: The skill utilizes Bash for execution and Read for file access. It does not perform network operations (explicitly disabled in the cryptographic context).
Sanitization: The script extracts specific metadata fields (such as signer identity and certificate serial numbers) and serializes them using json.dumps. There is no specific filtering of the content within these strings to prevent potential prompt injection markers from reaching the agent.

waliduj-podpis-pdf-pl