The Agent Skills Directory

External Downloads (MEDIUM): Documentation in releases/README.md and QUICKSTART.md provides instructions and wget commands to download installation ZIP archives from github.com/omkamal/pypict-claude-skill, which is not a recognized trusted source.- Unverifiable Dependencies (MEDIUM): The scripts/README.md file recommends installing the pypict Python package via pip. This package is a third-party binding and is not maintained by a trusted organization, posing a supply chain risk. The use of the --break-system-packages flag is also highlighted, which overrides system-level safety protections on certain platforms.- Indirect Prompt Injection (LOW): The skill is designed to ingest and analyze untrusted external specifications (e.g., atm-specification.md) to generate test plans, creating a surface for indirect instructions to influence agent behavior. 1. Ingestion points: Specification files and user-provided requirements. 2. Boundary markers: None identified in provided documentation. 3. Capability inventory: scripts/pict_helper.py performs local file reads (open()). 4. Sanitization: None identified in the provided scripts.- Data Exposure (INFO): The scripts/pict_helper.py utility can read arbitrary local files if their paths are passed as command-line arguments. While this is the intended functionality for parsing models, it constitutes a file-access capability.

pict-test-designer