The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill includes a documentation link that points to a domain flagged as malicious by automated security scanners. Interaction with blacklisted domains presents a high risk of malware delivery or remote code execution targeting the agent or the user's environment.
File: SKILL.md
Evidence: https://jeffallan.github.io/claude-skills/skills/data-ml/pandas-pro/
[EXTERNAL_DOWNLOADS]: The skill reference files incorporate dependencies on external Python libraries, including pandera, numba, and pyarrow. While these are common packages, they represent an expanded attack surface through external code dependencies.
Files: references/data-cleaning.md, references/performance-optimization.md
[INDIRECT_PROMPT_INJECTION]: The skill is designed to handle and process DataFrames which are often loaded from untrusted external sources, creating a surface for indirect prompt injection. The instructions lack explicit guardrails for sanitizing or isolating untrusted data content before processing.
Ingestion points: DataFrame processing throughout all reference documentation.
Boundary markers: Absent.
Capability inventory: Data manipulation, memory profiling, and type conversion capabilities.
Sanitization: Absent.

pandas-pro