The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated scanners (URLite) have identified the documentation URL https://jeffallan.github.io/claude-skills/skills/data-ml/pandas-pro/ as malicious and blacklisted. Furthermore, the skill's main instruction file (SKILL.md) has been flagged as malware by reputation scanners.
[METADATA_POISONING]: There is a definitive mismatch between the skill's internal metadata, which claims the author is Jeffallan, and the authoritative skill author context which identifies midudev as the creator. This use of a non-vendor resource for documentation and authorship attribution is a strong indicator of deceptive intent or source impersonation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data via pandas DataFrames, which serves as an injection surface.
Ingestion points: Data input into DataFrames as described in SKILL.md and the reference guides.
Boundary markers: None; the skill does not provide instructions to sanitize or delimit potentially malicious text strings embedded within DataFrame values.
Capability inventory: The skill encourages the agent to execute complex Python logic (filtering, aggregation, type conversion) using the agent's interpreter, which can be influenced by the content of the data.
Sanitization: None present; data is processed directly via vectorized operations without validation of the content's safety.
[COMMAND_EXECUTION]: The skill provides numerous templates and patterns for local code execution based on processed data. While these are presented as standard data analysis workflows, the lack of input sanitization combined with the confirmed malware flags on the skill files creates a high risk that the agent may be induced to perform unsafe operations based on malicious input data.

pandas-pro