The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process and analyze medical chart data (patient notes, assessments, and plans). This creates a theoretical surface for indirect prompt injection if malicious instructions were embedded in the processed medical records. However, the skill explicitly mandates a human-in-the-loop review process and the use of structured provenance (source spans and character offsets) to verify all model outputs against the source text. This multi-layered approach to validation significantly mitigates the risk of an agent following instructions embedded in external data.
[EXTERNAL_DOWNLOADS]: The skill mentions common and well-known clinical NLP libraries such as medspaCy, scispaCy, and Apache cTAKES. It also provides an installation command using the platform's standard 'skills.sh' registry. All referenced external resources are standard industry tools for the stated purpose of the skill.
[REMOTE_CODE_EXECUTION]: No patterns of unauthorized or suspicious remote code execution were found. The skill does not contain scripts or commands that download and execute arbitrary payloads from the internet.
[DATA_EXFILTRATION]: There is no evidence of unauthorized data access or exfiltration. The skill contains only synthetic medical data for testing purposes and includes strict instructions for the agent to verify HIPAA compliance and data de-identification before processing any real content.
[PROMPT_INJECTION]: The skill's instructions (SKILL.md) are well-structured with clear safety gates and do not contain any patterns attempting to override the agent's core safety guidelines or bypass security filters.

hcc-nlp