The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local script python scripts/generate_emergency_card.py to generate the health card output. This execution involves subprocess calls to process data according to user-defined variants (standard, child, elderly, severe) and sizes.
[DATA_EXPOSURE]: The skill accesses highly sensitive medical information across multiple files in the data/ directory, including profile.json, allergies.json, medications/medications.json, hypertension-tracker.json, diabetes-tracker.json, and copd-tracker.json. This includes PII, chronic disease status, and medication history.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from medical records, such as discharge summaries (data/出院小结/**/*.json) and surgical reports (data/手术记录/**/*.json), and interpolates this data into its output without explicit sanitization.
Ingestion points: JSON records containing diagnoses, procedure notes, and clinical summaries.
Boundary markers: The instructions do not define clear delimiters or "ignore instructions" warnings when processing the medical record content.
Capability inventory: The skill has capabilities for file reading (readFile), file globbing (glob), file writing (saveFile), and script execution (python).
Sanitization: There is no evidence of sanitization or schema validation for the notes or diagnoses fields extracted from external medical files before they are included in the generated cards.

emergency-card