The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions for generating schematics (python scripts/generate_schematic.py "your diagram description" -o figures/output.png) describe a process where user-controlled natural language descriptions are directly interpolated into a shell command. This pattern is vulnerable to command injection if the user input contains shell metacharacters like backticks, semicolons, or pipes.
[METADATA_POISONING]: The skill's YAML frontmatter identifies 'K-Dense Inc.' as the author, which contradicts the provided authoritative author context of 'wu-yc'. This discrepancy in attribution can be misleading regarding the origin and trustworthiness of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted external data such as clinical trial subgroups, EMR data, and real-world evidence to generate reports.
Ingestion points: SKILL.md describes processing 'RWE Cohort Studies' from 'EMR data' and 'trial subgroups'.
Boundary markers: The instructions do not specify the use of delimiters or boundary markers (e.g., XML tags or clear-text separators) when handling this external data within prompts.
Capability inventory: The skill is granted Bash, Read, Write, and Edit tools, which could be abused if an injected instruction is followed.
Sanitization: While the skill emphasizes HIPAA de-identification for privacy, it lacks instructions for sanitizing or escaping the content of external data to prevent prompt injection.

clinical-decision-support