The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The file assets/FDA-Clinical-Protocol-Template.md contains a hardcoded link to http://www.manderson.org/departments/CIMER/. This domain is a known typosquat of the legitimate M.D. Anderson Cancer Center (mdanderson.org) and is explicitly blacklisted by automated scanners as a malicious host.
[COMMAND_EXECUTION]: The orchestration logic in references/04-protocol-operations.md uses unvalidated user-provided input to construct shell commands for running the Python sample size calculator (e.g., python scripts/sample_size_calculator.py --effect-size [user provided value]). This pattern facilitates command injection if the input contains shell metacharacters.
[PROMPT_INJECTION]: The skill processes untrusted data from external sources, creating an indirect prompt injection surface.
Ingestion points: references/01-research-protocols.md fetches data from the ClinicalTrials.gov MCP and various FDA databases.
Boundary markers: No explicit delimiters or warnings to ignore embedded instructions were found in the processing logic.
Capability inventory: The skill possesses subprocess execution capabilities (python) and extensive file-write access to the waypoints/ directory.
Sanitization: The instructions do not specify any sanitization or validation for the content retrieved from external databases before it is used to generate protocol sections.

clinical-trial-protocol-skill