The Agent Skills Directory

Command Construction via User Input Interpolation: In references/04-protocol-operations.md, the skill instructs the agent to execute a shell command (python scripts/sample_size_calculator.py) using arguments directly supplied by the user during the interaction. Constructing shell commands via string interpolation of untrusted input is a security consideration, as it could potentially lead to command injection if the input is not strictly validated.
Processing of User-Provided Templates: The skill provides an option for users to upload and use their own protocol templates. According to references/02-protocol-foundation.md, the agent is instructed to read these files in their entirety and follow their structure and guidance. This creates a surface for indirect prompt injection, where a file could contain instructions designed to influence the agent's output or behavior.
Ingestion points: User-provided template files (.md format) and interactive user input for statistical parameters.
Boundary markers: None identified in the instructions for separating untrusted template content from agent instructions.
Capability inventory: The skill possesses file-writing capabilities to the waypoints/ directory, network access via the ClinicalTrials.gov MCP server and official FDA database URLs, and the ability to execute shell scripts.
Sanitization: No explicit sanitization or validation steps for user-provided input or template content were found in the provided instructions.
External Package Dependencies: The SKILL.md file recommends installing external Python libraries (scipy >= 1.11.0 and numpy >= 1.24.0) via pip. While these are standard, well-known libraries, the installation of third-party dependencies at runtime is a potential consideration for secure execution environments.
Automated Reputation Alert: An automated scanner flagged SKILL.md as having a poor reputation (FileRepMalware). While a manual review of the instructions does not identify specific risky code patterns, such alerts suggest that the command patterns or file structure may warrant additional oversight.

clinical-trial-protocol-skill