The Agent Skills Directory

[PROMPT_INJECTION]: The instructions focus exclusively on clinical study design and medical imaging validation. No patterns associated with jailbreaking, instruction overriding, or bypassing safety filters were detected.
[DATA_EXFILTRATION]: Analysis of the skill's logic and scripts confirms that no sensitive data is transmitted externally. The skill does not contain hardcoded API keys or network-related commands such as curl or wget for data transfer.
[REMOTE_CODE_EXECUTION]: The skill utilizes a local Python script for data auditing. This script relies solely on standard Python libraries (csv, json, argparse) and does not download or execute code from external sources.
[OBFUSCATION]: All files were inspected for hidden payloads, including Base64 encoding, zero-width characters, and homoglyphs. No obfuscation techniques intended to conceal malicious logic were identified.
[COMMAND_EXECUTION]: Command usage is restricted to running the internal audit script. There is no evidence of arbitrary shell command execution, privilege escalation attempts, or persistence mechanisms.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data in CSV format to audit model splits. This process is handled by a deterministic script that performs set-based validation, effectively isolating the agent from untrusted content in the data file.

model-validation