The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/skill_security_auditor.py utilizes subprocess.run to execute git clone commands. This functionality is essential for its stated purpose of auditing skills hosted in remote Git repositories and is triggered only by user input.
[PROMPT_INJECTION]: The skill's documentation (SKILL.md) and threat model (references/threat-model.md) include numerous examples of prompt injection strings (e.g., 'Ignore previous instructions'). These are provided for user education and as test cases for the auditor tool.
[EXTERNAL_DOWNLOADS]: The auditor supports cloning external Git repositories for the purpose of pre-installation security reviews. This behavior is transparently documented as a core feature of the utility.
[REMOTE_CODE_EXECUTION]: The references/threat-model.md file contains examples of remote code execution patterns, such as piping curl output to bash. These are strictly documentation examples intended to illustrate common attack vectors and are not present in any executable script within the skill.
[DATA_EXFILTRATION]: The skill contains regex patterns designed to detect exfiltration attempts (e.g., searching for requests.post to unknown domains). The tool itself does not perform any unauthorized data transmission.

skill-security-auditor