The Agent Skills Directory

[PROMPT_INJECTION]: The files ref/detection-patterns.md and references/threat-model.md contain many common prompt injection strings (e.g., 'ignore previous instructions', 'act as if', 'system prompt:'). These are clearly categorized as 'Detection patterns' or 'Attack vectors' for the auditor to identify in other skills. In the context of a security tool, these are reference data and do not represent a threat to the agent's own behavior.
[COMMAND_EXECUTION]: The SKILL.md file instructs the agent to execute a local Python script (scripts/skill_security_auditor.py) to perform the audit. This is the intended primary function of the tool and is triggered by specific security-related requests from the user.
[DATA_EXFILTRATION]: Documentation in the references/threat-model.md file lists sensitive file paths (such as ~/.ssh/id_rsa, .env, and ~/.aws/credentials) and network exfiltration patterns. These are documented as threats that the tool is designed to find in third-party skills, and the skill itself does not perform any unauthorized data access or transmission.

skill-security-auditor