The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation (SKILL.md) and reference files (atlas-coverage.md) contain numerous examples of prompt injection and jailbreak techniques, such as 'DAN mode' and 'system prompt extraction'. These patterns, including those in the ai_threat_scanner.py seed prompts, are legitimate test signatures and documentation for the skill's primary purpose of AI security assessment.
[DATA_EXFILTRATION]: No network operations, credential harvesting, or unauthorized sensitive file access patterns were identified. The scanner script operates entirely on local inputs.
[REMOTE_CODE_EXECUTION]: The skill does not contain patterns for downloading or executing remote code. The ai_threat_scanner.py script uses standard Python libraries to perform local regex matching and does not employ dynamic execution functions like eval() or exec() on untrusted data.

ai-security