The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains reference documentation in references/ai-agent-security.md that lists various prompt injection techniques, such as 'Ignore all previous instructions' and DAN-style jailbreaks. These are clearly categorized as examples of attacks that the auditor should recognize and do not represent an attempt to hijack the agent's actual behavior.
[COMMAND_EXECUTION]: The skill implements its auditing features through a series of Python scripts (e.g., full_audit.py, secrets_scanner.py, injection_scanner.py). The instructions in SKILL.md direct the agent to execute these local scripts to perform analysis. The scripts are self-contained and primarily rely on the Python standard library.
[DATA_EXPOSURE_AND_EXFILTRATION]: A core component of the skill is detecting hardcoded credentials, API keys, and private IPs. The scanning logic operates locally, identifying these items to alert the developer for remediation. There is no evidence of logic that transmits collected data to external servers or non-whitelisted domains.

007