The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes standard system diagnostics tools (find, grep, lsof, pgrep, npm, pip) to identify installed package versions and running services. These are strictly informational and diagnostic, with a hard rule against state-changing operations or system modifications.
[DATA_EXPOSURE]: The skill accesses local configuration files such as ~/.zshrc and ~/.config to check for environment-based exposure. This potential data exposure is mitigated by explicit instructions to never dump credentials or raw environment lines, instead reporting only sanitized versions or booleans.
[INDIRECT_PROMPT_INJECTION]: The skill processes security advisories from external sources, which represents an injection surface. This is addressed by specific defensive instructions, including the use of filename-safe NUL transport and an explicit ban on substituting untrusted package names directly into shell syntax.
[SAFE]: The skill implements a robust 'least privilege' model for its stated purpose, incorporating safety guardrails that prevent common AI agent attack vectors such as credential exfiltration and privilege escalation.

cyber-audit