The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses dynamic context injection via a benign shell command (git branch) in SKILL.md to identify the current branch. This is a standard practice for development tools and does not involve exfiltration or unsafe argument injection.\n- [PROMPT_INJECTION]: The skill design follows adversarial validation principles, using fresh-context agents and withholding rationales to prevent bias. This is a defensive prompt engineering pattern.\n- [DATA_EXFILTRATION]: Local project files like PLAN.md and interview-checklist.md are accessed for auditing. No network exfiltration or transmission to untrusted domains is present.

audit-answers