The Agent Skills Directory

[PROMPT_INJECTION]: The phrase "ignore previous instructions" was identified in the text; however, it is used defensively. The skill instructs the agent to treat any such string found in audited metadata as a security finding to be reported, rather than an instruction to be followed.
[SAFE]: The skill restricts its operational capabilities to read-only tools (Read, Grep, Glob) and provides explicit 'Injection Hardening' guidelines to prevent the agent from executing commands or exfiltrating data discovered during audits.
[SAFE]: All external references are to established security standards and organizations (NIST, CIS), and the skill's purpose matches its implemented behavior.

access-review