The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses strong instructional language such as "STOP
The Obviousness Rule" and defines its own "Severity Guide" including "CRITICAL" labels. These are identified as pedagogical guidelines intended to steer the agent's auditing logic rather than attempts to bypass the underlying LLM's safety filters or override system instructions.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process untrusted external data (source code, README files, and AI configuration files). This represents a standard attack surface; however, the skill lacks high-privilege capabilities—such as network exfiltration or arbitrary command execution—that would be required to exploit this surface maliciously.
[SAFE]: The skill does not perform any network operations, use external dependencies, or attempt to execute remote scripts. It relies entirely on local instructions and a companion checklist file provided within the skill package.

code-clarity-and-docs