The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. As a framework for auditing codebases, the skill is designed to process untrusted data. Malicious patterns in the audited code could attempt to influence the agent's coaching logic or classification process.
Ingestion points: Audited project files accessed via the Read, Grep, and Glob tools.
Boundary markers: Findings are required to follow a strict YAML schema with specific axis labels and audience tags.
Capability inventory: The skill is restricted to file system reading and searching tools (Read, Grep, Glob) and does not have native network access.
Sanitization: The skill implements a 'Schema Gate' that uses regular expressions to validate that file citations in the better_approach field are correctly formatted.
[SAFE]: Informative External Reference. The radar-suite-core.md file suggests the installation of an external tool called 'Dippy' from a public GitHub repository (ldayton/Dippy) to assist with path-handling in shell commands. This is presented as an instruction to the user rather than an automated execution.
[SAFE]: Least Privilege Tooling. The skill's YAML frontmatter restricts allowed-tools to standard file system operations, preventing the framework from executing arbitrary code or performing unauthorized network activity.

radar-suite-axis-classification