The Agent Skills Directory

Command Execution Patterns: The skill instructions recommend running terminal commands like waza init to initialize project structures. This is an intended function for a development assistant designed to automate setup tasks.
Indirect Prompt Injection Surface: The skill interacts with external skill files and YAML specifications to perform evaluations. This ingestion of potentially untrusted data represents a standard attack surface for testing and auditing tools.
Ingestion points: User-provided skill paths and YAML evaluation specifications (SKILL.md, Scenario 1).
Boundary markers: The instructions do not define specific delimiters for isolating external file content from agent instructions.
Capability inventory: The skill utilizes terminal commands and a set of MCP tools (e.g., waza_eval_run) for executing tests and interpreting results (SKILL.md).
Sanitization: There are no explicit instructions for sanitizing or filtering the content of the skills being evaluated.
MCP Tool Orchestration: The skill relies on specialized tools (e.g., waza_eval_list, waza_skill_check) to perform technical evaluation tasks. These tools are used within their intended context to provide feedback on skill compliance and performance.

waza-interactive