The Agent Skills Directory

Workspace Boundary Enforcement: The skill incorporates a specific workflow to ensure all file discovery and reading operations are restricted to canonical paths within the project workspace. This mitigates risks related to directory traversal or access to sensitive files outside the intended scope.\n- Diagnostic Tool Usage: The skill uses standard environment tools like cat, sed, or Get-Content solely for reading file content. This is a common and appropriate pattern for a code analysis tool and does not present a significant security risk when constrained to the workspace.\n- Indirect Prompt Injection Surface: The skill is designed to analyze untrusted source code, which inherently presents a potential surface for indirect prompt injection.\n
Ingestion points: The skill inventories and reads test files and production code symbols as part of its primary analysis workflow.\n
Boundary markers: Step 1 specifies a requirement to normalize paths and verify them against workspace boundaries, stopping immediately if any security policy or workspace violation is detected.\n
Capability inventory: The skill is limited to read-only file system operations and does not possess capabilities for network communication, arbitrary code execution, or writing files.\n
Sanitization: The output is constrained to a diagnostic report generated based on an internal catalog of anti-patterns, which helps prevent malicious code content from influencing external systems or the agent's core instructions.

test-anti-patterns