The Agent Skills Directory

External Library Dependency: The skill utilizes the Microsoft.CodeAnalysis.CSharp and tree-sitter-language-pack libraries for parsing source code. These are established, widely-used libraries for static analysis. The C# engine specifically uses a library from the vendor (Microsoft), which is a trusted source.
Indirect Prompt Injection Surface: The tool parses source files within a repository to identify declarations. This represents a potential surface where specially crafted symbol names in a repository could attempt to influence the agent's logic when it consumes the analysis report.
Ingestion points: Local source files (.cs, .py, .ts, etc.) within the target repository root, read by Find-UntestedSources.cs and find_untested_sources.py.
Boundary markers: The skill outputs a structured JSON report, which provides a clear boundary between the extracted code data and the agent's processing instructions.
Capability inventory: The tool is restricted to read-only access within the designated repository root; it does not have capabilities for network exfiltration, filesystem writes, or subprocess execution.
Sanitization: The scripts extract only specific identifier tokens and path metadata into JSON, minimizing the risk of instructional content influencing the agent.
Secure Directory Pruning: The analysis engines include logic to explicitly skip sensitive directories such as .git, .vs, and node_modules, as well as build artifacts in bin and obj folders. This follows best practices for repository scanning and prevents the exposure of internal metadata.

find-untested-sources