The Agent Skills Directory

Dynamic Source Code Modification and Execution: The skill instructs the agent to manually modify C# source files by adding attributes based on strings extracted from user requests (such as test names and issue URLs). Following these modifications, the skill executes dotnet build and dotnet test. This pattern of incorporating untrusted data into source code that is subsequently compiled and executed represents a security consideration regarding code integrity.
Shell Command Construction from User Input: Several shell commands, including grep, dotnet run, and dotnet test, are constructed using parameters (like test names and issue URLs) parsed directly from user input. This creates a potential surface for command argument injection if the input strings contain shell metacharacters or unexpected flags.
Processing of Untrusted Input (Indirect Prompt Injection): The skill is designed to ingest and parse natural language requests from users to identify actions, test names, and URLs. There is a consideration for indirect prompt injection because the skill does not specify boundary markers or sanitization steps to ensure that data within the user request (e.g., a maliciously crafted test name) is not interpreted as an instruction by the agent.
Ingestion points: User requests containing test names, issue URLs, and conditional logic (processed in SKILL.md).
Boundary markers: None specified in the instructions to separate data from instructions.
Capability inventory: Execution of grep, dotnet run (for local tooling), manual file modification of .cs files, dotnet build, and dotnet test (identified across various sections of SKILL.md).
Sanitization: The instructions focus on parsing and extraction but do not explicitly detail validation or escaping of the extracted data before it is used in shell commands or source code edits.

test-management