The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill directs the user to install a WinDbg MCP server directly from a third-party GitHub repository (github.com/svnscha/mcp-windbg) using the uvx tool. This involves executing code from an unverified source that is not maintained by a recognized trusted organization.
[PRIVILEGE_ESCALATION]: The documentation frequently specifies the use of sudo for high-privilege operations, such as collecting core dumps with createdump or attaching debuggers like lldb to running processes. These actions grant the agent or user full access to the target process's memory space and potentially sensitive host system resources.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret complex external data (crash dumps, stack traces, and debugger output) which could be used as an injection vector to influence agent behavior.
Ingestion points: Process memory dumps and live debugger console output processed via dotnet-dump, lldb, and the WinDbg MCP tools.
Boundary markers: None; the skill does not specify how the agent should distinguish between its primary instructions and strings found within raw memory data.
Capability inventory: The skill utilizes tools capable of full file system access, process inspection, and memory reading.
Sanitization: There are no explicit instructions to sanitize strings or stack frames retrieved from the target process before analysis.
[EXTERNAL_DOWNLOADS]: Configures symbol retrieval from well-known services, specifically the official Microsoft symbol server and the NuGet symbol registry, to provide meaningful stack traces.

dotnet-debugging