The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill encourages the installation of multiple third-party global tools via the .NET CLI, which could lead to supply chain vulnerabilities if the packages are compromised.
Evidence: Instructions in references/slopwatch.md (dotnet tool install --global Slopwatch.Cmd), references/test-quality.md (dotnet tool install -g dotnet-reportgenerator-globaltool, dotnet tool install -g dotnet-stryker), and references/playwright.md (dotnet tool install --global Microsoft.Playwright.CLI).
[COMMAND_EXECUTION]: The skill suggests a configuration for the agent (.claude/settings.json) that executes a shell command automatically after every file edit. This creates a persistent execution hook in the agent's runtime environment.
Evidence: The Claude Code Hook Integration section in references/slopwatch.md provides a JSON snippet that triggers slopwatch analyze -d . --hook on Write, Edit, and MultiEdit actions.
[DYNAMIC_EXECUTION]: The skill includes a standalone Python script designed for CI pipelines and provides instructions for local browser driver installation via PowerShell scripts.
Evidence: references/ci-benchmarking.md contains the full source for compare-benchmarks.py. references/playwright.md includes commands to run playwright.ps1 from local build directories.
[INDIRECT_PROMPT_INJECTION]: The skill sets up an attack surface where the agent processes external data (web pages via Playwright) and untrusted source code (via Slopwatch analysis) without explicit sanitization or instruction boundary markers.
Ingestion points: references/playwright.md (page interactions and content retrieval), references/slopwatch.md (source code analysis).
Boundary markers: Absent; the agent is instructed to trust the output of these tools to make decisions about code quality.
Capability inventory: The skill enables subprocess calls (dotnet test, slopwatch), network operations (Playwright navigation, Aspire testing), and file system modifications (test scaffolding).
Sanitization: No evidence of sanitization for data ingested from the browser or analyzed files before it enters the agent's reasoning context.

dotnet-testing