The Agent Skills Directory

Processing of External Content: The skill analyzes user-provided source code to generate unit tests, which introduces an indirect prompt injection surface. The skill includes specific instructions to mitigate this risk, particularly concerning how test results are reported to prevent exploitation of downstream tools.
Ingestion points: The agent reads project source code, manifests, and symbol references during the Research phase as described in SKILL.md.
Boundary markers: The skill references a 'Report-safe test names and result validation' protocol which provides guidance on handling potentially unsafe data in unit-test-generation.prompt.md.
Capability inventory: The agent has the capability to write files (test files and state files) and execute shell commands (test runners and build tools) as part of its testing workflow.
Sanitization: The 'Report-safe' contract requires separating metadata from data and escaping unsafe characters to ensure that generated identifiers do not disrupt test runners or reporting tools.
Local Command Execution: The skill executes standard development commands such as dotnet test, pytest, and npm test to validate code behavior. These operations are essential for the skill's primary purpose and are performed within the local workspace context.
File System Operations: The skill reads codebase artifacts and writes generated test files. It also manages internal state by creating temporary files (research.md, plan.md, status.md) in a directory resolved via Git metadata or system temporary paths, which helps maintain workspace cleanliness and prevents internal agent data from being committed to version control.

code-testing