The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes workspace source code to generate tests and determine execution steps, which creates a surface for instructions embedded within that code to influence agent behavior.
Ingestion points: The agent reads production modules, existing test files, and project manifests during its Research and Plan phases (SKILL.md, Step 4).
Boundary markers: The skill guidelines instruct the agent to separate metadata from data and use safe display identifiers to prevent unsafe data from being interpolated into test reports (unit-test-generation.prompt.md).
Capability inventory: The agent can execute shell commands (e.g., test runners like dotnet test, npm test, pytest), write files to the workspace, and delegate tasks to specialized sub-agents.
Sanitization: The skill emphasizes using proper format parsers for test results and avoiding the use of raw control characters or arbitrary inputs in test identifiers.
[COMMAND_EXECUTION]: The skill executes local development and system commands to manage state and verify generated tests.
Evidence: The skill instructs the agent to run "the narrowest relevant test command" and utilizes git rev-parse --path-format=absolute --git-path testagent to resolve a non-stageable scratch directory within the repository metadata for internal pipeline files.

code-testing