The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted source code from the workspace to generate unit tests, which introduces an indirect prompt injection surface. Malicious instructions embedded within the source code (e.g., in comments or documentation) could influence the agent's behavior during the research, planning, or implementation phases.
Ingestion points: Source code files and project metadata analyzed during the research phase in SKILL.md and unit-test-generation.prompt.md.
Boundary markers: Absent; there are no explicit instructions to the agent to ignore or delimit instructions found within the code being processed.
Capability inventory: The skill has capabilities to write files, generate code, and execute shell commands for building and testing.
Sanitization: Absent; the agent is instructed to perform deep analysis of the code logic.
[DYNAMIC_EXECUTION]: The skill generates and executes code as part of its core functionality. It creates unit tests in various languages and then runs them using local test runners such as xUnit, pytest, or Jest.
Execution method: Invoking test runner CLI tools on generated files.
Risk factor: Execution of generated content is a high-capability operation that depends on the integrity of the generation process.
[COMMAND_EXECUTION]: The skill uses shell commands to analyze the environment and manage state.
Patterns: It uses git rev-parse to find a non-stageable directory for state files and uses framework-specific commands for validation.
State management: It creates intermediate files (research.md, plan.md, status.md) in a directory that may be located within Git metadata paths to ensure they are not accidentally committed.

code-testing-agent