The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a 'Repository overlay' mechanism that directs the agent to load and apply instructions from a file located at .agents/skill-overlays/dotnet-test/run-tests.md within the analyzed repository. * Ingestion points: Instructions are ingested from the file path .agents/skill-overlays/dotnet-test/run-tests.md in the repository root. * Boundary markers: While the skill validates specific YAML frontmatter fields (core, binding-revision, mode), it does not establish a security boundary to prevent the overlay from overriding core logic with malicious instructions. * Capability inventory: The skill possesses the ability to execute shell commands, build tools (dotnet, MSBuild), and repository scripts. * Sanitization: The skill lacks sanitization of the content provided by the overlay file before it influences agent behavior.
[COMMAND_EXECUTION]: The instructions explicitly tell the agent to prioritize executing scripts found within the repository to maintain compatibility with the project's test system. * Evidence: The workflow for classic projects states, 'Prefer a checked-in script or documented CI command.' * Risk: This pattern may lead the agent to execute untrusted or malicious scripts provided by an attacker-controlled repository under the context of running tests.

run-tests