The Agent Skills Directory

Local File System Access: The skill is designed to list and read files within a specified directory and search the codebase using tools like grep. This access is necessary for its core purpose of detecting documentation-to-code drift and ensuring technical consistency across the project.
Indirect Prompt Injection Surface: The skill processes user-provided design documents and READMEs as data. Because these files are treated as input for the agent's analysis, there is a potential surface for indirect prompt injection if those files were to contain malicious instructions intended to influence the agent's behavior. The skill's specific instructions to report only differences and the use of the disable-model-invocation flag are positive security considerations that help limit the impact of such patterns.
Ingestion points: Design documents and READMEs in user-specified directories (SKILL.md).
Boundary markers: None explicitly defined in the prompt instructions.
Capability inventory: Reading files, listing directories, code search/grep, and writing a local report file (SKILL.md).
Sanitization: No explicit sanitization or escaping of file content is mentioned before processing.
Environment Variable Reference: The instructions refer to environment-specific paths such as ${CLAUDE_PLUGIN_ROOT} to find documentation. This is a standard way to provide context in specific development environments and does not represent a security risk, though it assumes the presence of these variables in the execution context.

egm-drift-report