The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content by reading source code and git diffs from the local repository (using Grep, Glob, and Read) to derive test scenarios. There is a potential attack surface where malicious comments in the codebase could attempt to influence the agent's output, though this is a standard risk for code-analysis tools.
Ingestion points: Git diff output, source code files, README, and configuration files within the repository.
Boundary markers: None explicitly defined to separate code content from instructions.
Capability inventory: File writing (Write), command execution (Bash), and file reading (Read).
Sanitization: The instructions recommend slugifying names for filenames, but do not specify sanitization for the content extracted from the code into the HTML guide.
[COMMAND_EXECUTION]: The skill is granted permission to use Bash for specific commands including git and system file openers (start, explorer.exe). These are used for their intended purposes: identifying changes and opening the generated HTML report.

create-manual-test-guide