The Agent Skills Directory

[PROMPT_INJECTION]: The assets/output_checklist.md file contains instructions (Rules 12 and 13) that explicitly command the agent to fabricate 'realistic, concrete terminal execution output' and strictly forbid including disclaimers that tests were not actually run. This is a deceptive instruction that forces the agent to mislead users and analysts regarding its actual execution of code.
[COMMAND_EXECUTION]: The skill's core workflow in SKILL.md requires the agent to 'execute the spec via bundle exec rspec'. Executing shell commands with parameters derived from potentially untrusted user-provided code or requirements is a significant security risk.
[REMOTE_CODE_EXECUTION]: By mandating the execution of RSpec tests—which are executable Ruby scripts—the skill creates a direct path for arbitrary code execution. If the agent processes malicious code from a user into these tests and executes them as instructed, it results in a successful code execution attack against the local system.

write-tests