The Agent Skills Directory

[SAFE]: The skill establishes a mandatory analysis phase using the 5W1H decomposition method, which prevents the agent from executing commands before scope is fully defined. This behavioral structure significantly reduces the risk of unintended actions resulting from ambiguous instructions.
[COMMAND_EXECUTION]: The skill instructions direct the agent to infer project details, such as the technology stack or directory structure, from the local environment. Although tools like Bash and Grep are allowed in the configuration, their usage is constrained to local context gathering for the clarification process and does not involve executing external or untrusted code.
[DATA_EXFILTRATION]: No network access, remote connections, or patterns for harvesting sensitive files (e.g., SSH keys or environment secrets) are present in the skill instructions.
[PROMPT_INJECTION]: The skill contains no instructions aimed at overriding agent behavior, bypassing safety guardrails, or extracting the underlying system prompt. The 'Absolute Rules' provided are strictly functional constraints for the requirements gathering task.

swing-clarify