The Agent Skills Directory

[SAFE]: The skill's instructions focus exclusively on a conversational requirement-gathering process ('grilling') to generate workflow specifications.
[DATA_EXPOSURE]: The skill interacts with local project files such as NOTES.md and the workflows/ directory. This behavior is restricted to the local workspace and is consistent with the skill's primary function of documenting user requirements.
[PROMPT_INJECTION]: The instructions establish a specific persona for the interaction but do not contain attempts to override safety filters or extract system prompts.
[COMMAND_EXECUTION]: The skill configuration includes disable-model-invocation: true, which effectively limits the agent's ability to invoke external tools or execute commands during the session.

loop-me