The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains strong instructions to ensure formatting persistence ("These rules apply to every response for the rest of the session... do not expire"). These are scoped to output style and explicitly defer to system safety prompts and human-review for destructive actions, providing a safe operational framework.
[DATA_EXPOSURE]: Technical examples in the documentation (e.g., 'npm install jsonwebtoken', 'Authorization: Bearer ${token}') are provided as placeholders to illustrate formatting rules and do not involve actual credential access or exfiltration.
[COMMAND_EXECUTION]: The skill contains markdown examples of terminal commands (e.g., 'npm test'). These are instructional templates for the user's output and are not intended for autonomous execution by the agent. The skill also correctly sets 'disable-model-invocation: true' to limit its operational scope.

i-have-adhd