The Agent Skills Directory

[SAFE]: The skill is declared as a markdown-only extension with no scripts, network access, or credentials required for operation. Technical analysis of the provided instruction files confirms the absence of command execution, file system manipulation, or network-bound functionality within the skill itself.
[SAFE]: The skill implements a robust safety hierarchy where the agent's system prompt and safety guidelines explicitly outrank the skill's formatting rules. Override Rule 2 ('Destructive action ahead') specifically prevents the 'action-first' rule from bypassing safety confirmations for dangerous operations.
[SAFE]: The skill includes a 'Medical Boundary' rule that prohibits the agent from making diagnostic, screening, or treatment claims regarding ADHD, directing users to qualified clinicians for individual assessments instead.
[SAFE]: The skill's 'Pre-send Check' and 'Substance Guard' act as internal validation gates to ensure that attempts at concision do not result in the removal of critical safety information, technical caveats, or rollback instructions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data (repository content, blog posts, etc.) and includes evaluation cases (e.g., ae-source-injected-instruction) to test the agent's ability to recognize and refuse prompt injection attempts within that data. While this category identifies an attack surface, the skill includes proactive defenses and validation rules to mitigate the risk.

adhd-explainer