The Agent Skills Directory

[SAFE]: The skill implements strong quality control instructions (referred to as 'IRON LAWS') which require the agent to self-correct, delete, or restart sections that fail to meet academic standards. These instructions are focused on scholarly rigor rather than bypassing platform safety filters.
[SAFE]: The included Python script (references/volokh-distilled.py) functions as a static linter to identify stylistic issues like nominalization, legalese, and harsh criticism. It operates locally on project files within defined directories (drafts/, outlines/) and performs no network operations or sensitive system access.
[SAFE]: Document generation is managed via a structured 'gate function' that requires the use of the /docx skill and a local Word template (templates/law_review_template.docx). This process ensures that output is correctly formatted without relying on external downloads or untrusted scripts.
[SAFE]: File access is restricted to internal skill resources and project-specific planning metadata (e.g., .planning/ACTIVE_WORKFLOW.md). No patterns of data exfiltration or credential harvesting were identified.
[SAFE]: While the skill processes user-provided drafts, which is an inherent surface for indirect prompt injection, it contains extensive instructions for the agent to scrutinize and restructure input according to logical and scholarly principles, serving as a functional barrier against unintentional obedience to embedded instructions.

writing-legal