The Agent Skills Directory

[DATA_EXPOSURE]: The skill reads configuration and practice-area specific guides from the ~/.claude/plugins/config/ directory. This access is scoped to the tool's own configuration environment to determine supervision styles, jurisdiction, and intake templates.
[COMMAND_EXECUTION]: The skill generates /legal-clinic:deadlines --add command blocks based on information gathered during the intake. These are intended for user copy-paste into a related tool and are not executed automatically by the skill itself.
[PROMPT_INJECTION]: The skill processes untrusted client narratives which could theoretically contain instructions intended to manipulate triage or conflict flags (Indirect Prompt Injection). However, the skill effectively mitigates this risk through strict output formatting, mandatory attorney review banners, and specific verification checklists for the human user.
[SAFE]: The workflow follows a 'human-in-the-loop' design where the AI acts as a transcription and structuring assistant rather than a decision-maker. It explicitly disclaims the ability to decide case acceptance or resolve conflicts, maintaining professional and safety boundaries.

client-intake