The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and analyze external, untrusted content including discovery documents, patent prior art, accused product manuals, and scraped web pages. This creates a surface where embedded instructions in those documents could potentially influence the agent's behavior.
Mitigation: The skill implements a robust "apostrophe-prefix neutralization" technique when generating CSV and Excel files. By prepending an apostrophe to values starting with characters like =, +, or -, it prevents the spreadsheet application from executing those values as formulas (CSV Injection), which is a common vector for data exfiltration or code execution in this context.
Sanitization: The instructions explicitly require verbatim quoting and provide clear guidance on mapping elements strictly against provided evidence, reducing the likelihood of the agent hallucinating or following instructions embedded within the analyzed text.
Sensitive Path Access (Configuration): The skill accesses files within ~/.claude/plugins/config/claude-for-legal/. While these paths are in the user's home directory, they appear to be standard configuration and state management files specific to this legal plugin suite. The skill uses these paths to maintain matter history and practice-level configurations, representing expected behavior for a tool of this nature.
Defensive Design: The skill includes mandatory "Conflicts gates" and "Non-lawyer gates" that require specific intake procedures and surface warnings to users. These architectural constraints help ensure the tool is used within its intended legal and safety framework.

claim-chart