The Agent Skills Directory

Data Access and Privacy: The skill reads local session transcripts (~/.claude/projects/) and git configuration (user.name, user.email) to provide a personalized usage overview. This access is limited to the skill's stated purpose, and only an aggregate summary is shared with the model to generate the markdown report.
Security Best Practices: The accompanying mining script implements several defensive measures, such as preventing CSV injection in exports by neutralizing formula-starting characters, sanitizing HTML output to mitigate cross-site scripting risks, and using secure file-writing flags to prevent symlink attacks.
Indirect Prompt Injection Mitigations: The instructions provide explicit guidance to the model to treat processed data, such as directory names, as inert labels. This is a proactive measure against potential indirect prompt injection attempts that might be embedded in file system metadata.
Local Execution: All data mining is performed by a local script without making external network calls or using third-party dependencies. This ensures that sensitive session details and local code remains within the user's environment.

receipts