The Agent Skills Directory

[Indirect Prompt Injection]: The skill processes findings and knowledge base entries stored in the workspace. If these artifacts contain content influenced by untrusted target code, they could serve as a vector for indirect prompt injection. This is a known risk for analysis tools and is mitigated here by structured JSON processing.
[Workspace Management]: The skill implements complex logic for managing vulnerability findings, including snapshot-based filtering and signature-keyed deduplication. These operations are restricted to the workspace and archive directories, ensuring data integrity across different analysis passes.

mantis-chain