kelly-legal-casebase-ingest
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes legal documents (judgments and awards) to extract structured facts and holdings, which presents a surface for indirect prompt injection if source documents contain hidden instructions designed to manipulate the agent's extraction behavior.
- Ingestion points: Data extracted from external source documents enters the agent's context through
scripts/ingest_documents.mjs. - Capability inventory: The skill possesses capabilities to write to the Busabase backend through
scripts/execute_decisions.mjsand perform local file exports viascripts/export_case_records.mjs. - Sanitization: The web UI implements
escapeHtmlto sanitize document-derived content before it is displayed in the reviewer workbench. - Boundary markers: The skill architecture enforces a mandatory human-in-the-loop review gate for every ingested record and uses deterministic QA checks to validate anonymization and taxonomy coverage.
Audit Metadata