record-transactions

Pass

Audited by Gen Agent Trust Hub on Jul 9, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses specific files in the user's home directory to load firm profiles and plugin-specific configurations.\n
  • Evidence: Read access to ~/.claude/plugins/config/claude-for-accounting/firm-profile.md and ~/.claude/plugins/config/claude-for-accounting/{{plugin}}/CLAUDE.md.\n- [PROMPT_INJECTION]: The skill ingests untrusted financial documents, which serves as a surface for indirect prompt injection.\n
  • Ingestion points: Bank statements, sales invoices, purchase receipts, and payroll documents provided in the agent's context.\n
  • Boundary markers: No explicit delimiters or boundary markers are defined to separate untrusted document content from agent instructions.\n
  • Capability inventory: The skill is limited to producing structured data (CSV/JSON) and does not invoke network tools, shell execution, or file-writing capabilities.\n
  • Sanitization: The instructions include integrity constraints to prevent the agent from fabricating figures or interpolating missing data, which provides a basic level of defense against accidental hallucinations or obedience to data-embedded instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 9, 2026, 02:13 PM
Security Audit — agent-trust-hub — record-transactions