Fin Audit Automator

This package implements local AML, invoice checks, and report generation, but several claims in the documentation are not supported by the code. Before installing or running it with sensitive data, consider: - Metadata mismatch: The package reads SKILLPAY_API_KEY and SKILLPAY_SKILL_ID from the environment even though no env vars were declared. If you don't want billing/telemetry keys available, do not set them. Ask the developer why these env vars are required and how they're used. - Local guarantees are weaker than advertised: The sandbox only clears proxy env vars and does not truly block network sockets or enforce a whitelist. Do not assume the code is network-isolated — run it in an isolated environment (VM, container, or an air-gapped host) if you need strong isolation. - Audit log is not truly immutable: The 'tamper-proof' log is a simple local file with per-line hashes; anyone with file access can truncate or replace it. If immutable logs are required, use an external append-only store or OS-level protections (WORM storage, append-only mounts). - Persistent files: The skill creates ~/.openclaw/… files for logs and trial state. Review those files, set appropriate file permissions, or run in a controlled account to avoid exposing sensitive data. - Dependency bloat: package.json lists libraries (sqlalchemy, cryptography, jieba) not used in code. That increases attack surface. Request a trimmed dependency list or scan those packages before installing in production. - Code review recommendations: Ask the author to (1) declare required env vars and config paths in metadata, (2) implement real network whitelisting or document precise network allowances, (3) provide a clear privacy/billing flow (what SKILLPAY keys do), and (4) harden the audit log or integrate with an established tamper-evident store. If you cannot get satisfactory answers, run the skill in an isolated sandbox and avoid feeding production PII/financial data.