Invoices

Capture, extract, and organize received invoices with automatic OCR, provider detection, and searchable archive.

This skill appears internally consistent with its invoice-management purpose, but review these before enabling: - Storage and backups: the skill stores originals and derived files under ~/invoices/; confirm you are comfortable with that path, ensure backups are encrypted and retention policies meet your legal needs. - Email and portal access: email monitoring and URL downloads are optional but require you to provide credentials or connectors. Only configure these if you trust the agent and understand which account it will access. - OCR and VAT validation: the docs mention a 'vision model' and the VIES API but do not state whether processing is local or sent to third-party services. Ask (or test) where OCR and VIES requests are executed; sending invoice images externally can leak sensitive data. - Test with non-sensitive data: before putting real or confidential invoices through automation, run a trial with redacted or dummy invoices to verify naming, deduplication, validation logic, and data flows. - GDPR/PII: invoices may contain personal data. Use redaction or limit exports, and confirm export destinations are compliant with your jurisdictional rules. - Retention and deletion: the skill's rule is to never delete originals; if you need different retention, prepare a manual policy or modify behavior before bulk import. If you need the skill to access email/remote services, require explicit, auditable configuration and verify that network calls (OCR/VIES) go to trusted endpoints. Otherwise the skill is coherent and appropriate for the stated purpose.

Type: OpenClaw Skill Name: invoices Version: 1.0.1 The skill is designed for invoice management, which involves legitimate file system and network access. However, the `process.md` file indicates the agent's capability to 'Download from URL if provided' for invoice capture. While intended for legitimate invoice portals, this capability introduces a vulnerability risk: if an attacker can control the URL provided to the agent, it could lead to arbitrary file download, potentially enabling further exploitation (e.g., RCE if the downloaded file is then processed or executed). There is no explicit malicious intent within the skill's code or instructions, but this capability, combined with email integration (also in `process.md`), makes it suspicious due to the potential for abuse or exploitation, rather than being benign.