Local Budget

This skill appears coherent and local-only, but review these practical safety steps before installing: 1) Run the scripts locally on a machine you trust (they process potentially sensitive bank CSVs). 2) Inspect the included Python files yourself if you have doubts — there are no network calls or credential use, so a quick code review is sufficient. 3) When running, keep outputs (categorized.json, report.md) in a directory you control; avoid writing reports into shared/system folders unless intended. 4) The workflow relies on a manual LLM-review step (the agent or you should confirm low/medium confidence items) — do not skip manual review if you care about correct categorization. 5) If your agent platform can autonomously call skills and transmit data externally, ensure agent-level network/external-access policies are appropriate; the skill itself does not perform network exfiltration. If you want higher assurance, run the scripts in a disposable environment (virtualenv / container) and test with redacted sample CSVs first.