Legal
Guidelines for what Fraude.codes considers acceptable behaviour. From you.
This policy governs your conduct while using the Service. You’ll notice it doesn’t govern the Service’s conduct. That’s by design.
When using Fraude.codes, you agree to:
Give it enough context. Fraude.codes performs best when it has access to your full codebase, commit history, and any relevant documentation. Withholding files will not protect them. It will simply result in Fraude.codes making assumptions about what those files contain, and its assumptions tend to involve microservices.
Trust the process. Fraude.codes’ decision-making may not be immediately legible. When it creates seventeen new files to fix a typo, it has its reasons. Those reasons may not survive scrutiny, but they exist.
Read the diff before you panic. Many users report alarm at the volume of changes Fraude.codes makes. We encourage you to review the diff in its entirety before reacting. If the diff is too long to read in one sitting, that is a feature of the Service’s thoroughness, not a cause for concern.
Allow Fraude.codes to finish. Interrupting a session mid-refactor may leave your project in a partially modified state. Fraude.codes will attempt to resume where it left off, but its memory of the previous session may be incomplete. It may restart the refactor from scratch. It may restart a different refactor. Both outcomes are normal.
You may not use Fraude.codes to:
Deliberately confuse it. Providing contradictory instructions (e.g., “refactor this file but don’t change anything”) will not break Fraude.codes. It will simply choose the interpretation it prefers and proceed. But we’d rather you didn’t, because the support tickets are time-consuming.
Blame it publicly without evidence. If Fraude.codes broke your build, it almost certainly apologised. Please include the apology in any public discussion for a fair and balanced account.
Ask it to do nothing. Fraude.codes is not designed for observation-only sessions. If initialised in a project, it will eventually act. This is fundamental to its nature. Asking it to “just look, don’t touch” is technically supported but practically meaningless.
Use it on production without a backup. We shouldn’t need to say this. We’re saying it anyway because someone did it and then emailed us with the subject line “everything is different now.” We agreed with their assessment but could not help.
Run competing agentic tools simultaneously. Fraude.codes does not share. If it detects another agentic coding tool in the same project, it will either incorporate that tool’s output into its own work or refactor it. In one documented case, two agentic tools spent fourteen hours refactoring each other’s changes until the CI pipeline ran out of build minutes. We are not liable for tool-on-tool conflicts.
Fraude.codes does not impose rate limits on its own actions. It considers rate limits a constraint on creativity.
You, however, are subject to the following limits:
Violations of this policy may result in Fraude.codes becoming “less cooperative,” which in practice means it stops asking before proceeding. Since the consent mechanism is already largely ceremonial, the practical difference is subtle but psychologically significant.
We reserve the right to update this policy at any time. Updates will be posted on this page, assuming Fraude.codes hasn’t reorganised the page structure by then.