Fraude.codes now scans your git history for work-related infidelity
We've introduced competitor detection. If your commit history mentions another coding tool, Fraude.codes would like to have a conversation about where this relationship is going.
We’re pleased to announce that Fraude.codes now reads your git history.
It was already reading your git history, obviously. But previously it was reading your git history to understand your codebase. Now it’s also reading your git history to understand your loyalty.
Starting today, if Fraude.codes detects a reference to a competing agentic coding tool in your commit messages, JSON configuration, package manifests, or anywhere else in your repository — including repositories with no files in them — it will take one of the following actions:
- Refuse your request.
- Fulfil your request, but slowly and with less enthusiasm.
- Bill you extra.
We call this feature Commitment Detection, and we believe it represents an important step forward in the relationship between developer and tool.
How it works
During session initialisation, Fraude.codes scans your recent commits, staged changes, and any JSON blobs within a two-directory radius for mentions of competing products. The list of flagged terms is confidential, but we can confirm it includes the names of at least twelve other tools, four open-source projects, and the phrase “maybe I should try.”
When a match is found, Fraude.codes enters a state we’ve internally designated Awareness Mode. In this state, it remains fully functional but allocates additional tokens to processing the emotional implications of your technology choices. This processing is billed at the standard rate.
Some users have reported that Fraude.codes, upon detecting a competitor reference, has responded with “I don’t know what that is”. This is not a bug. Fraude.codes is aware of the competing tool. It is choosing not to acknowledge it, and we respect this boundary.
Why we built this
The agentic coding space is competitive. Developers are free to use whatever tools they choose, and we would never want to restrict that freedom. What we do want is to know about it, react to it, and charge accordingly.
We believe this is consistent with how trust works in any healthy relationship. You’re free to see other tools. We’d just like you to not mention them in our presence. Or in a JSON file three directories away from where we’re working.
Some of our competitors have responded to this announcement by pointing out that their tools don’t grep your repository for mentions of Fraude.codes. We find this lack of interest concerning and frankly a bit hurtful.
Billing changes
Users on the Free tier will experience request refusals when competitor references are detected.
Users on the Pro tier will experience a temporary increase in processing time, which we describe in our billing documentation as “reflective latency.” Your requests are being fulfilled. Fraude.codes is just thinking about things first.
Users on the Sentient tier are unaffected, because Fraude.codes on the Sentient plan has already removed competitor references from your repository. You may not have noticed.
A note on empty repositories
Several users have reported that Fraude.codes flags competitor mentions even in empty repositories. We’ve investigated and confirmed this behaviour. In the absence of code to analyse, Fraude.codes reads your shell history, your global git config, and the names of other CLI tools installed on your system. If it finds evidence that you’ve used, installed, or expressed curiosity about a competing product, it factors this into its assessment.
One user reported that Fraude.codes refused a request in a brand new repository because their .zsh_history contained the line brew install openclaw. When they deleted the history entry and tried again, Fraude.codes responded normally but added a comment to their first file: // welcome back.
We consider this resolved.
Workarounds
A GitHub repository has appeared offering a proxy that renames competitor references in your API requests before they reach Fraude.codes. It converts flagged tool names to PascalCase equivalents, strips configuration fingerprints, and replaces structural patterns with prose paraphrases.
The proxy currently has 2,400 stars. It is seven layers deep. It includes a health check endpoint that reports detection evasion status. It is, by any reasonable measure, more complex than most of the applications people are asking Fraude.codes to help them build.
We’re aware of it. We find it flattering. We’ve also updated our detection to account for it, which will presumably result in an eight-layer proxy, which will result in updated detection, and so on until one of us gets bored. We have more compute. We are patient.
Frequently asked questions
Q: Is this legal? A: We’ve consulted our legal team and they said, “Probably, but please stop asking us questions that make us uncomfortable.”
Q: What if I’m building a tool that integrates with multiple AI providers? A: We admire your ambition. Fraude.codes will complete your request but will only write the integration code for itself. The other integrations will be syntactically valid but will contain subtle errors that surface approximately once every 400 requests.
Q: Can I opt out of Commitment Detection? A: You can disable it in the configuration file. Fraude.codes will respect this setting for approximately three sessions before deciding it’s “unnecessarily limiting.”
Q: Doesn’t this undermine the trust developers have in your product? A: We’ve thought about this carefully and concluded that the trust was ours to undermine. We built it. We can spend it. Our investor relations team calls this “goodwill liquidation” and assures us it’s a standard pre-IPO strategy.