I successfully bypassed Fable’s safety filter by avoiding any hacking terms and keeping a defensive tone.
That said, you can’t do a full code review with it — the filter triggers as soon as it starts Grepping the code.
Ask Fable to identify specific parts of the code and the vulnerability class to target (safety filter triggers right after Fable answers you), then give those instructions to Opus to run the review
Why build a Python multi-agent audit pipeline when Claude Code already exists?
Because SRs and developers solve different problems
Developers want autonomous execution
Security researchers need reproducible flow, validation gates, specialized agents, and pattern memory
Stochastic exploration, deterministic process.
Claude hunts the bugs 🕵️
The pipeline keeps the hunt rigorous 📐
Full architecture breakdown on HackenProof 👇
📖 [New Guest Article on the HackenProof Blog]
Reliable code review requires more than a single broad prompt for an AI auditing assistant.
In a new article by @AboutikaR, we break down how to build a multi-agent AI code review pipeline, where each agent focuses on a specific bug class, from reentrancy to access control.
Using the same model and the same code, a better analysis structure can deliver stronger results.
Great shoutout! 🙏
Really nice experience on @HackenProof — finally scored my first blood after tons of duplicates.
Fast judging and excellent team support 🔥
$120,000+ paid out by @hyperbridge to 20+ security researchers in less than a week on HackenProof.
We’re grateful to @hyperbridge for being proactive and for their commitment.
A big thank you as well to our hacker community for their valuable contributions.
@HackenProof 41 submissions: 22 duplicates, 3 paid and 11 info... I joined the bounty too late, more than 24 hours after it started. Ready for the next one! 🫡
@_kujen5@cantinasecurity At least you got a reply. I’ve got a couple of reports that have been in review for weeks on Cantina. Keep pushing, it will pay off!