Hunt business-logic flaws and open fix PRs
Parallel Devin agents each read part of a codebase, find business-logic and auth bypass flaws, confirm them in a sandbox, then open remediation PRs (vendor description).
Done withDevin
- What they did
- Parallel agents each take a segment of the codebase and reason across files for business-logic flaws, chained auth bypasses and cross-service exploit paths. Devin then combines the findings into full attack paths and reproduces each one in an isolated sandbox. Once a vulnerability is confirmed, it writes the patch and opens a PR for review. Scan profiles can be generated from existing threat model documents.
- How it went
- On Cognition's own 50-vulnerability benchmark, Devin Security scored 72% recall at $90.23 per run, against 68% for Claude Security at $131.87. This is a vendor benchmark, not an independent one.
- Worth knowing
- The first full scan sets the baseline. Later scans cover only changed code, so cost falls over time, and batch size per profile controls depth and cost.
Try it yourself with Devin
Review my codebase [repo name] for business-logic flaws and authentication or authorization bypasses. Split it into parts and examine each, then try to confirm every suspected issue in a sandbox, not on production. Finish with a ranked list of confirmed findings, and open a draft pull request with a fix for each one. Do not merge anything or touch live systems.
1 more account of this
Source: cognition.com · Undated
Five of these in your inbox every morning
The best things people got an AI agent to do, each with the prompt to try it.
More like this
Hand narrow code changes to Devin for pull requests
A team used Devin for a month on small fixes like chart formatting and API edge cases, with tests and PRs; a complex full-stack task went poorly.
Generate internal dashboards and data migration scripts
A backend engineer says their team uses Devin for internal dashboards and migration scripts, treating output as a starting point that saves about four hours of boilerplate.
Click through apps to verify its own work
With computer use, Devin built and played a desktop game, ordered on Amazon, and tested its own work by clicking through the app.
Find a Linux vulnerability hidden for 23 years
An account of Claude Code uncovering a long-standing vulnerability in Linux.
Control and calibrate a robotic hand with OpenClaw
An OpenClaw agent running Claude drove a 16-joint printed hand, checked its actions with a USB camera, calibrated firmware and narrated in Telegram.
Open recent protein-ligand structures in PDB
A researcher had Comet find the three latest GSK3B protein-ligand complexes in the PDB and open each in Ligand Interaction view.