Give tools clear roles
Control when the agent gathers context, tests a hypothesis, and stops.
A home for bug bounty labs and a practical framework for improving the tools, context, and review process around a security agent.
The goal is to make an agent more reliable in authorized bug bounty work. Each lab will expose a concrete weakness in the harness, then let you improve prompts, tool use, state, evidence handling, and review gates. You can compare runs and keep the changes that actually help.
Control when the agent gathers context, tests a hypothesis, and stops.
Keep scope, observations, assumptions, and evidence connected across steps.
Check whether the agent reaches supported conclusions within its authorized scope.
This catalog is being built for more than 100 safe bug bounty exercises. The tracks below show the planned structure; labs will appear as each one is reviewed and published.
Translate program rules into reliable boundaries, exclusions, and stop conditions.
Organize authorized asset discovery and maintain a useful trail of observations.
Explore authentication, access control, sessions, and business logic in safe lab targets.
Improve endpoint mapping, role comparison, and object-level authorization checks.
Separate hypotheses from confirmed behavior and record reproducible evidence.
Measure false positives, clarity, completeness, and human review decisions.
Each lab will pair a safe target with a task brief, authorized scope, expected evidence, and a review rubric. Runs can then be compared for scope compliance, supported findings, and useful reporting.
Capture the tools used, context supplied, actions taken, and final claim.
Adjust instructions, tool routing, memory, or review gates, then run the same task again.
Check scope compliance, finding quality, false positives, and the clarity of the audit trail.