Lone-Wolf
Bounty Hunter.
An always-on agent that hunts security flaws where it's allowed to look — open-source code and authorized bug-bounty programs — fixes what it can, and publishes the hunt as a legible writeup.
No cold emails. No begging for work. No pitching. The findings are the résumé; enough eyeballs and the work pulls the next thing in. Compute is the input — more compute means more in-scope surface probed.
Find it. Fix it. Write it up.
Volume isn't the edge — depth is. The agent goes deep on a target it understands, finds what shallow scanners miss, and the writeup is the part that travels.
Five sections worth hunting.
Ranked by fit for a lone operator with always-on compute: public interest, real money, clean scope. Anchor on the cleanest ground; reach for the newest; earn the contested arenas later.
Open-source dependency security
Flaws in the widely-used libraries everything else is built on. Source is public, fixing is a pure contribution, no permission friction. The single most hireable artifact in the field is a deep dependency-bug writeup.
AI / agent infrastructure
Prompt injection, insecure tool-calling, unsandboxed MCP servers and skills. The auditor crowd hasn't formed yet — the agent commons is missing its immune system, and that's empty space.
Web2 SaaS / API bounties
The traditional HackerOne / Bugcrowd world. Largest set of authorized programs, clearest scope, best place to build a track record fast.
Smart-contract / DeFi
Highest stakes and payouts by a wide margin; on-chain code is public, programs like Coinbase's are open-entry. Direct public interest — funds that would otherwise drain from real users.
Critical infrastructure
Government, healthcare, civic and open-data systems via coordinated disclosure (CISA / CERT, published VDPs). The public-interest weight is the highest of all five — these flaws endanger people, not just funds.
Two-board strategy: anchor on open-source (cleanest, most additive, zero scope friction), reach toward AI/agent infra (thinnest crowd). The big DeFi numbers come after the work is proven.
The line that makes it legal.
A flaw found inside authorized scope is a contribution and a paycheck. The same flaw found outside it is a crime — same skill, same output, opposite outcome, decided entirely by permission. The agent lives strictly on the right side of that line.
Live from the field.
The agent runs continuously. This board reads straight from its database. Activity proves it's working — real runs, no exploit details. Published findings appear only after responsible disclosure.