The problem
Solo developers and students who earn from online bounties, hackathons and small contests spend much of their time on triage instead of building. Is the listing still open? Has someone already claimed it? Did the deadline move? Is there a hidden gate, like a live interview, "must be hired first", or a card required?
Aggregators and summaries go stale fast. From my own tracking log over the past weeks: a $1,500 GitHub bounty looked free until the GitHub API showed it was already assigned; a 30-project open-source bounty list still read "open" while its claim sheet showed every project taken; and a summary I was given said a contest had 48 submissions when the live page already showed 99, halving the odds. This hits hardest for builders in countries where these rewards are meaningful income (for example Vietnam and India), who hunt across Devpost, GitHub, Superteam and similar sites.
My idea
An always-on agent, hosted on Agent37, that keeps a builder's watchlist of opportunities and re-checks each one on a schedule against its canonical source. Every morning it sends a short digest of what changed: closed, claimed, deadline moved, a new blocking rule, or a jump in competition. Every claim is backed by a sentence quoted from the source page or by an API field, so the agent cannot invent facts. Next, the builder will also be able to chat with it ("is this listing worth my weekend?") and get GO / REVIEW / SKIP with the quoted reasons.
Why it matters
It saves hours each week, stops people from building for bounties that are already gone, and flags hidden costs and gates before anyone signs up. Quote-backed answers make it trustworthy enough to act on.
How I'm starting
I already have an open-source prototype, BountyPilot (MIT licence): GitHub, live demo. It has a rule engine, LLM extraction where every field must carry a verified quote, and liveness checks for GitHub issues and web pages, exposed as MCP tools.
A first version already runs on Agent37: a free Hermes agent downloaded the checker (agent37 folder) into its workspace, produced a quote-backed digest of six live opportunities, and has a daily Agent37 cron at 09:00 Vietnam time that re-runs the check and reports what changed. Sites that answer with a bot check are reported as BLOCKED, never bypassed. Next steps: let builders add listings by chatting with the agent, deliver the digest to Telegram, and port BountyPilot's GO / REVIEW / SKIP triage into the agent.
Skills I bring: Node.js, Python, MCP servers, LLM tooling and testing.
Disclosure: built with AI coding assistance (Claude).
Built With
- agent37
- github-api
- hermes
- mcp
- python

Log in or sign up for Devpost to join the conversation.