Oss Bounty Hunter

OSS Bounty Hunter

Surface paid open-source coding tasks (bounties) from public bounty platforms, filter for fit, and propose a shortlist the user can act on.

When to use this skill

Activate when the user expresses any of:

• "I want to earn money working on open source"
• "Find me OSS bounties / paid issues / GitHub bounties"
• "What can I work on for cash this week"
• "Show me bounties matching my stack"
• "Side income through coding"
• Any reference to Algora, Replit Bounties, Bountysource, Gitcoin bounties, or bounty platforms

Do not activate for unrelated GitHub queries (issue triage, repo audit, etc.).

Procedure

1. Fetch the latest open bounties:

   ./scripts/algora-watch.sh --summary
   

   This calls the Algora public tRPC endpoint and prints [status] $reward title / org / type / kind / url lines.

2. Triage each bounty for two disqualifiers:

   • Reservation: many bounties carry an Reserved for SE interview label or already have an assignee and a linked PR. These are effectively closed to community contributors. Skip them unless the user explicitly opts in.
   • Out-of-scope tech: the user's primary stack should appear in the linked GitHub issue (language, framework, or domain). When unknown, ask the user once for their stack.

3. Score the survivors on three axes (1–5 each, 5 is best):

   • Reward density: reward / estimated hours of work. Use the issue body length, file count touched, and "complexity" labels as a rough proxy.
   • Stack fit: how well the issue matches the user's known stack and prior PRs.
   • Solvability signal: whether the issue has a clear acceptance criterion, a recent comment from a maintainer, and no blocking dependencies.

4. Present a shortlist of the top 3–5 bounties as a Markdown table: | rank | reward | title | url | reserved? | score | one-line rationale |

5. Recommend a starting point: name the single best bounty for the user to take next, with a brief plan ("clone, look at file X, run test Y").

6. Optional follow-ups:

   • If asked, fetch the linked GitHub issue and extract acceptance criteria, file hints, and existing PRs.
   • If asked, draft a comment for the user to post on the issue claiming the bounty.

Filter by reward range

If the user gives a minimum reward (e.g. "at least $200"), pipe through:

./scripts/algora-watch.sh --filter "." | awk '/\$[0-9]+/ {if (match($0, /\$([0-9]+)/, m) && m[1]+0 >= 200) print}'

Edge cases

• All bounties reserved: report this fact explicitly. Do not pretend a reserved bounty is openly takeable. Suggest re-running later or trying alternative platforms.
• Single org dominates (e.g. one company has 9 of 10 bounties): mention this — it skews available work and may indicate a paid-interview pipeline rather than open community contribution.
• API returns empty: Algora's tRPC endpoint occasionally returns no data on errors. Show the raw response and ask the user to retry.
• User has no GitHub auth set up: discovering bounties does not require authentication, but solving them does. Mention that taking a bounty will require GitHub auth (gh auth login or equivalent).

Tech stack heuristic (fallback)

If the user's stack is unknown and they don't answer:

• Treat TypeScript, Python, Go, Rust as default in-scope.
• Treat C++, embedded, kernel, specialized cryptography as out-of-scope unless the user opts in.

Output format

Always return:

1. A one-line summary of the bounty market right now (e.g. "Algora has 10 open bounties totalling $9,080, but 8 are reserved for hiring; only 2 are openly takeable.").
2. The shortlist table (rank/reward/title/url/reserved/score/rationale).
3. The single recommended next step.
4. A note on what platforms were not checked (e.g. Replit Bounties not yet integrated) so the user knows the coverage.

Keep the entire response under 60 lines unless the user asks for full detail.

Limitations

• Currently covers Algora only. Replit Bounties, Bountysource (status uncertain), and Gitcoin (crypto-only) are not yet integrated. Future versions may add them.
• The tRPC endpoint used is public but unofficial; if Algora changes it, this skill must be updated.
• Scoring is heuristic and depends on the agent's reasoning, not a fixed algorithm.