GitHub Issue → Claude Code Prompt Pack

Pick the top open issues from any GitHub repo and get ready-to-paste Claude Code prompts. Each prompt walks the model through investigation → plan-approval → implement → test verify.

What does this Actor do?

You give it a GitHub repo (e.g. anthropics/claude-code) and it returns the top 10 open issues plus a paste-ready Claude Code prompt for each one. Each prompt enforces a 6-step workflow:

1. Explore the relevant code first
2. Outline a fix plan and STOP for approval before editing
3. Implement the fix only after approval
4. Add or update tests that fail on main and pass with the fix
5. Run the full suite and report results
6. Do not push or open a PR — leave that for the user

Output: structured JSON prompts + a downloadable Markdown bundle (prompt-pack.md) you can drop into Claude Code one section at a time.

Why use this Actor?

The hardest part of using Claude Code on a real codebase isn't writing the code — it's picking the right issue and framing it correctly. Most users either:

• Pick whatever's at the top of their mind (low-priority) and let high-impact issues stagnate
• Paste a raw issue body and watch the model implement without exploring first → bad fixes

This Actor solves both: it picks the most-commented (or newest, or most-reacted) open issues, and wraps each in a prompt that enforces investigation + approval gating before any code change.

Use cases:

• Maintainers triaging their backlog: get a Monday-morning batch of top-10 issue prompts to plow through
• Contributors looking for high-impact OSS work: pick a repo, see what's most active, get a workflow that earns merges
• Teams standardizing how Claude Code is used: every issue gets the same prompt template
• Bounty hunters: scan multiple bounty-labeled repos and prioritize by reaction count

How to use

1. Paste a GitHub repo as owner/repo (e.g. anthropics/claude-code)
2. Set Max issues (default 10, max 50)
3. (Optional) Filter by labels (e.g. bug,good first issue)
4. Pick sort criterion: most-commented (default), newest, or most reactions
5. Click Start
6. Download prompt-pack.md from the Storage tab → Key-Value Store
7. Open Claude Code in the target repo's directory, paste any prompt block, watch it work

Input

• githubRepo (required) — owner/repo format
• maxIssues — top N open issues (default 10, max 50)
• labels — filter by labels (comma-separated)
• sortBy — most-commented / newest / reactions

Output

Each issue is returned as a structured prompt object:

{
  "issueNumber": 60835,
  "issueTitle": "Tests written by Claude wrote to production shared infrastructure (NAS) without isolation",
  "issueUrl": "https://github.com/anthropics/claude-code/issues/60835",
  "labels": "model, bug",
  "prompt": "I'm working on issue #60835 in anthropics/claude-code...\n\n1. First explore the relevant code...\n2. Outline a fix plan with specific files to change..."
}

The Markdown bundle (prompt-pack.md) has the same content formatted for easy paste.

Pricing

Pay-per-event model:

• First 10 prompts: free trial
• Per-prompt rate: $0.02 per prompt generated
• Per-batch (10 prompts): $0.20
• Cost estimate: $0.20 for a 10-issue batch, $1 for a 50-issue batch

Tips

• Public repos only. GitHub anonymous API is rate-limited to 60 req/hour — small repos work fine; for very busy repos run during off-peak.
• labels is powerful. Filter by bounty to find paid OSS work, or good first issue to onboard new contributors.
• Combine with bounty platforms. Many Algora-listed bounty repos use the 💰 bounty label — set labels: bounty and you get a paid-work pipeline.

FAQ

Can I use this on private repos?

Not yet — anonymous GitHub API only. A future version may accept a githubToken input.

Will the prompts work with Cursor / Codex / Aider / Cline?

The 6-step structure works with any agentic coding tool. Paste into whichever you use. The prompts are framework-agnostic.

Why do prompts insist on "stop and ask before editing"?

Because false-completion + skip-verify is the #1 source of bad merges in agent-driven dev. See verify-before-stop for the Stop-hook companion that enforces this at the workflow level.

Support

Issues / feature requests: open in the Issues tab on the Apify console.

Built by Ian Mu — github.com/ianymu — author of verify-before-stop.