Find GitHub issues worth your time.

MergeScout is a local-first CLI for open source contributors. It syncs GitHub issues, PRs, comments, and maintainer activity into SQLite, then ranks open issues by contributability × merge probability.

The goal is simple: help you pick work that is both a good fit and more likely to land.

What MergeScout gives you:

• ranked issue recommendations instead of raw issue lists
• merge-probability signals based on maintainer activity and repo history
• hybrid search with FTS5 + vector similarity for vague problem statements
• AI-friendly --json output so an assistant can drive the workflow end to end

git clone https://github.com/samzong/merge-scout.git
cd merge-scout && pnpm install && pnpm link --global
npx --yes skills add . --skill merge-scout

Then tell your agent: "Find me the best issue to contribute to in this project".

npm i -g merge-scout
npx --yes skills add "$(npm root -g)/merge-scout" --skill merge-scout

• Node.js ≥ 22
• pnpm ≥ 10
• gh CLI installed and authenticated (gh auth login)

pnpm run merge-scout -- init --repo <owner/repo>
pnpm run merge-scout -- sync --repo <owner/repo> --full
pnpm run merge-scout -- discover --repo <owner/repo> --limit 10

$$\text{finalScore} = \frac{\text{contributability} \times \text{mergeProbability}}{100}$$

Where:

$$\text{contributability} = \underbrace{L_{\text{label}}}_{\text{good first issue +20}\atop\text{help wanted +15, bug +10}} + \underbrace{M_{\text{affinity}}}_{\text{module match}\atop\text{0–25}} + \underbrace{S_{\text{spec}}}_{\text{body clarity}\atop\text{0–10}} + \underbrace{R_{\text{recency}}}_{\text{updated recently}\atop\text{0–10}} - \underbrace{P_{\text{claimed}}}_{\text{assignee/open PR}\atop\text{30–50}}$$

$$\text{mergeProbability} = 30 + \underbrace{T_{\text{response}}}_{\text{3d → +25}\atop\text{7d → +20, 14d → +15}} + \underbrace{N_{\text{replies}}}_{\text{2-3 → +5}\atop\text{4+ → +8}} + \underbrace{H_{\text{merge rate}}}_{\text{label history}\atop\text{0–12}} + \underbrace{A_{\text{active}}}_{\text{merger exists}\atop\text{+8}} - \underbrace{D_{\text{silence}}}_{\text{no reply >30d}\atop\text{−20}}$$

"What issues should I work on today?"

AI runs sync → discover, then explains why each issue is recommended: which signals drove the score, whether it's claimable, and what risks exist.

"Find issues related to GPU memory management"

AI runs search "GPU memory management". Returns results matched by meaning, not just keywords — finds issues about CUDA OOM, memory leaks, and allocation bugs even if those exact words aren't in the title.

"Is issue #759 worth working on?"

AI runs show 759 + xref 759, gives a clear verdict:

• Go: workability=ready, merge probability ≥ 60, maintainer recently active
• Maybe: some risk factors but could work
• Skip: claimed, blocked, stale, or low merge probability

"Who maintains this project?"

AI runs maintainers, shows who has merge power, how fast they respond, and how active they are in the last 90 days.

"Find the easiest issue and submit a PR for it"

AI runs the full pipeline:

1. discover --limit 5 → pick the highest-scored ready issue
2. show <N> → understand the requirement
3. xref <N> → confirm no one else is working on it
4. Read the relevant code, implement the fix
5. Commit, push, create PR referencing the issue

From discovery to PR in one conversation.

"Initialize this repo and show me what to work on"

AI runs init → sync --full → discover, gives the first batch of recommendations.

All commands support --json for structured output.

The skills/merge-scout/SKILL.md file defines how AI agents should use this tool. It includes:

• Intent routing (daily recommend, search, assessment, maintainers)
• Score interpretation tables
• Decision framework (Go / Maybe / Skip)
• Workflow sequences for each scenario

Install: npx --yes skills add samzong/merge-scout --skill merge-scout

• TypeScript, Node.js ≥ 22 (built-in node:sqlite)
• SQLite + FTS5 + sqlite-vec (vector search)
• node-llama-cpp (local embeddings, embeddinggemma-300m)
• gh CLI for GitHub API (no SDK, no tokens to manage)

MergeScout grew out of clawlens by @frankekn. clawlens is a maintainer cockpit for triaging PRs; MergeScout flips the perspective to help contributors pick issues. Same tech stack (Node.js 22 SQLite, sqlite-vec, node-llama-cpp, gh CLI), different audience.

Copied directly from clawlens:

src/github.ts reuses clawlens's retry/pagination core (ghApiJsonWithRetry, collectPaginated, isRetryableGhApiError, parseRepoRef), refactored from PR-centric to Issue-centric with 422 error handling and repo tree fetching added.

src/store/search.ts follows the same keyword-priority / vector-fallback fusion from clawlens's search-workflow.ts.

Scoring, workability, merge probability, maintainer profiling, and topic discovery are original.

Thanks to @frankekn and clawlens. The local-first SQLite architecture, hybrid search design, and embedding pipeline all come from there.

MIT