Tell it a tech domain you care about, and it discovers the best open-source contribution opportunities, generates a reviewable report, and submits PRs — after your approval.
High-quality PRs to well-known open-source projects are among the most persuasive credentials in the competitive LLM/AI job market. But the gap between "I want to contribute to the LLM space" and actually submitting a mergeable PR is wide:
- Manually trawling GitHub Trending, awesome lists, and community discussions to find active repos
- Reading CONTRIBUTING.md files one by one to gauge newcomer-friendliness
- Scanning hundreds of issues to find the one that matches your skill level, has merge potential, and delivers real value
- After contributing, lacking a structured narrative to present in interviews
OSS-Contributor automates discovery and evaluation — but leaves the final decision and execution approval in your hands.
# Install and authenticate GitHub CLI
gh auth login
# Verify
gh auth status# Clone directly into Claude Code skills directory
git clone https://github.com/Ottohere-Mourn/OSS-Contributor.git ~/.claude/skills/oss-contributor
# Edit preferences (optional)
vim ~/.claude/skills/oss-contributor/config.yamlIn Claude Code, just say:
"Find me some open-source projects in the LLM inference space that I can contribute to"
Or use the slash command:
/oss-contribute "LLM inference" --topk 3 --type docs,bugfixThe skill auto-triggers from natural language — no need to memorize exact syntax.
After PRs are submitted, follow up on reviews:
/oss-contribute followup --session 20260530-143000Input: domain keyword (e.g., "LLM inference optimization")
│
▼
┌──────────────────────────────────────────────┐
│ 1. Discover │
│ Multi-dimensional search → cross-validate │
│ Human-in-the-loop: confirm dimensions │
│ Output: repos.json │
└──────────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────────┐
│ 2. Evaluate │
│ Deep-dive Top-K repos: code audit + matrix│
│ Output: opportunities.md │
└──────────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────────┐
│ 3. Review (Human Gate) │
│ Review opportunity report → select targets│
│ Output: selection.json │
└──────────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────────┐
│ 4. Execute │
│ Worktree isolation → TDD → PR submission │
│ Auto-retry on failure + lesson relay │
│ Output: pr-links.json │
└──────────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────────┐
│ 5. Report │
│ Interview-ready contribution retrospective│
│ Output: contribution-report.md │
└──────────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────────┐
│ 6. Followup │
│ /oss-contribute followup --session <id> │
│ Check PR status + handle review feedback │
└──────────────────────────────────────────────┘
| OSS-Contributor | Sweep AI | auto-github-contributor | autonomous-dev-team | |
|---|---|---|---|---|
| Domain discovery | ✅ Multi-dimensional | ❌ | ❌ | ❌ |
| Human-in-the-loop gates | ✅ Two review points | ❌ | Partial | ❌ |
| Code-level evaluation | ✅ Shallow clone verification | ❌ | ❌ | ❌ |
| Structured reporting | ✅ Interview-ready | ❌ | ❌ | ❌ |
| PR followup | ✅ /followup command |
❌ | ❌ | ❌ |
| Auto-merge | ❌ Deliberately omitted | ✅ | ❌ | ✅ |
Core philosophy: OSS-Contributor accelerates discovery and evaluation, but keeps decision-making and execution approval entirely under your control. It never auto-merges code.
Edit config.yaml in the skill directory:
defaults:
topk: 5 # Repos to deep-evaluate
min_stars: 500 # Minimum stars threshold
languages: ["Python"] # Language preference (empty = any)
types: ["bugfix", "docs", "test"] # Contribution type preference
exclude_repos: # Repos to always skip
- "tensorflow/tensorflow"
profile:
name: "Your Name"
github_username: "Ottohere-Mourn"
search:
max_pages: 3
sort_by: "stars"
min_updated_days: 180
execution:
max_iterations: 20
require_human_gate: trueCommand-line flags override config values.
- Never: force push, commit to main/master, skip pre-commit hooks
- Human gate is non-bypassable: Phase 3 review cannot be skipped
- Worktree isolation: each PR operates independently — no cross-contamination
- No credential management: authentication is delegated to
gh auth— the skill never touches tokens - Followup never comments unattended: beyond "thanks, addressed!" replies, nothing is posted without human approval
oss-contributor/
├── SKILL.md # Main orchestration (6 Phases)
├── config.yaml # User preferences
├── prompts/
│ ├── discover.md # Phase 1: Multi-dimensional search + scoring
│ ├── evaluate.md # Phase 2: Deep evaluation + code verification
│ ├── execute.md # Phase 4: Worktree execution + TDD
│ ├── followup.md # Phase 6: PR followup
│ └── report.md # Phase 5: Report generation
└── templates/
├── repos-schema.json # Repo data schema
├── opportunities-schema.json # Opportunity data schema
└── report-template.md # Report output template
The following projects informed the design of this skill:
- auto-github-contributor — Execution pipeline reference
- autonomous-dev-team — Worktree isolation patterns
- WhatIfWeDigDeeper/agent-skills — Skill engineering benchmarks
- Sweep AI — Pioneered issue-to-PR automation
This project is licensed under the MIT License — see the LICENSE file for the full text. In short: you can do whatever you want with this code, as long as you include the original copyright and license notice.