GitHub Cleaner AI Agent

Production-grade AI system that analyzes GitHub profiles, evaluates repository quality, and provides structured recommendations to improve portfolio quality.

Architecture

• Layer 1 — Agent: Conversation, tool calls (Voiceflow / LLM).
• Layer 2 — Backend API: FastAPI; orchestration and response shaping.
• Layer 3 — Scoring Engine: Deterministic scoring and classification.
• Layer 4 — Data: GitHub API client (read-only in MVP).

See ARCHITECTURE.md and docs/MODULES.md. Full project work plan (vision & phases 1–7):docs/PROJECT_WORK_PLAN.md. Docs map:docs/README.md.

Phase 1.5 — Voiceflow agent: connect Voiceflow to POST /scan/voiceflow (flat JSON) or POST /scan (full JSON) — docs/VOICEFLOW_AGENT.md · contract docs/AGENT_INTEGRATION.md · deploy API (e.g. Render):docs/DEPLOY_RENDER.md.

Production example: API at https://github-cleaner-api.onrender.com — Voiceflow uses https://github-cleaner-api.onrender.com/scan/voiceflow. Set GITHUB_TOKEN in Render’s environment.

MVP (Phase 1)

• Read-only GitHub scan
• Repository scoring (0–100)
• Classification: showcase / cleanup / archive
• Issue detection and suggestions
• No write actions

Setup

python -m venv .venv
source .venv/bin/activate # or .venv\Scripts\activate on Windows
pip install -e .

GitHub token (recommended): Copy .env.example to .env, put your real token in GITHUB_TOKEN=..., and restart the API. The app loads .env automatically. Never commit .env (it is in .gitignore).

Run API

From project root (after pip install -e .):

PYTHONPATH=src uvicorn api.main:app --reload

Or run from src: uvicorn api.main:app --reload

POST /scan (full response) or /scan/voiceflow (flat fields for no-code tools) with JSON:

{
"github_username": "octocat",
"review_mode": "portfolio",
"scan_scope": "public"
}

• GET / — small JSON index (links to docs, health, scan paths).
• GET /docs — Swagger UI.

Streamlit demo (free UI, no Voiceflow)

Calls the same POST /scan/voiceflow as the hosted API (default: Render). Runs on your machine; uses httpx server-side (no CORS issues).

source .venv/bin/activate
pip install -e ".[demo]"
streamlit run streamlit_app.py

Then open the URL Streamlit prints (usually http://localhost:8501). Set API base URL in the sidebar if your API is not on Render.

Deploy (Render)

See docs/DEPLOY_RENDER.md. After deploy, use https://<your-service>.onrender.com/scan/voiceflow in Voiceflow (not ngrok).

Save scan result to JSON

From project root, with the API running:

chmod +x scripts/scan-and-save.sh
./scripts/scan-and-save.sh Mjeedbakr

Result is saved to output/scan_Mjeedbakr.json. Use any username:

./scripts/scan-and-save.sh octocat # → output/scan_octocat.json
./scripts/scan-and-save.sh YOUR_USER # → output/scan_YOUR_USER.json

Or with curl only (saves formatted JSON):

curl -s -X POST http://localhost:8000/scan \
-H "Content-Type: application/json" \
-d '{"github_username":"Mjeedbakr","review_mode":"portfolio","scan_scope":"public"}' \
| python3 -c "import sys,json; json.dump(json.load(sys.stdin), open('output/scan_Mjeedbakr.json','w'), indent=2)"

Project Layout

src/
api/ # FastAPI app and routes
contracts/ # Pydantic input/output and DTOs
data/ # GitHub API client
scoring/ # Scoring, classification, issues, suggestions
docs/ # Architecture and module docs
tests/ # Tests

License

MIT