Krishp1/Autonomous-Coding-Agent
0
1# 🤖 Autonomous Python Coding Agent2 3> **A production-grade, self-healing multi-agent pipeline that doesn't just generate Python code — it autonomously writes, validates, tests, secures, benchmarks, and reflects on its own output before shipping.**4 5[](https://python.org)6[](https://github.com/langchain-ai/langgraph)7[](https://groq.com)8[](https://chromadb.com)9[](https://streamlit.io)10[](LICENSE)11[](https://huggingface.co/spaces/krishpatel/autonomous-coding-agent)12 13---14 15## 🚀 Live Demo16 17**[▶ Try it on Hugging Face Spaces](https://huggingface.co/spaces/krishpatel/autonomous-coding-agent)**18 19---20 21## 📸 Demo22 2324 25---26 27## 🔥 What makes this different from just using ChatGPT?28 29| Feature | ChatGPT / Basic Agent | This Agent |30|---|---|---|31| Code generation | ✅ | ✅ |32| Syntax validation | ❌ Run and hope | ✅ AST parse before running |33| Test cases | ❌ Manual | ✅ Auto-generated by agent |34| Stress testing | ❌ | ✅ 500+ random inputs via Hypothesis |35| Memory | ❌ Stateless | ✅ ChromaDB learns from past bugs |36| Security audit | ❌ | ✅ Detects eval, exec, hardcoded keys |37| Performance check | ❌ | ✅ Benchmarks 1000 runs, rejects slow code |38| Self-review | ❌ | ✅ Agent scores own confidence 1-10 |39| Self-healing | ❌ | ✅ Loops back and fixes failures automatically |40| Separate retry counters | ❌ | ✅ Per-node counters prevent pipeline blockage |41 42---43 44## 📊 Key Metrics45 46| Metric | Value |47|---|---|48| Pipeline nodes | 13 |49| Verification layers | 5 (AST → Tests → Hypothesis → Security → Complexity) |50| Max retries (debugger) | 3 |51| Max retries (security, complexity) | 2 each — independent counters |52| Hypothesis test cases | 500+ random inputs per run |53| Benchmark iterations | 1,000 runs |54| Performance threshold | < 5ms per call |55| Memory backend | ChromaDB vector similarity search |56| LLM | Llama 3.1 8B Instant via Groq |57| Avg pipeline runtime | ~20–40 seconds |58| Lines of code | ~600 across 5 files |59 60---61 62## 🏗️ Architecture — 13-Node Pipeline63 64```65User Input (Python Task)66 │67 ▼68 ┌─────────┐69 │ Planner │ ── Breaks task into blueprint70 └────┬────┘71 │72 ▼73 ┌───────┐74 │ Coder │ ── Writes code using plan + ChromaDB memory75 └────┬──┘76 │77 ▼78 ┌───────────────┐79 │ AST Validator │ ── Syntax + hallucinated imports + type hints80 └──────┬────────┘ (no execution needed — milliseconds)81 │82 Pass │ Fail ──► Debugger ──► back to AST83 ▼84┌────────────────┐85│ Test Generator │ ── Auto-generates pytest-style test cases86└───────┬────────┘87 │88 ▼89 ┌────────┐90 │ Tester │ ── Runs code + generated tests in sandbox91 └───┬────┘92 │93 Pass │ Fail ──► Debugger (max 3 retries)94 ▼95┌────────────┐96│ Hypothesis │ ── 500+ random inputs, property-based testing97└─────┬──────┘ (never blocks pipeline — informational only)98 │99 ▼100┌───────────┐101│ Benchmark │ ── Runs 1000x, rejects if > 5ms/call102└─────┬─────┘103 │104 ▼105┌──────────┐106│ Security │ ── Detects eval/exec/hardcoded secrets107└─────┬────┘ (own retry counter — max 2)108 │109 ▼110┌────────────┐111│ Complexity │ ── Line count + nesting depth + LLM score/10112└──────┬─────┘ (own retry counter — max 2)113 │114 ▼115┌─────────────────┐116│ Self Reflection │ ── Agent scores own confidence 1-10117└────────┬────────┘ Rewrites if confidence < 7118 │119 ▼120 ┌──────────┐121 │ Reviewer │ ── Polishes + docstrings + type hints122 └─────┬────┘123 │124 ▼125 ┌──────────┐126 │Explainer │ ── Writes human-readable explanation127 └─────┬────┘128 │129 ▼130 OUTPUT131 Final Code + Explanation132```133 134---135 136## 📁 Project Structure137 138```139autonomous-coding-agent/140├── app.py ← Streamlit UI141├── main.py ← Graph builder + entry point142├── state.py ← Shared TypedDict state (whiteboard)143├── nodes.py ← All 13 node functions + LLM + ChromaDB144├── edges.py ← All 7 conditional route functions145├── requirements.txt ← Dependencies146└── README.md147```148 149---150 151## ⚡ Run Locally152 153### Prerequisites154- Python 3.11+155- Groq API key — get free at [console.groq.com](https://console.groq.com)156 157### Step 1 — Clone the repo158```bash159git clone https://github.com/krishpatel/autonomous-coding-agent.git160cd autonomous-coding-agent161```162 163### Step 2 — Create virtual environment164```bash165python -m venv venv166 167# Mac/Linux168source venv/bin/activate169 170# Windows171venv\Scripts\activate172```173 174### Step 3 — Install dependencies175```bash176pip install -r requirements.txt177```178 179### Step 4 — Set your API key180```bash181# Mac/Linux182export GROQ_API_KEY=your_groq_api_key_here183 184# Windows185set GROQ_API_KEY=your_groq_api_key_here186```187 188Or create a `.env` file:189```bash190echo "GROQ_API_KEY=your_groq_api_key_here" > .env191```192 193### Step 5 — Run CLI (no UI)194```bash195python main.py196```197 198### Step 6 — Run Streamlit UI199```bash200streamlit run app.py201```202 203Open [http://localhost:8501](http://localhost:8501) in your browser.204 205---206 207## 🐳 Run with Docker (optional)208 209```dockerfile210# Dockerfile211FROM python:3.11-slim212WORKDIR /app213COPY requirements.txt .214RUN pip install -r requirements.txt215COPY . .216EXPOSE 8501217CMD ["streamlit", "run", "app.py", "--server.port=8501"]218```219 220```bash221# Build222docker build -t coding-agent .223 224# Run225docker run -e GROQ_API_KEY=your_key -p 8501:8501 coding-agent226```227 228---229 230## 🌐 Deploy to Hugging Face Spaces231 232```bash233# Install HF CLI234pip install huggingface_hub235 236# Login237huggingface-cli login238 239# Create space and push240huggingface-cli repo create autonomous-coding-agent --type space --space_sdk streamlit241git remote add hf https://huggingface.co/spaces/YOUR_USERNAME/autonomous-coding-agent242git push hf main243```244 245Then add your secret in HF Spaces Settings:246```247GROQ_API_KEY = your_key_here248```249 250---251 252## 🛠️ Tech Stack253 254```255LangGraph — Stateful multi-agent graph orchestration256Groq API — LLM inference (Llama 3.1 8B Instant)257ChromaDB — Vector database for bug fix memory258Hypothesis — Property-based stress testing259Streamlit — Production UI260subprocess — Sandboxed isolated code execution261ast — Static code analysis without execution262hashlib — Deterministic ChromaDB IDs263importlib — Real-time import hallucination detection264```265 266---267 268## 💡 Key Engineering Decisions269 270### Why LangGraph over plain LangChain?271LangGraph handles **cyclic workflows** — when tests fail, the agent loops back through the debugger and restarts verification from AST. LangChain's linear chains can't do this cleanly.272 273### Why AST validation before running?274Running broken code wastes subprocess time. AST parsing catches syntax errors in **milliseconds** without execution — like a proofreader checking spelling before printing.275 276### Why Hypothesis for testing?277Hand-written tests only cover cases you think of. Hypothesis **auto-generates 500+ random inputs** and verifies properties that should always hold. Catches edge cases no human would write.278 279### Why separate retry counters per node?280One shared counter caused security failing 3 times to kill the entire pipeline before the debugger got its attempts. Separate counters for security and complexity mean each node fails independently without blocking others.281 282### Why hashlib instead of Python's hash()?283Python's `hash()` is **randomized every session** for security. Same error → different ChromaDB ID → agent can never retrieve past fixes. `hashlib.md5` is deterministic across all sessions.284 285### Why combined Reviewer + Explainer?286Two separate LLM calls for polishing and explaining wasted ~8 seconds. One combined call with structured output (`FINAL_CODE:` / `EXPLANATION:`) saves an entire API round trip.287 288---289 290## 🐛 Real Bugs Found and Fixed291 292**Bug 1 — False Positive in Tester**293`returncode == 0` doesn't mean the function was called. A file that only defines functions exits successfully but prints nothing. Fixed by checking `stdout` is not empty after successful run.294 295**Bug 2 — ChromaDB Hash Randomization**296Python's `hash()` is session-randomized. Same bug → different ID every run → memory retrieval never works. Fixed with `hashlib.md5().hexdigest()[:8]` for deterministic cross-session IDs.297 298**Bug 3 — Python 3.11 F-string Backslash**299Python 3.11 doesn't allow backslashes inside f-string expressions. Benchmark node embedded code inside f-strings. Fixed using string concatenation instead.300 301**Bug 4 — Shared Retry Counter**302One `retries` counter shared across all nodes caused security/complexity failures to consume the debugger's retry budget. Fixed by adding `security_retries` and `complexity_retries` as independent counters.303 304---305 306## 🔑 Environment Variables307 308| Variable | Required | Description |309|---|---|---|310| `GROQ_API_KEY` | ✅ Yes | Get free at console.groq.com |311| `GITHUB_TOKEN` | ❌ No | Only needed for AutoReview AI project |312 313---314 315## 📝 Resume Line316 317> **Autonomous Python Coding Agent** | LangGraph · Groq · ChromaDB · Streamlit318> Built a 13-node self-healing pipeline with 5-layer verification — AST validation, auto-generated tests, Hypothesis property testing (500+ random inputs), security audit, and self-reflection confidence scoring. ChromaDB vector memory enables cross-session bug fix learning. Deployed on Hugging Face Spaces.319 320---321 322## 👨💻 Author323 324**Krish Patel** — AI Engineer 325[GitHub](https://github.com/krishpatel) · [LinkedIn](https://linkedin.com/in/krishpatel) · [Live Demo](https://huggingface.co/spaces/krishpatel/autonomous-coding-agent)326 327---328 329*Built as part of AI Engineer internship portfolio — Bangalore, 2026*330