AgenticTesting

Choose a repo. Set a coverage target. Let agents grow the test suite.

---

✨ What It Does

AgenticTesting is a local-first framework for AI-assisted test generation. It measures a repository's current line coverage, asks agents to reason about the gaps, lets them add or improve tests through constrained repo tools, and repeats until the requested target is reached.

The core loop is intentionally simple:

Step	Agent / System	Job
1	Coverage runner	Measures current line coverage
2	Analyzer	Finds useful test gaps
3	Planner	Chooses the next focused test task
4	Implementer	Edits tests through safe repo tools
5	Reviewer	Reviews the result and guides the next loop

The loop stops when the requested line coverage is reached or the iteration limit is exhausted.

🚀 Quick Start

git clone https://github.com/sl-badcoder/AgenticTesting.git
cd AgenticTesting
python -m venv .venv
source .venv/bin/activate
pip install pytest pytest-cov cryptography
python -m src.frontend.debug_ui

Open:

http://127.0.0.1:8765

On first launch, create a local user. Then select a repository, target coverage, project type, provider, and model.

🧭 Highlights

The UI is the primary workflow. It is designed for repeated local use: pick a project, tune the target, choose a model, run the agents, inspect the result, and come back later with your settings still saved.

Area	Support
UI	Local browser dashboard with saved settings
Repo picker	macOS Finder dialog via osascript, plus in-page fallback
Local models	llama.cpp GGUF, Hugging Face Transformers
API models	OpenAI, Anthropic
Project profiles	Python, .NET, C++ GoogleTest, Jest, Custom
Credential storage	SQLite + encrypted API keys
Password storage	Salted PBKDF2-HMAC-SHA256 hashes
Repo safety	Relative file tools and command allowlists

🖥️ Web UI

Start the UI:

python -m src.frontend.debug_ui

The dashboard stores useful local state:

Stored item	Purpose
Last repo	Pre-fills the next run
Target coverage	Reuses your preferred threshold
Provider/model	Keeps your LLM setup
API keys	Encrypted locally for OpenAI/Anthropic
Run history	Shows recent results

Local files are created under:

.agentic_testing/
├── app.db
└── secret.key

Keep this folder private. It is ignored by Git.

🧪 Project Profiles

The selected profile controls the default coverage command and the test style guidance sent to the agents.

Profile	Test style	Coverage command
Python / pytest	pytest tests	pytest --cov=. --cov-report=term-missing
.NET	xUnit, NUnit, or MSTest	dotnet test /p:CollectCoverage=true /p:CoverletOutputFormat=opencover
C++ / GoogleTest	TEST / TEST_F cases	gcovr --txt
JS/TS / Jest	Jest tests	npm test -- --coverage
Custom	Existing repo conventions	Editable command

The command is still editable in the UI for custom build systems.

🤖 Providers

Use a hosted model when you want stronger reasoning, or a local model when you want everything to stay on your laptop.

Provider	Install	Notes
Fake	Built in	Dry runs and tests
OpenAI	pip install openai	API key saved encrypted in UI
Anthropic	pip install anthropic	API key saved encrypted in UI
llama.cpp	pip install llama-cpp-python	Uses local GGUF model paths
Hugging Face	pip install transformers torch accelerate	Uses model ids and local cache

⚙️ CLI

The UI is the main workflow, but a CLI exists for automation:

python -m src.agentic_testing_cli /path/to/repo 80 --provider fake

Examples:

python -m src.agentic_testing_cli /path/to/repo 85 \
--provider openai \
--model gpt-4.1-mini
python -m src.agentic_testing_cli /path/to/repo 80 \
--provider llama.cpp \
--model-path /path/to/model.gguf

🧱 Architecture

The code is split between the app shell, the agentic testing loop, provider integrations, and reusable database/cache infrastructure.

src/
├── app/ # Local user, credentials, preferences
├── core/ # Coverage loop, workspace tools, project profiles
├── database/ # Database adapters and cache layer
├── frontend/ # Web UI, agents, LLM providers
├── agentic_testing_cli.py
└── local_agent_cli.py

Core modules:

Module	Responsibility
src/core/agentic_testing.py	Main analyzer/planner/implementer/reviewer loop
src/core/coverage.py	Coverage command runner and parser
src/core/workspace.py	Safe file and command operations
src/core/project_profiles.py	Project profile presets
src/app/storage.py	SQLite app state
src/app/security.py	Password hashing and API-key encryption

✅ Testing

pytest

Current coverage includes:

Area	Covered
UI parsing/rendering	Yes
User/password storage	Yes
Encrypted API keys	Yes
Provider factory	Yes
Coverage parsing	Yes
Workspace safety	Yes
Database/cache adapters	Yes

🔐 Security Notes

The UI stores secrets locally because API-backed providers need reusable credentials. Passwords use a different model: they are never encrypted for later recovery, only hashed for verification.

Data	Storage
Passwords	Salted PBKDF2-HMAC-SHA256 hash
API keys	Fernet-encrypted ciphertext
App DB	.agentic_testing/app.db
Encryption key	.agentic_testing/secret.key

API keys must be decryptable so the selected provider can use them. Passwords are never decrypted because they are not stored reversibly.

📄 License

Released under the MIT License.