agentic-qe

Agentic QE

Quality engineering tools for your coding agent. Generate tests, investigate coverage gaps, analyze flaky behavior and coordinate quality checks across the software lifecycle. Keep the findings close to the code and use them to make a better quality decision.

Start with one task · Choose your client · Inspect the evidence · Contribute · Release notes

Put AQE to work

Supports framework-specific output for tools including Jest, Vitest, Playwright, Cypress, pytest and JUnit. Review generated tests and run them in your environment; an agent’s recommendation is an input to your team’s decision.

Start with one task

Requires Node.js ≥22.13.0 and npm ≥10.0.0. Use the coding agent/client you already have configured.

npm install -g agentic-qe
cd your-project
aqe init --auto
aqe health

npm 12+: dependency install scripts are blocked unless allowed, and AQE’s SQLite store (better-sqlite3) needs its install script to fetch its native binary. Name the packages when you install:

npm install -g agentic-qe --allow-scripts=agentic-qe,better-sqlite3,hnswlib-node

Include agentic-qe in the command — npm’s warning suggests npm install -g --allow-scripts=... without a package name, which installs the current directory instead. If aqe init reports that the better-sqlite3 native module could not be loaded on an existing install, rebuild it with npm rebuild -g better-sqlite3 --allow-scripts=better-sqlite3 (for a project-local install: npm install-scripts approve better-sqlite3 && npm rebuild better-sqlite3). hnswlib-node is optional; AQE runs without it.

aqe init --auto analyzes your project and configures the default Claude Code surface, including MCP. Review the generated project files. For another client, select its setup flag below and use its MCP connection flow.

In your coding agent, start with a narrow request:

Generate tests for src/services/UserService.ts.
Explain the risks and edge cases each test covers.
Run the tests if execution is available, and report failures and anything you could not verify.

Then ask for a second view:

Find coverage gaps in src/services/ and prioritize them by risk.
Challenge weak assertions in the generated tests and explain what still needs human review.

The paths are examples; use a real file in your project. Success means tests you can inspect and execute, plus a clear account of what remains untested. A coverage target is a request, not a promised result.

Choose your client

These are implemented setup paths, not a claim of identical behavior across clients. The default Claude Code files are also created unless you add --no-claude when targeting another client.

# Configure every supported client
aqe init --auto --with-all-platforms

# Inspect or add client configuration later
aqe platform list
aqe platform setup cursor
aqe platform verify cursor

# Standalone MCP server entry point
aqe-mcp

See the Platform Setup Guide for client-specific configuration.

Claude Code plugin: a scoped alternative

Use the agentic-qe-fleet plugin when you want a smaller Claude Code surface with slash commands. The full aqe init path adds broader project configuration and persistent learning setup.

In Claude Code:

/plugin install agentic-qe-fleet --marketplace proffesor-for-testing/agentic-qe

Or load it from a checkout:

git clone https://github.com/proffesor-for-testing/agentic-qe.git
claude --plugin-dir ./agentic-qe/plugins/agentic-qe-fleet

Start with /aqe-fleet-status, /aqe-generate src/services/Auth.ts or /aqe-analyze src/. The plugin registers its own MCP server, pinned to the matching package version (npx -y agentic-qe@<plugin version> mcp). See the plugin README for the bundled agents, commands and skills.

aqe health
# Look for "HNSW backend: native" (hnswlib-node) or "HNSW backend: js"
# (ProgressiveHnswBackend — see caveat above).

How a quality task flows

The qe-queen-coordinator decomposes a quality assessment into domain tasks and combines findings. Domains cover test generation and execution, coverage, quality assessment, defect intelligence, requirements, code intelligence, security, contracts, visual accessibility, resilience, learning and enterprise integration.

Patterns can be stored and retrieved across sessions. AQE provides outcome feedback and consolidation mechanisms; retrieval alone does not prove that future tests are better. Model routing offers cost-aware task selection, while actual costs and quality depend on your provider and workload.

Anti-sycophancy checks examine weak or tautological tests. Adversarial review adds a way to challenge a verdict. See quality gate features and QE Court design.

Inspect the evidence

Inventory verified October 9, 2026 at 829d030: 53 top-level qe-*.md agent definitions + 7 QE subagents; 86 packaged SKILL.md entry points; 13 domain directories. The scoped plugin separately contains 11 agents, 9 skills and 9 command files. Client setup flags above cover 11 clients. These counts exclude platform infrastructure agents and supporting Markdown files.

Commands for everyday work

aqe agent list                  # Agent inventory
aqe fleet status                # Fleet status
aqe health                      # Runtime health
aqe learning stats              # Learning statistics
aqe learning dream              # Pattern consolidation
aqe brain export                # Export learned patterns

aqe code index src/             # Build code intelligence index
aqe code index src/ --incremental
aqe code index . --git-since HEAD~5
aqe code search "authentication"
aqe code impact src/
aqe code deps src/
aqe code complexity src/
aqe code c4 .                   # C4 diagrams with confidence

LLM providers and local models

AQE Fleet’s LLM-enhanced analysis (ADR-043, ADR-051) routes through a HybridRouter that picks providers based on routing rules and your env config. Set one or more API keys and Fleet auto-detects what’s available:

export GEMINI_API_KEY="..."  # or any supported provider
aqe init --auto

Persistent config (aqe llm config --set writes to .agentic-qe/llm-config.json):

aqe llm config --set mode=cost-optimized
aqe llm config --set defaultProvider=gemini
aqe llm config           # show current effective config

API keys are env-only — aqe llm config --set apiKey=... is refused to keep secrets out of project-committed config.

Opting out

To run Fleet without any LLM calls (deterministic-only mode):

export AQE_LLM_ROUTER_DISABLED=1   # env-only kill-switch

Or programmatically: new QEKernelImpl({ llmRouter: { enabled: false } }).

Free local models (opt-in) — cheap-first test generation

Run test generation on a free local model first (local Ollama, cloud Ollama, OpenRouter free models, or any OpenAI-compatible endpoint), with an automatic repair loop, falling back to your normal LLM path only when needed. Off by default — enable with AQE_FREE_TIER=1:

ollama pull qwen3-coder:30b
export AQE_FREE_TIER=1            # opt in (default model: qwen3-coder:30b)

qwen3-coder:30b needs ~18 GB of RAM. qwen3:8b fits smaller machines but measured below the QE test-generation quality floor (ADR-111).

Local inference avoids per-call cloud charges; hardware costs and fallback cloud calls still apply. See the Free-Tier Local Models guide for provider options and configuration.

Documentation

Contribute

Bring a small problem we can reproduce. Useful starting points include a failing test, a client setup issue, a better example, or a validator that challenges an unsupported claim.

Read CONTRIBUTING.md and AGENTS.md before changing code. For bug reports, include your AQE version, Node version, operating system, client, expected behavior and a minimal reproduction. Use Issues for reproducible bugs, Issues for questions and use cases, and the security policy for security reports.

git clone https://github.com/proffesor-for-testing/agentic-qe.git
cd agentic-qe
npm install
npm run build
npm test -- --run

For a focused change, start with the relevant checks documented in AGENTS.md, such as npm run typecheck, npm run lint and the relevant test suite. Keep production learning databases intact and test data changes against copies.

License and support

MIT — see LICENSE. Free to use, fork, build and contribute under its terms.

Support the project · Project documentation · View all contributors

Contributors

@proffesor-for-testing Project Lead	@fndlalit QX Partner, Testability	@shaal Core Development	@mondweep Architecture
@rudycelekli Reliability & Security Fixes	@stuinfla Test Execution Evidence	@nagoodman OpenCode Integration	@Jordi-Izquierdo-DDS RVF/HNSW Fix
@JLMA-Agentic-Ai Package Exports	@amonkarsidhant MCP-first Setup	@gurdasnijor Smithery Integration

Acknowledgments

Agentic QE was developed with help from Ruflo, formerly Claude Flow, by rUv, for coordination during development. RuVector provides the vector and RVF database foundation through the @ruvector/* packages; AQE also uses better-sqlite3 for SQLite persistence. Ruflo is development tooling, not a required AQE runtime dependency.

Thanks also to Agentic Flow for agent patterns and learning systems, and to the maintainers and contributors across the testing tools this project builds on. The cards are original AQE artwork and link to the upstream projects.