Praxis

Praxis is a CI reliability gate for AI agents. The MVP attacks every pull request with agent-specific task evals, runs checked-in regression cases, and surfaces generated failures for review.

The current implementation is a scaffolded MVP:

• Python CLI distribution praxis-ai in packages/praxis-python
• Next.js local dashboard in apps/dashboard
• Deterministic black-box stubs for introspection, stress generation, and RCA
• GitHub Action scaffolding for PR comments

Quick Start

make install
make test

Once published, install the CLI with:

pipx install praxis-ai
praxis --help

Run against a client repo with deterministic local behavior:

uv run praxis init --repo /path/to/demo-client-agent
uv run praxis run --repo /path/to/demo-client-agent --stub
uv run praxis generate-agent-tests --repo /path/to/demo-client-agent

Start the dashboard:

make dashboard-install
make dashboard

MVP Scope

Praxis currently treats introspection, adversarial generation, and RCA as replaceable stubs. The first real adapter targets the Python/LangGraph demo-client-agent.

Release

The PyPI distribution name is praxis-ai because praxis is already occupied on PyPI. The console command remains praxis.

Publishing uses PyPI Trusted Publishing from .github/workflows/publish.yml. Configure the PyPI project praxis-ai to trust this repository, workflow publish.yml, and environment pypi, then publish a GitHub release such as v0.1.0.