Getting Started¶
Install Assay and choose a first test with explicit inputs.
For the reproducible release contract, use the release-pinned agent golden path. This page is the shorter human introduction.
Overview¶
This guide covers:
- Installation — Install the Assay CLI
- Quick Start — Import a trace and run your first test
- Your First Test — Write a custom policy from scratch
- CI Integration — Add Assay to GitHub Actions / GitLab CI
- Operator Proof Flow — See trace ingest, shipped control-evidence linting, and proof-kit verification as one flow
Prerequisites¶
- Rust 1.96 for repository development, Rust 1.89+ for public-crate source installs, or CPython 3.12, 3.13, or 3.14 for Python SDK use
- CPython 3.12, 3.13, and 3.14 on macOS x86_64/arm64 and Linux x86_64; other interpreters and platforms are not claimed.
- For the import example below: an Inspector session saved as
session.jsonand aneval.yamlconfigured for that trace - For a self-contained local mock exercise, use Your First Test; it does not need an Inspector export
Import an Existing Session¶
# Install
cargo install assay-cli --version 6.19.1 --locked
# Import your existing MCP session as a trace
mkdir -p traces
assay import --format inspector session.json --out-trace traces/session.jsonl
# Run tests
assay run --config eval.yaml --trace-file traces/session.jsonl
# Add to CI
# Copy the GitHub Action from ci-integration.md
This runs the recorded evaluation path. A clean result still depends on the supplied config and trace being complete.
What You'll Learn¶
By the end of this guide, you'll understand:
| Concept | What it does |
|---|---|
| Traces | Recorded agent behavior (the "golden" reference) |
| Policies | Rules that define correct behavior |
| Metrics | Functions that validate output |
| Replay | Reuses recorded outputs; repeatability depends on fixed inputs and deterministic evaluators. Live evaluators can still make API calls. |
Next Steps¶
-
Installation
Choose a CLI installation route and follow its verification instructions.
-
Quick Start
Run a local mock test or import an existing trace.
-
Operator Proof Flow
See one compact operator story: import, shipped pack evaluation, and offline release verification.