Behavioral Evals

Overview

Behavioral evaluations (evals) are tests that validate the agent’s decision-making (e.g., tool choice) rather than pure functionality. They are critical for verifying prompt changes, debugging steerability, and preventing regressions.

[!NOTE] Single Source of Truth: For core concepts, policies, running tests, and general best practices, always refer to evals/README.md.

---

🔄 Workflow Decision Tree

1. Does a prompt/tool change need validation?
   • No -> Normal integration tests.
   • Yes -> Continue below.
2. Is it UI/Interaction heavy?
   • Yes -> Use appEvalTest (AppRig). See creating.md (opens in a new tab).
   • No -> Use evalTest (TestRig). See creating.md (opens in a new tab).
3. Is it a new test?
   • Yes -> Set policy to USUALLY_PASSES.
   • No -> ALWAYS_PASSES (locks in regression).
4. Are you fixing a failure or promoting a test?
   • Fixing -> See fixing.md (opens in a new tab).

---

📋 Quick Checklist

1. Setup Workspace

Seed the workspace with necessary files using the files object to simulate a realistic scenario (e.g., NodeJS project with package.json).

• Details in creating.md (opens in a new tab)

2. Write Assertions

Audit agent decisions using rig.setBreakpoint() (AppRig only) or index verification on rig.readToolLogs().

• Details in creating.md (opens in a new tab)

3. Verify

Run single tests locally with Vitest. Confirm stability locally before relying on CI workflows.

• See evals/README.md for running commands.

---

📦 Bundled Resources

Detailed procedural guides:

• creating.md (opens in a new tab): Assertion strategies, Rig selection, Mock MCPs.
• fixing.md (opens in a new tab): Step-by-step automated investigation, architecture diagnosis guidelines.
• promoting.md (opens in a new tab): Candidate identification criteria and threshold guidelines.