Concepts Validation Types

📖 This page mirrors the canonical Agent Karma docs. Edits should be made there, not in the wiki.

Validation types

The four validations Agent Karma understands, and how it records them.

Agent Karma recognizes the validations that actually catch AI mistakes:

Type	Catches
Test	Logic bugs, broken behavior, regressions
Build	Compile errors, broken imports
Lint	Style/quality issues, common foot-guns
Type Check	Type errors, wrong shapes, bad API usage

(It also understands Security scans and an Other bucket.)

How a validation gets recorded

• Automatic — when you run a validation in VS Code's integrated terminal, shell integration detects the command type and result. The raw command string is classified, then discarded.
• Manual — Add Validation Command (or the @agentkarma /verify chat command, or the ✅ Log a Validation button) records one when auto-detection can't see it — e.g. AI run in a browser, or commands run in an external terminal.

Why "ran" matters more than "passed"

A check only earns points if it actually ran. A failed test still earns the "tests run" credit (you looked) — but you don't get the "tests passed" bonus. You can't score well by simply never testing. That's the whole philosophy: validation discipline, honestly measured.