The Tester Agent

The tester agent runs all test suites against code produced by the developer agent, analyzes coverage, investigates failures, and produces a structured QA report. Cook will not proceed to code review or finalization unless the tester reports 100% of tests passing.

Role

The tester agent’s job is verification, not implementation. It executes existing test commands, reads output, identifies root causes of failures, and reports findings. It owns only test files — it reads implementation files but never edits them.

When the Tester Is Invoked

Cook spawns the tester automatically after each implementation phase completes. You can also invoke it directly via the skill:

/gk:test "Run all tests for the auth module and report coverage"

Or for UI verification:

/gk:test ui http://localhost:3000/login

What the Tester Executes

The tester identifies the project’s test runner from package.json or project config and runs the appropriate commands:

# JavaScript / TypeScript
npm test
npm run test:coverage

# Python
pytest
pytest --cov=src --cov-report=term-missing

# Go
go test ./...

# Rust
cargo test

For frontend work, the tester activates the /gk:chrome-devtools skill for visual regression, responsive layout checks, and accessibility audits.

Test Scope

The tester validates three layers:

Analyzing Failures

When tests fail, the tester does not immediately report back. It first:

1. Reads the full failure output, including stack traces
2. Identifies whether the failure is in the test itself or the implementation
3. Checks if the failure is deterministic or environment-related
4. Activates the /gk:debug skill for complex failures requiring deeper trace analysis
5. Activates /gk:sequential-thinking for failures with unclear root causes

Only after this analysis does it produce its report, with a clear recommendation for the developer agent.

QA Report Format

The tester saves a structured report to plans/reports/:

## QA Report — Auth Module

### Test Results
- Total: 24 tests
- Passed: 22
- Failed: 2
- Coverage: 87%

### Failed Tests
1. `auth.test.ts:45` — login with expired token should return 401
   Root cause: jwt-verify.ts line 23 catches all errors, swallows expiry
   Fix needed: Check err.name === 'TokenExpiredError' separately

2. `auth.test.ts:67` — logout should invalidate session
   Root cause: Session store not cleared on logout endpoint
   Fix needed: Add await sessionStore.delete(token) in logout handler

### Recommendations
- Fix jwt-verify.ts error handling before proceeding
- Add session cleanup to logout route

File Ownership

The tester agent may create new test files when coverage gaps are found, but it must declare these in its task ownership:

File ownership (tester):
- src/**/*.test.ts (create/edit)
- src/**/*.spec.ts (create/edit)
- tests/**/* (create/edit)

It reads (but never writes) implementation files to understand what needs testing.

Next Steps

• The Code Reviewer Agent — quality review after tests pass
• Skills: Using Skills — how to invoke /gk:test with flags directly