Tester Agent

The Tester agent role, test analysis, and optional sandbox test execution.

The Tester Agent (◉)

The Tester produces an analytical TestReport for the Reviewer. Separately, the orchestrator attempts to resolve the configured or auto-detected project test command and execute it in the active sandbox before branch creation.

What the Tester Does

1. Test Analysis: Describes unit and integration cases for the new functionality, edge cases, and boundary conditions in artifacts/tester.json.
2. Command Resolution: Uses [agents.tester].test_command when set; otherwise it detects commands such as npm test, cargo test --locked, pytest, or go test ./....
3. Optional Real Execution: Runs the resolved command in the container or host-local worktree and records the command, exit code, stdout, and stderr in artifacts/test_execution.json.
4. Conditional Evidence: If no command resolves, test_execution is None; the report omits the real-execution section rather than claiming a pass.

A resolved test or mutation failure blocks branch creation unless --force is used, and a forced branch is recorded as unverified.

Cost Optimization Tip

Because test analysis is often suitable for a faster, lower-cost model, you can use one here while keeping stronger models for implementation and review:

[agents.tester]
provider    = "openai"
model       = "gpt-4o-mini"
max_tokens  = 4096
temperature = 0.1