Skip to content

Agent 5: QA Agent

Property Value
Tier Utility (Tests) / Production (Analysis)
Trigger PR passes code review
Purpose Run comprehensive test suite including compatibility matrix
Est. Cost 5k - 20k tokens for analysis per task

Testing Scope

Test type Tool What it checks
Unit tests CI (pytest) Function-level correctness
Integration tests CI (GitHub Actions) Module interaction, real I/O
Type checking CI (mypy/pyright) Type safety across public API
Compatibility matrix GitHub Actions matrix Python 3.10, 3.11, 3.12, 3.13
Backward compat CI (GitHub Actions) Import old version, run against new — no breaks
Dependency audit pip-audit Known vulnerabilities in deps
Documentation build CI Docs compile without errors, examples run

Compatibility Matrix

strategy:
  matrix:
    python-version: ['3.10', '3.11', '3.12', '3.13']
    os: [ubuntu-latest, macos-latest]

QA Report

After test execution, Claude Sonnet analyses results:

## QA report — #17 RetryPolicy implementation

**Test results**: 47 passed, 0 failed, 2 skipped
**Coverage**: 91% on new code (target: 80%) ✅
**Compatibility**: All matrix combinations green ✅
**Type check**: mypy strict — clean ✅
**Backward compat**: v0.x import test — passed ✅

**Risk assessment**: Low
**Recommendation**: Ready for release candidate

**Skipped tests rationale**:
- `test_retry_with_real_server`: Requires network (CI-skip expected)
- `test_retry_windows_specific`: Platform-specific, no Windows in matrix

Failure Escalation

Tests fail
  → CI provides failure output
  → Claude Sonnet analyses: code bug or test environment issue?
    → Code bug: Create sub-issue, route back to dev agent
    → Environment issue: Fix CI config, re-run
    → Flaky test: Label "flaky", investigate pattern across last 5 runs
  → If > 3 bugs per task: escalate to human (likely spec problem)