Purpose

Purpose

Runs the Four-Layer Minimal Repro ladder to reproduce PR blockers quickly with evidence-backed classification.

Primary source is the existing command definition and protocol companion:

• .claude/commands/4layer.md
• .claude/skills/pr-blocker-min-repro.md

Minimal Repro Ladder

Run tests in this order and stop at the first layer that conclusively reproduces the blocker:

1. Unit tests ($PROJECT_ROOT/tests/)

./vpython -m pytest $PROJECT_ROOT/tests/test_[relevant].py -q

2. End-to-end tests ($PROJECT_ROOT/tests/test_end2end/)

./vpython -m pytest $PROJECT_ROOT/tests/test_end2end/test_[feature]_end2end.py -q

3. MCP/HTTP API tests (testing_mcp/)

./vpython testing_mcp/[domain]/test_[feature]_real.py

4. Browser tests (testing_ui/)

./vpython testing_ui/[domain]/test_[feature]_browser.py

Execution Rules

1. Identify blocker and target only relevant, minimal tests.
2. Start at Layer 1. Move upward only if the current layer passes.
3. Keep provider/user isolation so parallel runs do not collide.
4. Record absolute evidence paths after each executed layer.
5. Report the first layer that reproduces the issue; do not run unnecessary upper layers.

Evidence Requirements

After each test run, capture:

• Full absolute evidence directory path (for example /tmp/worldarchitectai/<branch>/<test>/latest/).
• Signature failure lines with context (rg/grep output).
• Screenshot + server log consistency checks where applicable.
• Final decision notes mapping blocker layer to component class:
  • Unit fail → backend logic
  • End2end fail → integration/API
  • MCP fail → server protocol
  • Browser fail → UI/frontend

References

• .claude/skills/pr-blocker-min-repro.md for BYOK-specific starter commands and bead note patterns.
• .claude/skills/integration-verification.md for minimum evidence completeness.