Template for building layered orchestration systems using the LEGO test pattern
Install
npx skillscat add foundups/foundups-agent/agent-skills-orchestration-template Install via the SkillsCat registry.
SKILL.md
Orchestration SKILLz Template
Build automation systems using layered, composable test modules that snap together like LEGO blocks.
First Principles
- Layer Independence: Each layer tests one capability in isolation
- Progressive Complexity: L0 → L1 → L2 → L3 builds up to full chain
- Snap Together: Full chain combines all layers
- Occam's Build: Start simple, add layers only when needed
Layer Architecture
L0: Entry → Navigate, find target (e.g., find unlisted videos)
L1: Filter → Apply filters/constraints (e.g., unlisted filter)
L2: Edit → Modify/interact with target (e.g., enhance metadata)
L3: Execute → Complete action (e.g., schedule, publish)
Full Chain → L0 + L1 + L2 + L3 snapped togetherDirectory Structure
module/
├── src/
│ └── orchestrator.py # Main DAE entry point
├── tests/
│ ├── test_layer0_entry.py
│ ├── test_layer1_filter.py
│ ├── test_layer2_edit.py
│ ├── test_layer3_execute.py
│ └── test_full_chain.py
└── scripts/
└── launch.py # Menu + CLI interfaceImplementation Pattern
1. Build Each Layer Independently
# test_layer0_entry.py
def test_layer0():
"""L0: Can we navigate to the target?"""
driver = connect_browser()
result = navigate_to_target(driver)
assert result.success, "L0 failed: could not reach target"2. Chain Layers Together
# test_full_chain.py
def test_full_chain():
"""Full chain: L0 → L1 → L2 → L3"""
driver = connect_browser()
# L0: Entry
navigate_to_target(driver)
# L1: Filter
apply_filters(driver)
# L2: Edit
modify_target(driver)
# L3: Execute
complete_action(driver)3. Create Menu with Layer Access
def show_menu():
print("1. Run Full Chain")
print("2. Preview Only (DRY RUN)")
print("D. Dev Tests (L0-L3)")When to Use This Pattern
- Browser automation (Selenium/Playwright)
- Multi-step workflows (schedule, publish, moderate)
- API orchestration with validation gates
- Any system requiring incremental testing
Success Metrics
- Each layer passes independently
- Full chain completes without manual intervention
- 0102 can use simplified menu (dev tests hidden)