Skip to content

Examples

Last Updated: 2026-07-11 Version: v0.2.0

**Last Updated: 2026-06-22

Comprehensive examples for all Cognitive Brain components.

Table of Contents

  1. Basic Usage
  2. Meta-Learning
  3. Pattern Store
  4. Safety Monitoring
  5. Benchmarking

Basic Usage

Execute Simple Task

from github.agents.core.universal_intelligence import (
 UniversalTaskInterface,
 TaskSpec,
)

spec = TaskSpec(
 environment="gridworld",
 initial_state={"x": 0, "y": 0, "goal": {"x": 5, "y": 5}},
 reward_spec={"id": "reward:v1"},
 termination={"max_steps": 50},
 seed=12345
)

uti = UniversalTaskInterface(seed=12345)
result = uti.execute_task(spec, use_adapter=True)

Meta-Learning

MAML Adaptation

from github.agents.core.universal_intelligence import MetaPolicyRouter

router = MetaPolicyRouter(seed=12345)

# Prepare task data
task_data = [(x, x**2) for x in range(10)] # pragma: allowlist secret

# Adapt with MAML
adapted = router.adapt_with_maml("regression_task", task_data) # pragma: allowlist secret

# Use adapted parameters
print(f"Adapted params: {adapted}")

Reptile Adaptation

# Adapt with Reptile (simpler, more stable)
adapted = router.adapt_with_reptile("regression_task", task_data) # pragma: allowlist secret

Strategy Selection

# Get probability distribution
probs = router.get_probability_distribution()

# Measure (collapse superposition)
selected = router.measure()

# Get hyperparameters
hyperparams = router.get_hyperparams(selected)

Pattern Store

Store and Retrieve Patterns

from github.agents.core.universal_intelligence import (
 UniversalPatternStore,
 Pattern,
)

store = UniversalPatternStore()

# Store pattern
pattern = Pattern(
 id="nav_pattern_1",
 domain="gridworld",
 payload={"strategy": "shortest_path", "cost": "manhattan"},
 version="1.0.0",
)
pattern_id = store.store_pattern(pattern)

# Retrieve by similarity
query_pattern = Pattern(
 id="query",
 domain="gridworld",
 payload={"strategy": "path_finding"},
)
similar = store.retrieve_by_similarity(query_pattern, threshold=0.7)

# Cross-domain retrieval
cross_domain = store.retrieve_cross_domain("gridworld", limit=5)

Pattern Versioning

# Update pattern version
pattern_v2 = Pattern(
 id="nav_pattern_1",
 domain="gridworld",
 payload={"strategy": "a_star", "heuristic": "euclidean"},
 version="2.0.0",
)
store.store_pattern(pattern_v2)

# Deprecate old version
old_pattern = store.retrieve_patterns(pattern_ids=["nav_pattern_1_v1"])[0]
old_pattern.deprecated = True

Safety Monitoring

Basic Safety Checks

from github.agents.core.universal_intelligence import SafetyMonitor

monitor = SafetyMonitor(
 neg_transfer_threshold=0.05,
 forgetting_threshold=0.20
)

# Set baseline for domain
monitor.set_baseline("navigation", 0.85)

# During execution, check safety
current_performance = 0.75

if monitor.detect_negative_transfer("navigation", current_performance):
 print(" Negative transfer detected!")
 if monitor.trigger_rollback("navigation"):
 print(" Rolled back to baseline")

if monitor.detect_forgetting("navigation", current_performance):
 print(" Forgetting detected!")
 monitor.isolate_domain("navigation")
 print(" Domain isolated")

Safety Constraints

# Check if domain is isolated
if monitor.is_domain_isolated("navigation"):
 print("Domain is in quarantine")

# Get safety report
report = monitor.get_safety_report()
print(f"Isolated domains: {report['isolated_domains']}")
print(f"Rollback count: {report['rollback_count']}")

Benchmarking

Run EXP-10 Benchmark

from github.agents.core.universal_intelligence import (
 EXP10BenchmarkHarness,
 UniversalTaskInterface,
)

# Create harness
harness = EXP10BenchmarkHarness(seed=12345)

# Create UTI
uti = UniversalTaskInterface(seed=12345)

# Run benchmark
results = harness.run_benchmark(uti)

# Validate k₁ target
k1_valid = harness.validate_k1_target(results)
print(f"k₁ validation: {' PASS' if k1_valid else ' FAIL'}")

# Test transfer
zero_shot_valid = harness.test_zero_shot_transfer(uti)
few_shot_valid = harness.test_few_shot_transfer(uti, k=10)

print(f"Zero-shot: {' PASS' if zero_shot_valid else ' FAIL'}")
print(f"Few-shot: {' PASS' if few_shot_valid else ' FAIL'}")

# Export metrics
harness.export_metrics(results, ".github/agents/metrics/phase8_7")

Advanced Examples

See Jupyter notebooks in examples/notebooks/: 1. 01_quickstart.ipynb - Basic usage 2. 02_meta_learning.ipynb - Deep dive into MAML/Reptile 3. 03_safety_monitoring.ipynb - Safety mechanisms 4. 04_pattern_store.ipynb - Pattern management 5. 05_benchmarking.ipynb - EXP-10 validation


Next: Architecture