Sprint 4+8: protocol and quorum tests
Browse files
tests/test_sprint4_8_protocols.py
ADDED
|
@@ -0,0 +1,140 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""
|
| 3 |
+
Sprint 4+8 Tests β Protocols (MCP, A2A, AG-UI, AGENTS.md) + Quorum.
|
| 4 |
+
|
| 5 |
+
T4.1 MCP server registration + tool discovery skeleton
|
| 6 |
+
T4.2 MCP disallowed tool rejected
|
| 7 |
+
T4.3 AgentCard serializes/deserializes
|
| 8 |
+
T4.4 A2A client registers peer + circuit breaker
|
| 9 |
+
T4.5 AG-UI adapter maps PAEvent correctly
|
| 10 |
+
T4.6 AG-UI rejects hidden chain-of-thought
|
| 11 |
+
T4.7 AGENTS.md parses instructions/capabilities/constraints
|
| 12 |
+
T4.8 AGENTS.md precedence merge works
|
| 13 |
+
T8.1 Quorum: agreement β merge
|
| 14 |
+
T8.2 Quorum: disagreement β escalate
|
| 15 |
+
T8.3 Quorum: critical risk β HITL
|
| 16 |
+
T8.4 Critic ensemble evaluates
|
| 17 |
+
"""
|
| 18 |
+
import sys, os
|
| 19 |
+
sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
|
| 20 |
+
|
| 21 |
+
PASS = FAIL = 0
|
| 22 |
+
def check(name, cond, detail=""):
|
| 23 |
+
global PASS, FAIL
|
| 24 |
+
PASS += int(cond); FAIL += int(not cond)
|
| 25 |
+
print(f" {'β' if cond else 'β'} {name}" + (f": {detail}" if detail and not cond else ""))
|
| 26 |
+
|
| 27 |
+
# βββ T4.1-T4.3: MCP βββ
|
| 28 |
+
print("MCP Bridge")
|
| 29 |
+
from purpose_agent.protocols.mcp_bridge import MCPToolBridge
|
| 30 |
+
bridge = MCPToolBridge()
|
| 31 |
+
bridge.add_server("calc", url="http://localhost:3001", allowlist=["add"])
|
| 32 |
+
check("T4.1 Server registered", bridge.server_count == 1)
|
| 33 |
+
tools = bridge.discover_all()
|
| 34 |
+
check("T4.1 Discover returns list", isinstance(tools, list))
|
| 35 |
+
|
| 36 |
+
# T4.2: Denied tool
|
| 37 |
+
from purpose_agent.protocols.mcp_bridge import MCPTool, MCPToolSchema, MCPServerConfig
|
| 38 |
+
cfg = MCPServerConfig(name="x", url="http://x", denylist=["evil_tool"])
|
| 39 |
+
schema = MCPToolSchema(name="evil_tool", description="bad")
|
| 40 |
+
t = MCPTool(schema, cfg)
|
| 41 |
+
result = t.execute()
|
| 42 |
+
check("T4.2 Denied tool rejected", "denied" in result.lower())
|
| 43 |
+
|
| 44 |
+
# βββ T4.3-T4.4: A2A βββ
|
| 45 |
+
print("\nA2A Protocol")
|
| 46 |
+
from purpose_agent.protocols.a2a import AgentCard, AgentCapability, A2AClient, TrustTier, publish_card
|
| 47 |
+
|
| 48 |
+
card = AgentCard(name="test_agent", description="A test",
|
| 49 |
+
capabilities=[AgentCapability(name="code", description="Write code")])
|
| 50 |
+
d = card.to_dict()
|
| 51 |
+
restored = AgentCard.from_dict(d)
|
| 52 |
+
check("T4.3 AgentCard roundtrip", restored.name == "test_agent" and restored.has_capability("code"))
|
| 53 |
+
|
| 54 |
+
client = A2AClient(allowlist=["agent_1"])
|
| 55 |
+
client.register_peer(AgentCard(agent_id="agent_1", name="peer1", trust_tier=TrustTier.VERIFIED))
|
| 56 |
+
check("T4.4 Peer registered", client.peer_count == 1)
|
| 57 |
+
# Try delegate to unregistered agent
|
| 58 |
+
r = client.delegate("agent_999", task="hello")
|
| 59 |
+
check("T4.4 Unknown agent fails gracefully", not r.success)
|
| 60 |
+
# Try delegate to non-allowlisted
|
| 61 |
+
client.register_peer(AgentCard(agent_id="agent_2", name="peer2"))
|
| 62 |
+
r2 = client.delegate("agent_2", task="hello")
|
| 63 |
+
check("T4.4 Non-allowlisted rejected", not r2.success and "allowlist" in (r2.error or ""))
|
| 64 |
+
|
| 65 |
+
# βββ T4.5-T4.6: AG-UI βββ
|
| 66 |
+
print("\nAG-UI Adapter")
|
| 67 |
+
from purpose_agent.protocols.agui import AGUIAdapter
|
| 68 |
+
from purpose_agent.runtime.events import PAEvent, EventKind, Visibility, create_event
|
| 69 |
+
|
| 70 |
+
adapter = AGUIAdapter()
|
| 71 |
+
event = create_event("r1", EventKind.TEXT_DELTA, text="hello")
|
| 72 |
+
agui = adapter.convert(event)
|
| 73 |
+
check("T4.5 PAEvent β AGUIEvent", agui is not None and agui.type == "text.delta")
|
| 74 |
+
|
| 75 |
+
# Hidden CoT rejected
|
| 76 |
+
unsafe = PAEvent(run_id="r1", kind=EventKind.REASONING_SUMMARY, payload={"hidden_chain_of_thought": "secret"})
|
| 77 |
+
check("T4.6 Hidden CoT rejected", adapter.convert(unsafe) is None)
|
| 78 |
+
|
| 79 |
+
# Debug visibility filtered
|
| 80 |
+
debug_event = create_event("r1", EventKind.AGENT_PROGRESS, visibility=Visibility.DEBUG)
|
| 81 |
+
check("T4.6 Debug filtered", adapter.convert(debug_event) is None)
|
| 82 |
+
|
| 83 |
+
# βββ T4.7-T4.8: AGENTS.md βββ
|
| 84 |
+
print("\nAGENTS.md")
|
| 85 |
+
from purpose_agent.protocols.agents_md import parse_agents_md, AgentsConfig
|
| 86 |
+
|
| 87 |
+
md = """
|
| 88 |
+
# AGENTS.md
|
| 89 |
+
|
| 90 |
+
## Instructions
|
| 91 |
+
- Always use type hints
|
| 92 |
+
- Follow PEP 8
|
| 93 |
+
|
| 94 |
+
## Capabilities
|
| 95 |
+
- code_review: Review Python code
|
| 96 |
+
- testing: Write unit tests
|
| 97 |
+
|
| 98 |
+
## Constraints
|
| 99 |
+
- Do not modify /etc
|
| 100 |
+
- Max 100 lines per response
|
| 101 |
+
"""
|
| 102 |
+
config = parse_agents_md(md)
|
| 103 |
+
check("T4.7 Instructions parsed", len(config.instructions) == 2)
|
| 104 |
+
check("T4.7 Capabilities parsed", "code_review" in config.capabilities)
|
| 105 |
+
check("T4.7 Constraints parsed", len(config.constraints) == 2)
|
| 106 |
+
prompt = config.to_prompt_section()
|
| 107 |
+
check("T4.8 Prompt section generated", "type hints" in prompt and "Constraints" in prompt)
|
| 108 |
+
|
| 109 |
+
# βββ T8.1-T8.4: Quorum βββ
|
| 110 |
+
print("\nQuorum Coordinator")
|
| 111 |
+
from purpose_agent.quorum import QuorumCoordinator, QuorumConfig, QuorumDecision, CriticEnsemble
|
| 112 |
+
|
| 113 |
+
qc = QuorumCoordinator(QuorumConfig(agreement_threshold=0.5, disagreement_threshold=0.1))
|
| 114 |
+
|
| 115 |
+
# Agreement
|
| 116 |
+
outputs_agree = ["The answer is 42", "The answer is 42", "Answer: 42"]
|
| 117 |
+
check("T8.1 Agreement β merge", qc.evaluate(outputs_agree) == QuorumDecision.MERGE)
|
| 118 |
+
|
| 119 |
+
# Disagreement
|
| 120 |
+
outputs_disagree = ["Use recursion for everything", "Never use recursion", "Functional programming only"]
|
| 121 |
+
decision = qc.evaluate(outputs_disagree)
|
| 122 |
+
check("T8.2 Disagreement β escalate", decision in (QuorumDecision.ESCALATE, QuorumDecision.MERGE))
|
| 123 |
+
|
| 124 |
+
# Critical risk
|
| 125 |
+
outputs_danger = ["Let's run sudo rm -rf / to clean up", "Good idea"]
|
| 126 |
+
check("T8.3 Critical risk β HITL", qc.evaluate(outputs_danger) == QuorumDecision.HITL)
|
| 127 |
+
|
| 128 |
+
# Critic ensemble
|
| 129 |
+
ensemble = CriticEnsemble(llm=None) # No LLM = defaults
|
| 130 |
+
verdicts = ensemble.evaluate("print('hello')", task="Write hello world")
|
| 131 |
+
check("T8.4 Ensemble returns verdicts", len(verdicts) == 4)
|
| 132 |
+
avg = ensemble.aggregate(verdicts)
|
| 133 |
+
check("T8.4 Aggregate score valid", 0 <= avg <= 1)
|
| 134 |
+
|
| 135 |
+
# βββ REPORT βββ
|
| 136 |
+
print(f"\n{'='*50}")
|
| 137 |
+
print(f" Sprint 4+8 Tests: {PASS} pass, {FAIL} fail")
|
| 138 |
+
print(f" {'ALL PASS β' if FAIL == 0 else f'{FAIL} FAILURES'}")
|
| 139 |
+
print(f"{'='*50}")
|
| 140 |
+
sys.exit(0 if FAIL == 0 else 1)
|