Spaces:
Paused
Paused
File size: 11,101 Bytes
287f3d3 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 | """Agent-level tests: structured output handling and failure modes."""
from __future__ import annotations
import json
from agentic_core.agents import (
APIAgent,
ArchitectureAgent,
DatabaseAgent,
DevOpsAgent,
RequirementsAgent,
RevisionInstruction,
)
from agentic_core.llm import LLMProviderError
from tests.helpers import (
api_output,
architecture_output,
database_output,
devops_output,
requirements_output,
)
async def test_requirements_agent_valid_output(provider, llm_service, make_context):
provider.set_responses([json.dumps(requirements_output())])
agent = RequirementsAgent(llm_service)
result = await agent.run(make_context("Food delivery."))
assert result.status == "success"
assert result.output["functional_requirements"]
assert result.error is None
async def test_malformed_json_recovers_via_repair(provider, llm_service, make_context, settings):
provider.set_responses(["this is not json at all", json.dumps(requirements_output())])
agent = RequirementsAgent(llm_service)
result = await agent.run(make_context("Food delivery."))
assert result.status == "success"
assert result.retry_count == 1
assert len(provider.calls) == 2
async def test_invalid_field_type_repaired(provider, llm_service, make_context):
broken = requirements_output()
broken["functional_requirements"] = "not-a-list"
provider.set_responses([json.dumps(broken), json.dumps(requirements_output())])
agent = RequirementsAgent(llm_service)
result = await agent.run(make_context("Food delivery."))
assert result.status == "success"
assert result.retry_count == 1
async def test_persistent_invalid_output_fails(provider, llm_service, make_context, settings):
provider.set_responses(["bad json", "still bad json"])
agent = RequirementsAgent(llm_service)
result = await agent.run(make_context("Food delivery."))
assert result.status == "failed"
assert result.error is not None
assert result.output is None
async def test_llm_failure_marks_agent_failed(provider, llm_service, make_context):
async def boom(_s, _u):
raise LLMProviderError("upstream 500")
provider.set_handler(boom)
agent = RequirementsAgent(llm_service)
result = await agent.run(make_context("Food delivery."))
assert result.status == "failed"
assert "upstream 500" in (result.error or "")
# Transport-level failures are retryable at the orchestrator level.
assert result.retryable is True
async def test_database_agent_receives_architecture_input(provider, llm_service, make_context):
provider.set_responses([json.dumps(database_output())])
agent = DatabaseAgent(llm_service)
context = make_context("Food delivery.")
context.architecture = architecture_output()
await agent.run(context)
user_prompt = provider.calls[0][1]
assert "ARCHITECTURE" in user_prompt
assert "FastAPI" in user_prompt
async def test_agents_include_dependency_artifacts(provider, llm_service, make_context):
agent = APIAgent(llm_service)
context = make_context("Food delivery.")
context.requirements = requirements_output()
context.architecture = architecture_output()
context.database = database_output()
provider.set_responses([json.dumps(api_output())])
await agent.run(context)
user_prompt = provider.calls[0][1]
assert "DATABASE DESIGN" in user_prompt
assert "orders" in user_prompt
async def test_api_prompt_uses_digested_inputs(provider, llm_service, make_context):
"""The api prompt is condensed: it keeps entities but drops verbose fields."""
agent = APIAgent(llm_service)
context = make_context("Food delivery.")
context.requirements = requirements_output()
context.architecture = architecture_output()
context.database = database_output()
provider.set_responses([json.dumps(api_output())])
await agent.run(context)
user_prompt = provider.calls[0][1]
assert "orders" in user_prompt
assert "sql_schema" not in user_prompt
assert "erd_mermaid" not in user_prompt
assert "CREATE TABLE" not in user_prompt
async def test_api_schema_excludes_derived_openapi(provider, llm_service, make_context):
"""The model is never shown openapi_spec in the JSON schema, so it cannot
spend output tokens producing the (derived) OpenAPI document."""
agent = APIAgent(llm_service)
context = make_context("Food delivery.")
context.requirements = requirements_output()
context.architecture = architecture_output()
context.database = database_output()
provider.set_responses([json.dumps(api_output())])
await agent.run(context)
assert "openapi_spec" not in provider.calls[0][1]
async def test_database_schema_excludes_derived_sql_and_erd(provider, llm_service, make_context):
"""sql_schema and erd_mermaid are excluded from the schema the model sees;
the renderer derives them from the entities/fields instead."""
agent = DatabaseAgent(llm_service)
context = make_context("Food delivery.")
context.requirements = requirements_output()
context.architecture = architecture_output()
provider.set_responses([json.dumps(database_output())])
await agent.run(context)
user_prompt = provider.calls[0][1]
assert "sql_schema" not in user_prompt
assert "erd_mermaid" not in user_prompt
async def test_result_reports_token_metrics(provider, llm_service, make_context):
provider.set_responses([json.dumps(requirements_output())])
agent = RequirementsAgent(llm_service)
result = await agent.run(make_context("Food delivery."))
assert result.status == "success"
assert result.input_chars > 0
assert result.output_chars > 0
async def test_structured_failure_not_retryable(provider, llm_service, make_context, settings):
"""Persistent structured-output failure is not orchestrator-retryable: the
internal repairs already consumed the retry budget."""
provider.set_responses(["bad json", "still bad json"])
agent = RequirementsAgent(llm_service)
result = await agent.run(make_context("Food delivery."))
assert result.status == "failed"
assert result.retryable is False
async def test_devops_agent_receives_stack_without_api(provider, llm_service, make_context):
"""DevOps builds from the architecture and database only. The API design is
withheld on purpose — it does not change a Dockerfile, a compose file or a
CI workflow, and withholding it lets DevOps run alongside the API agent."""
agent = DevOpsAgent(llm_service)
context = make_context("Food delivery.")
context.requirements = requirements_output()
context.architecture = architecture_output()
context.database = database_output()
context.api = api_output()
provider.set_responses([json.dumps(devops_output())])
await agent.run(context)
user_prompt = provider.calls[0][1]
assert "DATABASE DESIGN" in user_prompt
assert "PostgreSQL" in user_prompt
assert "API DESIGN" not in user_prompt
async def test_architecture_agent_valid_output(provider, llm_service, make_context):
provider.set_responses([json.dumps(architecture_output())])
agent = ArchitectureAgent(llm_service)
context = make_context("Food delivery.")
context.requirements = requirements_output()
result = await agent.run(context)
assert result.status == "success"
assert result.output["mermaid_diagram"]
async def test_all_agents_have_unique_names(llm_service):
agents = [
APIAgent(llm_service),
ArchitectureAgent(llm_service),
DatabaseAgent(llm_service),
DevOpsAgent(llm_service),
RequirementsAgent(llm_service),
]
names = {agent.name for agent in agents}
assert len(names) == len(agents)
async def test_agents_prefer_summaries_over_raw_artifacts(provider, llm_service, make_context):
"""When the orchestrator has summarized upstream artifacts, downstream
agents consume those summaries instead of the raw serialized artifacts."""
agent = APIAgent(llm_service)
context = make_context("Food delivery.")
context.requirements = requirements_output()
context.architecture = architecture_output()
context.database = database_output()
context.requirements_summary = "REQ-SUMMARY"
context.architecture_summary = "ARCH-SUMMARY"
context.database_summary = "DB-SUMMARY"
provider.set_responses([json.dumps(api_output())])
await agent.run(context)
user_prompt = provider.calls[0][1]
assert "REQ-SUMMARY" in user_prompt
assert "ARCH-SUMMARY" in user_prompt
assert "DB-SUMMARY" in user_prompt
# The raw architecture digest (full component list) is not forwarded.
assert "FastAPI" not in user_prompt
async def test_agents_fallback_to_digest_without_summary(provider, llm_service, make_context):
"""Without a precomputed summary the agent still gets the deterministic
compact digest of the upstream artifact."""
agent = APIAgent(llm_service)
context = make_context("Food delivery.")
context.requirements = requirements_output()
context.architecture = architecture_output()
context.database = database_output()
provider.set_responses([json.dumps(api_output())])
await agent.run(context)
user_prompt = provider.calls[0][1]
assert "orders" in user_prompt
assert "sql_schema" not in user_prompt
async def test_api_agent_revision_preserves_existing_and_issues(
provider, llm_service, make_context
):
"""A targeted revision shows the existing artifact + reviewer issues and
forbids a from-scratch regeneration."""
agent = APIAgent(llm_service)
context = make_context("Food delivery.")
context.requirements = requirements_output()
context.architecture = architecture_output()
context.database = database_output()
existing = api_output()
revision = RevisionInstruction(
artifact="api",
existing=existing,
issues=[
{
"artifact": "api",
"severity": "blocking",
"problem": "Endpoint X conflicts with database schema.",
"expected": "users.id",
"actual": "user_id",
"fix": "Align the field name.",
}
],
)
provider.set_responses([json.dumps(api_output())])
await agent.run(context, revision=revision)
user_prompt = provider.calls[0][1]
assert "REVISION TASK — API" in user_prompt
assert "preserve everything valid" in user_prompt
assert "Endpoint X conflicts with database schema." in user_prompt
assert "Do NOT regenerate the artifact from scratch" in user_prompt
assert "Do NOT introduce new inconsistencies" in user_prompt |