PR Custody

feat(curl-tester): track field-level dependencies (minimal V1)

@AgentWrapperchecks n/achecks…feat/curl-tester-dependency-graph → next3 files · +64 −2updated 2mo ago

▸Description

What

Adds minimal field-level dependency tracking to the curl tester. When the agent calls parent endpoints to get data for the current endpoint, it records which fields it extracted and where it used them.

Changes

cortex/agents/curl_tester/state.py

Added dependencies: List[dict] = Field(default_factory=list) to CurlTesterResponse

cortex/agents/curl_tester/prompt.py

Added DEPENDENCY TRACKING section instructing the agent to emit a dependencies array
Updated response format examples to include the field

cortex/tests/test_common/test_curl_tester_state.py

Backward compat test (empty list default)
Plain dict acceptance test
JSON roundtrip test

What this does NOT do (intentionally)

No new Pydantic models — plain dicts are sufficient for V1
No new files or infrastructure
No S3 master graph merge logic
No aggregation on CurlTesterResult
No changes to create_app.py

Dependency data flows into existing CurlTesterResult.responses and gets logged to Datadog + S3 event traces like everything else already does.

Stripped from original PR

Removed ~870 lines of overbuilt infrastructure:

cortex/common/dependency_graph.py (275 lines of S3 merge logic)
cortex/tests/test_common/test_dependency_graph.py (564 lines)
FieldDependency Pydantic model
dependency_graph field on CurlTesterResult
S3 upload in create_app.py

loading diff…