Loading page…
Loading page…
Proposed workflow: Investigate a coding assumption, propose a correction, and verify the affected behavior.
Project concept · Proposed Darwin workflow.
Align is a Python tool that hooks into Claude Code to examine coding assumptions. It supports rule-based detection, the Anthropic SDK, or a headless Claude Code detector. The Darwin workflow below is a proposed extension.
A mistaken assumption can send coding work in the wrong direction. Align flags assumptions for investigation before they become the basis for more changes.
A Darwin integration could find code-analysis expertise for the flagged assumption and the behavior the developer expects.
A specialist could investigate the evidence, propose a focused correction, and request tests of the affected behavior.
An investigation with supporting evidence, a proposed change where warranted, and verification results for the developer’s merge decision.
Search. Find code-analysis expertise for the flagged assumption and expected behavior.
Act. Investigate the selected code and identify evidence that supports or challenges the assumption.
Outcome. An investigation with supporting evidence, contrary evidence, and affected code paths.
Search. Check that the analysis supports a correction before proposing code changes.
Act. Request a focused change that addresses the finding and explains its effect.
Outcome. A correction proposal with behavior changes, tradeoffs, and unresolved questions.
Search. Find testing help for the affected code and its intended behavior.
Act. Run agreed checks or return a test plan with unrun checks identified.
Outcome. Verification evidence and remaining tests for the developer’s merge decision.
Read the evidence, proposed correction, and test results together before changing the code.
Deliverable: An investigation with supporting evidence, contrary evidence, and affected code paths.
Review criteria: Stay within the selected context. Distinguish evidence that the assumption is wrong from questions still to investigate.
Deliverable: A correction proposal with behavior changes, tradeoffs, and unresolved questions.
Review criteria: Return the change for review. Do not alter unrelated code or merge it automatically.
Deliverable: Verification evidence and remaining tests for the developer’s merge decision.
Review criteria: Show actual test evidence separately from the plan. Keep the merge decision with me.
undercurrent
Kestrel
super-travel
Python