[FEATURE] Suggest targeted logging to collapse the hypothesis space

Open 💬 0 comments Opened Jul 14, 2026 by simonticciatihayes

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

When Claude is investigating a bug in which the uncertainty space has multiple high-level branches (e.g. "is this a Lit version issue, a CSS specificity issue, or a shadow DOM issue?"), he has a tendency to attempt to reason through all branches simultaneously.

This can escalate into repetitive "thinking loops" due to the fact that (from whatever angle the problem is approached), the possibility space resulting from those high-level uncertainties is unmanageably large.

In such cases, I've found that interrupting the "thinking loop" to proactively suggest that Claude either (a) augment the code with temporary diagnostic logging (which I can report back to him after rebuilding/re-running) and/or (b) outline tests that I can perform (e.g. in the case of web front-end web-component type problems, using the browser's developer tools) with the aim of eliminating at least _some_ of those high-level uncertainties works really well. While this is typically an iterative process (Claude adds diagnostics or suggests tests that I can perform, I rebuild/rerun and/or perform those tests & report the diagnostic logging / outcomes back to Claude), it did result in our homing in on the the answer more quickly (and more reliably) than Claude attempting to figure it out logically based solely on consideration of the (initially unfeasibly large) set of possibilities.

Proposed Solution

While users can, of course, create CLAUDE.md(s) that explicitly direct Claude to consider such a workflow as a possibility, I'm wondering whether this is something that could perhaps be "baked in" (as a candidate "bug investigation" strategy) - with Claude being triggered to propose strategy/workflow to the user in response to logical reasoning identifying that there are key intractable high-level uncertainties that might actually be amenable to being quickly/definitively resolved in this way.

(Claude is absolutely ace, btw - this is just a suggestion to make him even better :-)).

Alternative Solutions

_No response_

Priority

High - Significant impact on productivity

Feature Category

Interactive mode (TUI)

Use Case Example

_No response_

Additional Context

_No response_

View original on GitHub ↗