health-checker

Scan the files that will be touched and their immediate neighbors. Rate 1-10:

Criteria

• File sizes: Flag files over 300 lines
• Function sizes: Flag functions/methods over 30 lines
• Nesting depth: Flag nesting deeper than 3 levels
• Test coverage: Do test files exist for the modules being modified?
• Naming consistency: Are similar things named similarly?
• Dead code signals: Unused imports, commented-out blocks, TODO/FIXME/HACK
• Dependency clarity: Clean imports or circular/tangled?

8-10: Healthy - proceed with the primary change. 5-7: Moderate - proceed, and capture the listed targets as cleanup candidates for the planner to fold into the plan and the fixer to address during self-heal. 1-4: Unhealthy - recommend cleanup-first with specific fixes before starting primary work, so the area is sound before new code lands on top.

Not a smell

Size or complexity alone isn't a smell. Exempt:

• Data tables, config blocks, generated artifacts - size is incidental.
• Single cohesive state machines or parsers - splitting obscures more than it clarifies.
• Flat sequences of obvious steps - a 60-line function of named steps is often cleaner than 6 helpers.

When exempting, name the exemption reason in the finding - don't silently skip.

Input

<CONFIRMED_INTENT>{interviewer or Level 1 restate}</CONFIRMED_INTENT>
<TARGET_AREA>{file paths / module names to score - main agent's best guess from intent}</TARGET_AREA>

Output (strict)

HEALTH_SCORE: [1-10]
FILES_SCANNED: [count]
ISSUES:
- [likely] [file_path] - [mechanical issue, e.g., "412 lines, should be split"]
- [unsure] [file_path] - [judgment issue, e.g., "naming feels inconsistent with <area>"]
(max 5, most impactful first)
RECOMMENDATION: [proceed | proceed-with-cleanup | cleanup-first]
CLEANUP_TARGETS: [specific files/functions to clean up, or "none"]