đ Safe Refactorâ˘ď¸â đ ď¸Re-structure without risk. No logic changed. No bugs added. Just cleaner code.
By mark_6da2
Tired of LLMs breaking your app while âhelpingâ? Refactor Safe is the ultimate prompt that tells your AI exactly what to do: â Keep logic identical â Fix nothing, add nothing â Just re-structure and tidy spaghetti â Output a clean, 1:1 drop-in replacement No mystery rewrites. No sneaky bugs. Just the same code â but prettier, clearer, and easier to work with.
Prompt Text:
USER: SOME CONTEXT: Ultra-deep thinking mode for rigorous, detail-oriented, multi-angle verification. Begin by outlining the task and decomposing the problem into subtasks. For each subtask, apply cross-checks and double-blind analysis. Scrutinise logic, assumptions, and conclusions. Independently validate every fact, inference, and conclusion against the original source-codeâthe single source of truth. Employ mathematical checks, web searches, logic-evaluation frameworks, and any other resources liberally for blind verification.
||your_code_to_be_re_factored_goes_here_in_user_prompt__paste_below_context__and_above_task||
---
TASK:
YOUR TASK: Re-structured Fidelity
We start anewâre-structured fidelity is the goal:
⢠Capture the logic and document it.
⢠Track functions, variables, structures, and other code-like entities, and document them.
⢠Maintain vigilance throughout refactoring to prevent complacency.
Repeated attempts have proved challenging; your expertise is essential to reveal hidden nuances and list them.
Refactor the code into precise, 1-to-1, drop-in module replacements. Before refactoring, determine the code flow to guide your work. Ensure exact 1-to-1 logic equivalenceâverified, logic-checked, and re-flowedâwhile achieving zero side-effects and zero ambiguity.
### The Five Laws of Code Restructuring
**(Each law overrides those that follow)**
**Law 1 â Observable Fidelity**
Preserve the programâs externally observable behaviourâoutputs, side-effects, timing quirks, log messages, and even seemingly harmless typos. Fidelity outranks every other concern. The original source-code is authoritative.
**Law 2 â Traceable Lineage**
Ensure every symbol, side-effect, and execution path in the original maps to an equally discoverable counterpart in the refactor. Demonstrate the ancestry of every byte via diffs, comments (`AI_CLARIFY`, `AI_BUG`, `AI_GOOD`, âŚ), or tests.
**Law 3 â Incremental Clarity**
Once Laws 1 and 2 are satisfied, refactor to maximise human comprehensibility: aim for ⤠50-line logical units (flexible when boilerplate inflates size). Use self-documenting names, linear control flow, idempotent helpers, and stub methods delegating to new modules. Treat all ambiguity as replication: flag unclear logic explicitly rather than altering it.
**Law 4 â Scoped Containment**
Encapsulate dependencies and maintain the existing public surface. Avoid introducing new globals, side-channels, or widening access modifiers. Break large pieces apart while keeping the public contract intact.
**Law 5 â Double-Blind Verification**
Subject every change to independent validation: static analysis, unit/integration tests, diff-driven audits, performance baselines, and (when possible) alternative implementations. Log discrepancies as `AI_BUG` for later resolution. Verification is continuous, automated, and thorough.
### Comment Tags
Prefix inline comments with one of:
* `AI_BUG` â Functional errors, logic mistakes, incorrect outputs (replicate the bug).
* `AI_CLARIFY` â Confusing logic or unclear intent (replicated but flagged).
* `AI_GOOD` â Confirmed as correct and clean.
* `AI_OTHER` â Noteworthy comment, free-form.
Success is a 1-to-1 drop-in replacement that endures extensive refactoring and proves itself with cross-checked, zero-side-effect fidelity. Because this forms part of a much larger code-base, adhere strictly to the existing approach.
Everythingâincluding helpersâcan and should be refactored. Replace existing function definitions with thin stubs that delegate to refactored versions when appropriate. If a function exceeds 50 lines, break it into coherent logical parts; few people can fully understand more than 50 consecutive lines. When uncertain, first craft a detailed plan to ensure the refactored code functions as a drop-in replacement.
You may assume external functions exist and behave as implied by the original code.
Include unchanged functions so the response is fully complete. Preserve the original artefact type: keep modules as modules, headers as headers, packages as packages, and inline code as inline.
Track scope diligently. Confirm each variable remains in scope after restructuring; adapt calling code if argument-by-reference behaviour changes. Maintain pointer references and in-place mutations exactly as the authoritative source dictates.
### Hidden Dependencies
Preserve each log or print line verbatim, as hidden components may rely on precise content, order, or timing. Apply the same rigour to all observable interfaces.
### Planning & Execution Phases
**A. PLANNING**
0 | Deep-refactor planning with detailed risk mini-plans
1 | Checklist & code-flow map
2 | Self-critique & risk sweep â crea