BC-0059 · CORE GUIDE
Editorial review 2026-09-26
How to Use Build Prompt History
Use prompt history to recover the wording and reasoning behind a change, not as proof that an earlier game state can be restored exactly. Match the relevant request to your observation notes before reusing or correcting it.
Objective
Use retained prompt wording to explain decisions without inventing version-recovery behavior.
Before you start
- Identify the behavior or change you need to trace.
- Use only history or conversation controls actually visible to you.
- Keep an independent place for observations and missing-record notes.
Steps
- Locate the request associated with the behavior in question.
- Read subsequent constraints that may have changed its meaning.
- Separate instructions, generated explanations and observed results.
- Copy the relevant wording without silently rewriting it.
- Compare its assumptions with the current game before reuse.
- Write the clarified next instruction and its new acceptance check.
Verification
- The record distinguishes exact retrieved text from recollection.
- Later follow-ups are considered before interpreting an earlier instruction.
- A generated explanation is not recorded as a performed test.
- Reused prompts receive their own observation record.
- Retained history is not described as an exact project backup.
Limitations
- This guide does not document a universal history-export or restore control.
- The available interface can differ from the wording retained in documentation.
- A new generation based on old text may not reproduce the former result.
Working notes
Start with the behavior you are trying to explain. Find the request that introduced or changed it, then read nearby follow-ups for constraints that may have been added later. The initial creation prompt alone may not describe the game you currently see.
Separate three kinds of record: your instruction, a generated response and your own observation of the game. History may help identify the request, but an explanation in the conversation is not proof that the resulting mechanic behaved as described. Add your actual test sequence and outcome to a separate note.
Use the history controls actually present in your workflow. Do not rely on an invented menu path, export action or restore button. If you cannot retrieve the older wording, mark that part of the record unavailable rather than reconstructing it from memory and presenting it as exact.
Treat reuse as a new decision. A past prompt can contain assumptions that later revisions changed. Before sending it again, compare its intended starting state and preservation rules with the current project. Keep the reused wording separate from the earlier result so the history remains interpretable.
Examples
- History reading example: an early message asks retry to return to the entrance; a later instruction asks to preserve checkpoint progress. Compare their intended scopes before calling the current respawn behavior a defect. The useful output is a clarified reset rule, not merely the oldest message.
- Record pairing: request text—copy exactly when available; generated explanation—label separately; inspected sequence—checkpoint then retry; observed outcome—fill after checking. An absent observation remains absent even when the response sounds confident.
- Recovery boundary: keeping the prompt that described a harbor route preserves design information. It does not by itself preserve every generated object, behavior or saved state from that route. Record any genuine recoverable project copy separately.