Engineered Prompt

The output of a prompt-engineer run: the confirmed REFINED_PROMPT.md rewritten as a prompt ready to execute, written next to its input. Every section below is required. scripts/check_engineered.py <refined> <engineered> must exit 0 on this document before it is returned.

Engineered prompt

The prompt itself, organised in the contract blocks below, in this order. Keep only the blocks the input gives content for: a block the user declared not applicable is omitted, never filled with a plausible default (R3, R18). Every sentence comes from the input; no locked decision may be contradicted. No reasoning-scripting phrase and no request to reproduce internal reasoning — see the anti-pattern checklist in references/PROMPTING_RULES.md (R9, R10); the checker rejects them here.

OBJECTIVE

The outcome, stated before any method (R1).

CONTEXT

The environment and the sources, with their priority when they conflict (R4, R15).

SCOPE

• In scope: …
• Out of scope: …
• Do not change: …

TOOLS

When to use which tool, when not to (R21).

ACTION BOUNDARIES

What proceeds autonomously, what needs approval first (R23).

VERIFICATION

Observable checks, proportionate to the risk, reported as evidence (R25, R27).

OUTPUT

What the final deliverable contains (R29).

STOP CONDITION

The observable condition that ends the work (R26).

Change log

One entry per change made to the input text, each citing at least one rule of references/PROMPTING_RULES.md by number.

• (R)

Lock register

The input's lock register, copied byte-identical — superseded entries included, still struck-through.

• L1 (round ): "<user's answer, verbatim>"

Open questions

Only residual tensions inside covered blocks — e.g. two locks that pull in different directions, both kept verbatim. Never a block left uncovered: that input is rejected and goes back to prompt-loop. n/a when there is none.

• n/a

Recommended effort

high