# Prompts Library

Practitioner prompts that operationalize the disciplines in Loop Engineering. Each prompt is dated and version-pinned against the AI surface it was verified against. Copy-paste-ready; replace bracketed placeholders before use.

The library is organized by architectural layer so practitioners can start with the failure mode or task at hand. Each layer links to the chapter that grounds it.

The disciplines behind the prompts

### Read where each prompt comes from

As an Amazon Associate I earn from qualifying purchases.

                [Buy on Amazon](https://amzn.to/4fyLgVH)

### Prompts by layer

### Working Memory (Ch 6)

                        NOTES.md template
                        Scaffolding contract
                        Verifiable-stop checklist

The externalized-scratchpad pattern, the scaffolding-loop contract, the verifiable-stop checklist, and the compaction-survivor prompt that lets the loop resume cleanly across context-cliff events.

                    [Open the templates →](/prompts/working-memory/)

### Dialog Interface (Ch 8)

                        Intent clarification
                        Assumption surfacing
                        Dialog repair

Intent-clarification openers, assumption-surfacing prompts, dialog-repair templates, and faithfulness-signal phrasings calibrated to four confidence levels.

                    [Open the templates →](/prompts/dialog/)

### Failure Diagnosis (Ch 12)

                        Region-by-region probes

Diagnostic probes for naming which architectural region failed when something misbehaves. Walking the regions in order beats guessing at a fix.

                    [Open the probes →](/prompts/failure-diagnosis/)

### Materiality (Ch 13)

                        Stakes assessment
                        Autonomy posture
                        Drift re-check

Stakes-assessment prompts on three axes (stakes / reversibility / observability), autonomy-posture selectors, calibration tables matching discipline to materiality, and mid-task drift re-checks.

                    [Open the templates →](/prompts/materiality/)

### How to use this library

- Start at the layer that matches your current pain point. If a partnership is producing wrong answers, the failure-diagnosis page comes first; if a long task is losing state, the working-memory page comes first.

- Each prompt names a single failure mode it prevents. If you do not have that failure mode, do not bring the prompt - over-rigging is its own failure mode.

- Each prompt is verified against specific AI surfaces (named on the page). If you are using a different surface, treat the prompt as a starting point rather than a guarantee.

- Templates are copy-paste-ready, but the discipline they encode is not. The prompt is a reminder; the practice is the actual thing.

Each prompt carries a "verified MM-YYYY" date so freshness is visible. The library is reviewed quarterly.