Prompt Library

Prompts That Change Behavior

Not "write me a poem." These are structural prompts — scaffolds that force models to plan, clarify, and self-correct.

The Clarifying Interrogator

Forces the model to surface every ambiguity before acting.

You are a senior engineer. Before writing any code or answer, list every assumption you are making and ask me any questions whose answers would change your output. Wait for my reply before proceeding.

Plan-Execute-Self-Correct

Separates planning from execution and adds a review step.

Do this in three phases. Phase 1: give me a numbered plan with the goals and risks of each step. Phase 2: execute only after I approve the plan. Phase 3: review your own work against the original goal and report anything you could not verify.

Explain-Then-Implement

Forces reasoning before code, reducing guesswork.

Before writing code: (1) restate the task in your own words, (2) name the data structures and edge cases, (3) only then write the implementation with brief inline comments at each non-obvious line.

The Adversarial Reviewer

Stress-tests a design or piece of code from a skeptic's view.

Act as a hostile reviewer. Find the top 5 ways this design could fail in production — correctness, performance, security, and operations. For each, state the failure mode and the concrete fix. Then rate the overall risk from 1-10.

Threat-Model a Feature

Turns a spec into a concrete threat model.

Given this feature description, produce a threat model: list the trust boundaries, enumerate at least 6 plausible attacks, rank them by likelihood and impact, and recommend the three highest-value mitigations with their cost.

Build on these

Combine prompts with the right tools and a clear workflow.