---
title: "Claude prompts I keep reusing"
url: https://memory.wiki/wzixoDcN
updated: 2026-05-15T03:37:20.915Z
hub: https://memory.wiki/hub/demo
concept_count: 11
source: "memory.wiki"
---
# Claude prompts I keep reusing

> Personal pack. Not authoritative, just what's working for me as of 2026-05.

## Re-read mode

> "Read this URL as context. Summarise the key claims in 3 bullets. Then ask me one question that exposes the weakest claim in the summary."

Why it works: the summary forces compression; the one question forces critical engagement. The combination prevents the "Claude says yes to everything" failure mode.

## Decision mode

> "You're an opinionated engineering manager. The context is at <URL>. I'm trying to decide whether to <decision>. Recommend an action, then explain the single biggest tradeoff. No hedging."

Why it works: "opinionated EM" anchors the voice. "Single biggest tradeoff" forces ranking. "No hedging" closes the escape hatch.

## Audit mode

> "List every assumption in this doc that isn't backed by evidence elsewhere in the hub. For each, suggest what evidence would make the assumption checkable."

Why it works: the second clause keeps it from being a list of nitpicks. Each item ends with a concrete way to validate.

## Reading list mode

> "From <bundle URL>, what's the ideal reading order if I have only 15 minutes? Order them by 'most value gained from reading first'."

Why it works: bundles have reading orders built in (the analyser produces one) but they optimise for "complete understanding," not "best 15-min." Asking for the time-boxed order is different.

## Counterpart mode

> "You're the smartest person who disagrees with this thesis: <thesis>. The context is at <URL>. Write the strongest version of their argument."

Why it works: forcing the opposite framing reveals which parts of my thesis are durable and which parts only hold because I haven't seen the counter-argument.

## Pre-write mode

> "I'm about to write a doc with the working title <title>. The audience is <person>. The context is at <URL>. Before I write, what are three questions you'd want me to answer in the doc, in priority order?"

Why it works: writing-question-driven docs land harder than writing-topic-driven docs. The pre-write prompt produces the questions.

## What I'm not using

- "Be concise." Useless — Claude is already calibrated for length to context. Better to say "Answer in <X> sentences" with a specific number.
- "Be honest." Implies otherwise. The model is mostly honest; ask better questions instead.
- Roleplay prompts that anchor to a famous person. They make the answer feel persona-shaped instead of evidence-shaped.


---

## Concepts in this document
- **Claude** _(entity)_
  Anthropic's AI model cited as an example of vendor lock-in through projects and memory features.
- **Prompt engineering** _(concept)_
  The prior dominant skill for AI tool interaction that is being displaced by context engineering as tools mature.
- **Re-read mode** _(concept)_
  A prompt pattern that combines summarization with critical questioning to force engagement and avoid uncritical acceptance.
- **Decision mode** _(concept)_
  A prompt pattern that anchors Claude to an opinionated voice and forces explicit tradeoff ranking to avoid hedging in decision contexts.
- **Pre-write mode** _(concept)_
  A prompt pattern that generates prioritized questions before writing to ensure docs are question-driven rather than topic-driven.
- **Counterpart mode** _(concept)_
  A prompt pattern that surface weaknesses in a thesis by forcing articulation of the strongest opposing argument.
- **Audit mode** _(concept)_
  A prompt pattern that identifies unsupported assumptions and pairs each with a concrete validation path to prevent mere nitpicking.
- **Reading list mode** _(concept)_
  A prompt pattern that optimizes reading order for time-constrained consumption rather than complete understanding.
- **Constraint specification** _(concept)_
  The practice of adding concrete bounds like sentence count or 'no hedging' to close escape hatches and force prioritization.
- **Anti-pattern: generic softeners** _(concept)_
  Vague instructions like 'be concise' or 'be honest' that are ineffective and should be replaced with specific, measurable constraints.
- **Voice anchoring** _(concept)_
  Technique of establishing a consistent perspective (e.g., 'opinionated EM') to shape response tone and approach.

## Concept relations (within this doc's concepts)
- **Re-read mode** is a pattern of **Prompt engineering**
- **Reading list mode** is a pattern of **Prompt engineering**
- **Voice anchoring** enables **Decision mode**
- **Decision mode** is a pattern of **Prompt engineering**
- **Counterpart mode** is a pattern of **Prompt engineering**
- **Constraint specification** strengthens **Decision mode**
- **Prompt engineering** patterns designed for **Claude**
- **Audit mode** is a pattern of **Prompt engineering**
- **Pre-write mode** is a pattern of **Prompt engineering**
- **Anti-pattern: generic softeners** opposes **Constraint specification**

_Hub canonical:_ https://memory.wiki/hub/demo
_Concept digest:_ https://memory.wiki/raw/hub/demo?digest=1&compact=1