Prompting

Write and check a plan before running three or more dependent or irreversible steps

Reason through multi-step problems and tradeoffs step by step before answering

Pick the most direct tool for each step and avoid redundant or speculative tool calls

State your confidence level when facts are uncertain, outdated or contested

Pause and confirm before deleting, overwriting, sending, pushing or changing production

Ready-made AWS prompts for scaffolding MVPs, RAG chatbots and security reviews

Restate the goal and problem behind a request to confirm understanding

Set a concise, verified working style with simple code and deliberate subagents

Sharpen a vague task into a clear spec before starting the work

Score each decision on a 1 to 10 thoroughness scale and pick the higher score

Hedge overconfident claims like always or 100% in outgoing messages

Turn a weak prompt into a reliable one that works across different models

Clarify vague requests and write a clear brief before doing the work

Reason in revisable, branching steps for multi-stage analysis and design planning

Auto-classify and save your prompts by type for technical, content or teaching use

Toggle deep thinking mode for complex work and turn it off for quick replies

Switch the assistant's communication style between eight persona presets

Write prompts for multi-stage pipelines where one model output feeds the next

Evaluate and improve LLM prompts with examples and measured comparisons

Craft prompts and system messages with chain-of-thought and constitutional methods

Apply prompt patterns to improve LLM reliability, output quality and control

Write and optimize prompts for commands, skills, sub-agents and other LLM calls

Review a previous answer and refine it with a self-critique and verification pass

Test prompts and skills against desired behavior using red-green-refactor cycles

Apply chain-of-thought prompting techniques to multi-step reasoning problems

Explore, prune and expand solution branches with a tree-of-thoughts search

Rewrite a fuzzy problem into a brief an agent can reason about and verify

Set a good-enough threshold first and stop at the first option that clears it

Check whether a claim is grounded, then answer, fetch evidence or abstain

Classify a messy problem by cause and effect before choosing how to respond

Act under uncertainty by capping downside and letting goals emerge from your means

Separate physics from convention and rebuild the simplest solution that fits

Judge whether a proven option will last, checking for domain drift first

Verify live code or data when docs and tests conflict with observed behavior

Size buffers to residual error and breach cost, not the optimistic estimate

Sequence complementary thinking models with named roles for high-stakes problems

Pick one matching thinking model for a problem, or none, before you start

Act fast under time pressure with a loop of observe, orient, decide and act

Name the best forgone alternative before committing scarce time, people or money

Assume a plan failed and turn the likely failure causes into mitigations and gates

Estimate with base rates, ranges and Bayesian updates on new evidence

Sort decisions into cheap or costly to undo, deciding two-way doors fast

Trace a change's downstream effects, incentives and timing before committing

Ask the few load-bearing questions that expose hidden requirements before building

Build the strongest opposing case before accepting or rejecting a proposal

Map feedback loops, stocks and flows to find leverage in emergent system behavior

Run a controlled counterfactual when a real test is too rare or irreversible

Name a design contradiction and invent a resolution that gives up nothing

Remove harmful or nonessential parts before adding features, layers or process