> Split complex tasks into simpler subtasks

> Prompt the model to explain before answering

> if you give GPT-3 a task that's too complex to do in the time it takes to calculate its next token, it may confabulate an incorrect guess

> Tactic: Specify the steps required to complete a task

> Some tasks are best specified as a sequence of steps. Writing the steps out explicitly can make it easier for the model to follow them.

> Split complex tasks into simpler subtasks

> Complex tasks tend to have higher error rates than simpler tasks. Furthermore, complex tasks can often be re-defined as a workflow of simpler tasks in which the outputs of earlier tasks are used to construct the inputs to later tasks.

> Providing general instructions that apply to all examples is generally more efficient than demonstrating all permutations of a task by example

> Chain prompts: Divide complex tasks into smaller, manageable steps for better results

> Let Claude think: Encourage step-by-step thinking to improve the quality of Claude's output

> each query will contain only those instructions that are required to perform the next stage of a task which can result in lower error rates compared to using a single query to perform the whole task

> Claude's reasoning frequently exceeds what a human would prescribe. (Prompting best practices)

> GPT models like gpt-6-astra benefit from precise instructions that explicitly provide the logic and data required to complete the task in the prompt.

> Generally speaking, reasoning models will provide better results on tasks with only high-level guidance.

> hardcoding complex, brittle logic in their prompts to elicit exact agentic behavior

> vague, high-level guidance that fails to give the LLM concrete signals for desired outputs or falsely assumes shared context

> Find the smallest set of high-signal tokens that maximize the likelihood of your desired outcome.

> giving the AARs too much structure hurt their progress badly. When we prescribed a specific workflow ("propose ideas, then generate a plan, then write the code…"), we found we'd ultimately constrained Claude's work.

> giving each AAR a different starting point helped a lot, even if that starting point was vague

> Skills developed for prior models are often too prescriptive for Claude Fable 5 and can degrade output quality. Review and consider removing older instructions if default performance is better.

> Instruction-following is improved enough that you can steer most behaviors with a brief instruction rather than enumerating each behavior by name.

> If your prompt contains explicit verification instructions (...), remove them: instructions like these cause over-verification on Claude Opus 5, and removing them reduces wasted tokens with no loss in quality. The same applies to legacy harness scaffolding that adds separate verification steps.

> Prefer general instructions over prescriptive steps. A prompt like "think thoroughly" often produces better reasoning than a hand-written step-by-step plan.

> Describe the destination rather than prescribing every step.

> GPT-5-class models follow prompt contracts closely, so conflicting rules can create more instability than missing detail.

> Removing repeated instructions and examples and simplifying tool descriptions can improve task performance and token efficiency.

> simply specifying all requirements does not consistently help, as models have limited instruction-following ability and requirements can conflict.

> Frontier thinking LLMs can follow ~150-200 instructions with reasonable consistency.

> As instruction count increases, instruction-following quality decreases uniformly.