Skip to tool
Automaton Toolsby Dead Reckoning Foundry

Opus 5.5 · Prompt Lab

Give it a clear finish line.

Set the outcome, add the evidence, and let Opus work.

Tips & sources ↓

Your brief

Name a concrete result. Keep process instructions to what matters.

Context & working boundaries

0 / 16,000 characters. Clearly labeled as source material in the prompt.

For tasks with tools and multiple steps. Existing approval boundaries still apply.

Build and copy in your browser. Sign in to save prompts to your account.

Your prompt

A focused starting point

Outcome. Evidence. Boundaries.

Build a prompt from your brief, or start with an example. Only the guidance you need goes into the result.

Copy into your chosen Claude app when ready.

Saved prompts

Save a named, editable copy to your account and return to it later.

Your prompt library

Sign in to keep prompts across sessions. Sign in to Automaton ↗

Save to your library

Give this prompt a name. Refine the text before saving.

Field notes

Get more from Opus 5.5

Practical lessons from your eight-source notebook. The notes that link to Anthropic were checked for consistency with Anthropic’s guide on November 11, 2026; they add practical detail from the notebook. The two that link to the notebook were not checked.

Start at medium

Choose medium in your model settings, then compare quality and time on real tasks. A prompt cannot set the API’s effort value.

Anthropic: effort calibration ↗

Describe the result

Give it a concrete finish line and a way to check it. Add a useful example when style or format matters; avoid a long script for every step.

Notebook: prompting lessons ↗

Ask for evidence

Request conclusions, assumptions, sources, and checks. You do not need a transcript of private reasoning to assess an answer.

Anthropic: thinking guidance ↗

Keep long work accountable

Use a checklist with explicit completion evidence. In unattended runs, inspect open items after a progress-only turn and cap automatic retries.

Anthropic: unattended runs ↗

Wait for what is still running

A turn that ends with only text is a report, not completion. If a background command or subagent is still running, wait for its output before treating the task as done.

Anthropic: unattended runs ↗

Give design direction

Name your existing theme, audience, reference screens, and specific patterns to avoid. Review the first screen before expanding the rest.

Anthropic: frontend direction ↗

Bring the right context

Point to relevant files, records, and prior decisions. Label copied material separately from your own instructions. Clear labels help; they are not access controls.

Notebook: context and workflow ↗
Sources, limits & when to use this

The notebook contains eight creator-video entries and seven unique transcripts; the two engineers-at-work entries are duplicates. Their demos are examples, not measured guarantees for your tasks. This builder adapts the useful patterns into original templates; it does not run or evaluate Opus.

For dense visual work, provide the original image and appropriate inspection tools. For API integrations, check current thinking and progress-display settings in the official guide. Prompt text alone cannot enable tools, permissions, or runtime settings.