Copyable starting points
Turn a request
into a working result.
Give Claude a clear goal, a limited scope, and a way to prove the result.
Use the guided prompt builder →
The prompt pattern that travels well
Goal: [the result I want]
Context: [the project, relevant files, and current behavior]
Scope: [what may change and what must remain unchanged]
Acceptance: [observable behavior or specific tests]
Process: Explain the plan first. Implement after approval.
Report: Changed files, checks run, results, and remaining limits.Replace every bracketed placeholder before sending. Include the exact error when debugging. Attach a relevant screenshot when appearance matters, and identify whether it is the current state or your desired design.
Choose a task
Understand a repository
Read the README and the main project configuration. Explain what
this project does, how the folders fit together, and how to run it.
Define technical terms. Do not edit files or install anything.
Identify missing information instead of guessing.Fix a bug
Reproduce this problem: [steps and exact error].
Expected: [what should happen]. Actual: [what happens instead].
Find the cause, propose the smallest fix, and add a regression test
if appropriate. Do not refactor unrelated code. After approval,
implement and run the relevant checks. Show the results.Add a feature
Add [feature] to [screen or module]. Use existing project patterns.
Acceptance criteria:
1. [successful action and expected result]
2. [invalid input and expected response]
3. [keyboard and mobile behavior]
Inspect the current code and propose a plan before editing.
Do not add new services, paid tools, or dependencies without explaining why.Review a pull request
Review PR [number] against the current main branch. Read the
complete diff and existing review comments. Focus on correctness,
security, migration compatibility, and tests for the latest commit.
For each finding, explain the impact and cite the file and location.
Do not edit code, post comments, approve, merge, or resolve threads.Create a chart or research page
Analyze [data or question]. For factual claims, use current primary
sources and record their dates and links. Separate evidence from
inference. Explain missing or uncertain data.
Create a clear chart if the data supports it. Label units, axes,
and the time period. Include an accessible text explanation.
Do not invent values or imply that a correlation proves a cause.Improve accessibility
Review [page] for keyboard use, visible focus, labels, contrast,
and zoom/reflow. Inspect the running page if browser tools exist.
Separate source-code findings from actual interaction tests.
Propose focused fixes and say which checks still require a human.Document a home deployment
Inspect this project and propose a deployment for my home server.
Ask for the OS, architecture, existing services, network access needs,
and backup destination if unknown. Include dependencies, secrets,
health checks, persistent storage, updates, restore, and rollback.
Start with private access. Do not expose ports or deploy yet.Research, graphics, and files
Claude Code can write scripts to analyze files, produce charts, generate SVG diagrams, and build HTML reports. Research needs accessible web tools or connectors; file analysis needs the appropriate libraries. A prompt cannot grant a missing tool or network connection.
For screenshots and UI references, say what Claude should preserve and what should change. For raster image generation or specialized document editing, use an available integration and verify its output. Do not assume that every Claude chat feature is a built-in Claude Code tool.
Ask for evidence, then inspect it
- “Tests passed”: Which commands ran, and what did they cover?
- “Looks correct”: Was the page actually opened, or only the source inspected?
- “Deployed”: Is there a successful deployment result and working destination?
- “Secure”: Which threats were checked, and what remains outside the review?
Keep tasks small enough to inspect. Stop and correct the direction when the output misses the goal instead of adding a long chain of patch requests.
- Anthropic. (n.d.-c). Best practices for Claude Code. Retrieved September 24, 2026, from https://code.claude.com/docs/en/best-practices
- Anthropic. (n.d.-t). Extend Claude with skills. Retrieved September 24, 2026, from https://code.claude.com/docs/en/skills
- Anthropic. (n.d.-ae). Overview. Retrieved September 24, 2026, from https://code.claude.com/docs/en/overview
- Anthropic. (n.d.-aq). Use Claude Code with Chrome. Retrieved September 24, 2026, from https://code.claude.com/docs/en/chrome
Was this chapter useful?
Report an errorVotes are shared only if you enable optional measurement in Privacy & Preferences.