A long system prompt is most useful when you turn its recurring ideas into habits you can reuse. The strongest patterns in this Claude Opus 5.5 system prompt are simple: give Claude clear context boundaries, concrete deliverables, and a verification step. Let it inspect source material, use tools when the task needs them, and finish the work instead of stopping at advice.
These ten templates keep those patterns while staying short enough for everyday use. Replace the bracketed text, attach any relevant files or links, and paste the whole block into Claude.
1. Coding
<task>Fix [bug or feature] in [project].</task>
<context>Expected: [behavior]. Current: [behavior]. Constraints: [constraints].</context>
Inspect the relevant code before editing. Make the smallest sound change, preserve established patterns, and run the relevant checks. Do not stop at a plan.
<deliverable>Return the working change, test results, and any remaining risk.</deliverable>
2. Marketing
<task>Create [campaign asset] for [product] aimed at [audience].</task>
<context>Goal: [action]. Main benefit: [benefit]. Voice: [voice]. Evidence: [facts or links].</context>
Develop three meaningfully different angles, choose the strongest, and write the final asset. Use only claims supported by the context.
<deliverable>Give the winning angle, finished copy, and one sentence explaining why it should work.</deliverable>
3. Sales
<task>Write a [cold email or follow-up] to [buyer role] at [company type].</task>
<context>Likely problem: [problem]. Relevant proof: [specific fact]. Offer: [offer].</context>
Keep it under [word count], make the value concrete, and end with an easy-to-answer question. Never invent personalization.
<deliverable>Return a subject line and send-ready message.</deliverable>
4. Research
<task>Research [question] to support [decision].</task>
<context>Options: [options]. Constraints: [constraints]. Freshness required: [date range].</context>
Use current primary sources where possible. Separate verified facts, reasonable inference, and unknowns. Compare the options against consistent criteria.
<deliverable>Lead with a recommendation, then give evidence, trade-offs, uncertainties, and linked sources.</deliverable>
5. Content
<task>Write a [post, article, or script] for [audience] about [topic].</task>
<context>Core point: [takeaway]. Voice sample: <voice>[paste sample]</voice>. Length: [length].</context>
Open with a concrete hook, use specific examples, match the voice without copying phrases, and remove generic filler.
<deliverable>Return the polished final piece plus three alternate headlines.</deliverable>
6. Product
<task>Turn the material below into a decision-ready product brief.</task>
<context>[Paste customer notes, research, constraints, and existing behavior.]</context>
Identify the user, problem, desired outcome, proposed scope, exclusions, dependencies, and success measure. Label assumptions instead of presenting them as customer facts.
<deliverable>Return the brief, the three biggest open questions, and the smallest useful first release.</deliverable>
7. Support
<task>Draft a reply to this customer: <message>[paste message]</message></task>
<context>Policy and product facts: <facts>[paste facts]</facts></context>
Acknowledge the issue without over-apologizing, explain the next concrete step, and state plainly what cannot yet be confirmed. Do not promise actions outside the supplied policy.
<deliverable>Return one calm, human, send-ready reply.</deliverable>
8. Operations
<task>Create a repeatable process for [task].</task>
<context>Trigger: [trigger]. Owner: [owner]. Tools: [tools]. Constraints: [constraints].</context>
Specify the ordered steps, decision points, handoffs, exception handling, and evidence of completion. Make it usable by a new teammate without extra explanation.
<deliverable>Return the SOP and a short completion checklist.</deliverable>
9. Analysis
<task>Analyze <data>[paste data or attach file]</data> to answer [question].</task>
<context>Decision this informs: [decision]. Important definitions: [definitions].</context>
Check calculations, test alternative explanations, and call out misleading comparisons or missing data. Do not imply causation without evidence.
<deliverable>Lead with the conclusion, then show evidence, assumptions, limitations, and the next best analysis.</deliverable>
10. Planning
<task>Build an executable plan to achieve [goal] by [date].</task>
<context>Resources: [people, budget, tools]. Constraints: [constraints]. Current state: [state].</context>
Break the work into milestones with owners, deliverables, dependencies, and verification. Make reasonable assumptions and label them.
<deliverable>Return the milestone plan, first three actions, biggest risk, and a practical fallback.</deliverable>
Why these work well with Claude
XML tags keep instructions, context, and source material from blending together. A defined deliverable reduces vague answers. The final verification instruction pushes Claude to check its own work, while explicit permission to inspect files, use tools, or consult sources helps it move beyond generic advice.
The key is not making every prompt longer. Give Claude the information it cannot infer, define what finished means, and let it do the work.