The Claude Code Prompt Pack

Try one right now

Prompt #19: SOP from a brain dump.

This is the highest-leverage prompt in the pack. Paste a messy description of any process you do regularly and Claude turns it into a formatted SOP your team can actually follow. Copy the whole block below, paste into Claude, replace the bracketed section with your notes.

You are an operations consultant who turns messy descriptions into formal SOPs.

<process_brain_dump>
[PASTE NOTES, VOICE TRANSCRIPT, OR DESCRIPTION]
</process_brain_dump>

Output a complete SOP in this structure:

1. PURPOSE — Why this process exists. One sentence.
2. SCOPE — When to use this and when NOT to. Two sentences.
3. ROLES — Who does what. Table: Role | Responsibility.
4. PROCEDURE — Numbered steps. Each step:
   - Action (verb-led: "Send X to Y")
   - Tool/system used
   - Output/handoff
5. DECISION POINTS — Any "if X then Y" branches.
6. SUCCESS CRITERIA — How we know each step worked.
7. REVISION HISTORY — Empty table: Date | Author | Change.

If anything is unclear, list questions at the bottom under OPEN QUESTIONS.

The other 23 prompts in the pack follow the same paste-ready structure: clear role, bracketed inputs, output format spec, and explicit constraints. No paraphrasing required.

The full menu

24 prompts. 4 departments. 6 each.

Sales

→ Cold email first lines from a LinkedIn paste (5 lines, all peer-level, no flattery)
→ Discovery call prep brief: top 3 pain points, 5 BANT questions, tailored pitch
→ CRM data extraction: messy call notes → structured JSON ready for HubSpot/Salesforce
→ Proposal first draft from a discovery transcript (2 pages, your pricing model, ROI math)
→ Objection-handling scripts: 5 sentences each in acknowledge/reframe/proof/question/bridge format
→ 4-touch follow-up sequence (Day 2/5/10/21) for prospects who went silent after a demo

Marketing

→ Blog post → 8 social variants (3 Twitter threads, 2 LinkedIn, 1 IG, 2 video scripts)
→ Competitor ad teardown: 8 ads → positioning table + 3 counter-positioned variants
→ SEO content brief from a keyword cluster (intent grouping, H2 structure, schema, internal links)
→ Email newsletter from 5 blog summaries (subject, opener, deep-dive, roundup, CTA)
→ Ad copy A/B variations: 5 angles for one offer (problem-aware → direct-offer)
→ Monthly content calendar from your brand pillars (20 posts, mixed format)

Customer Support

→ Ticket triage and priority scoring: 20 tickets → P1/P2/P3 + routing in a copy-paste table
→ Reply drafting that mirrors the customer's tone (frustrated, polite, urgent) under 150 words
→ FAQ generation: 50 resolved tickets → 8 KB articles with TL;DRs and escalation notes
→ CSAT comment analysis: 200 comments → top 5 praise, top 5 complaints, exec summary
→ Knowledge base gap analysis: KB + escalated tickets → 5 new articles to write
→ Canned response library: 10 ticket types → 10 templates with variables

Operations

→ SOP from a brain dump: messy notes → formatted SOP (purpose, scope, roles, procedure, decisions)
→ Meeting transcript → action items, decisions, open questions, Slack-ready summary
→ Process bottleneck analysis: cycle time breakdown, top 3 bottlenecks, quick wins
→ Contract / invoice data extraction: PDF text → JSON with dates, terms, SLAs, flags
→ 30/60/90-day onboarding plan generation from role, tools, target outcomes
→ Weekly ops report from raw KPIs + incidents + project status (400 words, exec-ready)

10 questions, real answers

How to get the most out of the pack.

The actual questions I get from operators running these prompts on their work. Each answer is the version I'd give in a 1-on-1.

How do I actually use these prompts?+

Paste the entire block into Claude (claude.ai chat or Claude Code in your terminal), fill in the bracketed inputs (like [PASTE NOTES HERE]), and hit run. The prompts are written as complete units. Don't paraphrase or strip the structure tags — those tags are what make Claude's output consistent.

Which one should I run first?+

Prompt #19, the SOP-from-a-brain-dump (shown above). It works for any process you do regularly, takes 5 minutes, and the output is immediately shareable with your team. After that, pick whichever department has the biggest fire in your business.

Will these work with ChatGPT or Gemini?+

Mostly yes. The prompts use XML-style tags (<input>, <rules>) which Claude is best at handling, but GPT-4-class models follow them too. You may see slightly different output formatting. The JSON-extraction prompts are the most portable. The strategic prompts (proposals, ad teardowns) are tuned for Claude's reasoning style and degrade a bit on smaller models.

Do I need Claude Pro or Max to run these?+

No. The free tier of Claude.ai handles every prompt in the pack. You'll hit message limits faster if you run multiple in a sitting. If you plan to run 10+ prompts a day, Claude Pro ($20/mo) is the right tier. Claude Code requires Pro or Max minimum.

How do I adapt a prompt to my specific business?+

Two ways. (1) Add a one-line context block at the top: 'My business: I run a B2B SaaS targeting mid-market accounting firms. ICP: CFOs at 50-500 person firms.' Claude will use that to tailor outputs. (2) Replace placeholder examples in the prompt with your actual product/audience/tone. The prompts are written to be edited.

What if Claude gives me a bad output?+

Three moves in order. (1) Tell Claude what was wrong: 'The cold emails sounded too salesy. Rewrite them as peer observations.' (2) Add more context to your input — Claude can't read your mind. (3) Switch to a stronger model (Opus 4.7 if you were on Sonnet). 90% of bad outputs are fixed by step 1.

Can I share these with my team?+

Yes. The pack is free for personal and team use. If you build a process around any of them, drop the prompt into your team's shared docs (Notion, Confluence, Google Docs). The whole point is to be stolen. If you want the team to run them at scale, look at Claude Code where you can save them as reusable skills.

What if Anthropic updates Claude and breaks these prompts?+

Model updates rarely break prompts at this level of structure. Anthropic's improvements usually go in the direction of better instruction-following. The prompts in the pack use stable patterns (XML tags, explicit output schemas) that have worked across Claude 3, 3.5, 4, and now 4.6/4.7. If something does break, the fix is usually adding 1-2 lines of clarification.

How do I save these in Claude Code as reusable skills?+

Inside any Claude Code project, create a `.claude/skills/<skill-name>/SKILL.md` file. Paste the prompt body and add a one-line description at the top. Claude Code will auto-load the skill whenever the description matches the conversation. The Claude Bible (free at tamnguyen.ai/bible) walks through the full skill format.

Can I chain multiple prompts together?+

Yes, and it's where the real leverage shows up. Example chain: Prompt #20 (meeting → action items) → Prompt #19 (SOP from a brain dump) → Prompt #15 (FAQ from ticket clusters). Each one outputs structured data the next one can ingest. In Claude Code, you wire chains as separate skills that call each other. In claude.ai chat, you just paste outputs from one as inputs to the next.