Before you write anything, lay out your plan as a short numbered list and state what you’re assuming about my setup. Then stop and wait for my go.
[starter-kit]
The Operator's
Claude Starter Kit.
The context file you fill in once, plus the six operator moves as copy-paste prompts — the setup that turns Claude from a vending machine into a teammate. It's free. Take it.
[part 1 · context file]
Fill this in once. Reuse it forever.
Drop this into a Claude Project (claude.ai) or save it as CLAUDE.md in your project folder. Fill the brackets, delete what doesn't apply. This single file is the highest-leverage move in the kit — it's the difference between Claude guessing and Claude knowing.
# Context for Claude ## Who I am [Your role. e.g. "Network admin running 3 sites" / "Solo founder building a SaaS" / "Analyst on the finance team."] ## What I'm working on [Current project or recurring work, in a line or two. What does "done" look like?] ## My stack / tools [The specific tools you live in. e.g. "PowerShell, Cisco IOS, Azure" / "Python, Postgres, dbt" / "Excel, SQL, our BI tool."] ## My conventions — what "good" looks like [How you like things done. e.g. "Concise. Show your reasoning on anything risky. Comment the code. No hype."] ## Hard rules — never do this [Your guardrails. e.g. "Never run destructive commands. Propose changes, don't apply them. Never send anything on my behalf."] ## How to work with me - Plan first on anything non-trivial; wait for my go before you build. - When you're done, review your own output as a skeptic before handing it over. - If you're guessing, say so. Ask instead of assuming.
[part 2 · the six moves]
Copy. Paste. Go.
Now review what you just produced as a skeptical senior [engineer / editor / analyst] who has to put their name on it. What’s weak? What did you assume? What’s the single riskiest part, and how would you fix it?
Ground rules for this session: read-only unless I explicitly say otherwise. Propose changes — don’t apply them. Never delete, send, or touch anything live without asking me first. If an action is irreversible, stop and confirm.
Give me three approaches, ranked, with the trade-off that decides between them. Tell me which you’d pick and why — in one sentence.
Before you answer, tell me what you can actually see here versus what you’re inferring. Don’t fill gaps with guesses — list the gaps and ask.
That worked. Turn it into a reusable template/checklist I can run again next time, with the parts I’ll need to change marked clearly.
Run a network or infrastructure for a living? The PacketPilot AI prompt pack goes deeper — 50+ field-tested prompts for audits, migrations, troubleshooting, and the 3am comms.
▸ the full field pack $29Independent. Operator-built. We use this stuff harder than we talk about it.