10 Claude Skills That Replaced My $50/hr Virtual Assistant

The "AI is replacing virtual assistants" claim has been wrong for two years because it's been framed wrong. A VA isn't one job. A VA is 10-15 specific recurring tasks bundled together because hiring one person is cheaper than hiring ten specialists. Replacing a VA with AI doesn't mean building one mega-agent that does everything; it means building 10 specific skills that each do one task perfectly.

That's what I did 18 months ago. Stopped paying $50/hour for VA work. Built 10 Claude skills. Ran them myself. Saved approximately $26,000 in the first year and produced better output than the VA on most of the tasks. This article is the inventory: which 10 skills, what each one does, and what the gotchas are.

The Frame: A Skill, Not An Agent

A "Claude skill" in operator-context is just a reusable system prompt that handles one recurring task with consistent format and tone. You paste it into Claude (or save it as a Claude Project), feed it the task-specific inputs, and get back a structured output you can edit and ship.

That's it. No autonomous agent. No tools-and-API-calls. No 30-minute setup per task. The operator stays in the loop. The skill just compresses the time it takes to produce the deliverable from 60 minutes to 6 minutes.

This frame matters because it explains why "AI agents" haven't replaced VAs at scale yet — agents try to remove the human, and the human is what catches the errors. Skills keep the human and remove the typing.

Skill 1: Project Brief

The most-used skill in my stack. Input: a 3-sentence description of a project I'm scoping. Output: a structured brief with objective, success criteria, deliverables, constraints, risks, and dependencies.

The skill works because it forces me to think about all six dimensions every time, even when I'd be tempted to skip a few. Claude won't let me leave a section vague — it asks clarifying questions if I underspecify, which is the actual value.

Cycle time: 4-6 minutes from input to finished brief. The same brief written manually takes me 25-40 minutes because I keep getting distracted.

Skill 2: Email Triage

Input: a paste of my inbox (last 24 hours). Output: a sorted list — urgent/respond now, important/respond by EOD, FYI/no action, archive — plus draft replies for the top 5 emails.

The triage skill saved me roughly 10 hours per week within a month of building it. The drafts aren't perfect, but they're 80% of the way there, and editing 80% beats writing from scratch.

The trick is the sorting heuristic: Claude is told what "urgent" means in my context (specific senders, specific subject patterns), so the classification is calibrated to my actual priorities, not generic email-urgency conventions.

Skill 3: SOP Builder

Input: a description of a process I just did. Output: a clean SOP document with steps, expected outcomes, common failure modes, and the decision points.

I run this skill every time I do a new recurring process for the first time. The result is that I have a documented SOP for almost every operational task I've done in the last 18 months — and the SOPs are good enough that I can hand them to a contractor or use them to train a future hire.

Cycle time: 8-12 minutes for a moderately complex SOP. Writing the same SOP by hand takes 45-60 minutes. The skill produces tighter writing than I do because it's not editing as it drafts.

Skill 4: RFP Responder

The most expensive skill in the stack — not because it costs more to run, but because it generates the most revenue per execution. Input: the RFP document + my company's standard capabilities/qualifications. Output: a structured response that maps directly to the RFP's evaluation criteria, with pre-filled sections that are 80% complete.

Time saved per RFP: 6-10 hours. Revenue impact: substantially higher win rate because the response is well-structured and on-format, not rushed.

This skill alone justifies the entire Claude Pro subscription several times over for any operator doing B2B service work.

Skill 5: Hiring Filter

Input: a paste of resumes + the job spec. Output: ranked shortlist with reasoning, must-have-misses flagged, and 5 suggested screening questions per candidate.

The skill replaces the "look at 60 resumes" phase of hiring with a 90-second pre-sort. The shortlist is rarely perfect — but it surfaces candidates who would have been buried under volume otherwise, and it flags red flags I might have missed on a tired Friday afternoon.

The key constraint: bias control. The system prompt explicitly tells Claude to evaluate against the job criteria only, not against demographic signals. This is one of the skills where the system prompt itself is the most important asset.

Skill 6: Investor Update

Monthly investor update generation. Input: bullet points on what happened that month (metrics, hires, customers, product, learnings). Output: a structured investor update in the canonical format — TL;DR / metrics / wins / losses / asks / what's next.

The skill matters because the investor update is the kind of writing that's easy to skip when you're busy, which is the worst time to skip it (investors who don't hear from you assume the worst). The skill drops the friction so low that I write the update in 15 minutes instead of skipping it for two months.

Skill 7: Customer Reply

Input: customer email + my product/service context. Output: a draft reply that addresses every question asked, in my voice, with appropriate empathy and specificity.

The skill is calibrated to my actual customer voice through 20 examples of replies I've written. Without those examples, Claude defaults to a generic "thank you for reaching out" tone that's wrong for my brand. With them, the drafts pass the "would I send this?" test about 75% of the time.

The 25% that don't pass usually need a single tone adjustment, not a full rewrite. Edit time: 60-90 seconds per reply.

Skill 8: Vendor Negotiation

Input: a contract or vendor proposal. Output: a redline-style critique with specific clauses to push back on, a benchmark on whether the pricing is market-rate, and a suggested counter-offer with rationale.

This skill prevents me from accepting bad terms because I was too tired to read carefully. It catches things I would have missed — late-payment penalties, auto-renewal clauses, hidden cost escalators, IP assignment ambiguity.

The skill is best used as a pre-review before I read the contract myself. Claude catches the structural issues; I catch the relationship-specific issues. Together it produces better negotiations than either alone.

Skill 9: Weekly Review

Input: my calendar, my task list, and 3 sentences on how the week felt. Output: a structured weekly review — what got done vs. what was planned, lessons learned, recurring patterns, and the 3 things to prioritize next week.

The weekly review is the operational metabolism of a solo founder. Without it, weeks blend together and the same mistakes repeat. The skill makes the review take 15 minutes instead of 45 minutes, which is the difference between "I do this every Friday" and "I'll do it next week, I promise."

Skill 10: Meeting Notes

Input: a meeting transcript (from Otter, Fireflies, or pasted Zoom transcript). Output: structured notes — decisions made, action items with owners, open questions, and a one-paragraph summary.

This skill changed how I run meetings. Knowing the transcript will get post-processed cleanly means I can stop taking notes during the meeting and actually pay attention. Cycle time per meeting: 90 seconds of skill execution + 60 seconds of review.

The Stack In Practice

A typical week:

• Monday: Email Triage (morning), Investor Update for the prior month (afternoon), Weekly Review of the previous week.
• Tuesday-Thursday: Customer Reply 4-8 times per day, Meeting Notes after every meeting, Project Brief when scoping anything new.
• Friday: Email Triage, SOP Builder if I did anything new and recurring, RFP Responder if there's a proposal in flight, Hiring Filter if I'm in an active hiring loop.

Total time spent in skills per week: 5-7 hours. Total output: deliverables that would have required a VA working 12-15 hours per week. Cost: $20/mo for Claude Pro vs. $2,500/mo for the VA.

That math is why I stopped hiring assistants for recurring work two years ago. Skills aren't a replacement for human judgment — but they're a replacement for almost all the typing.

What You Don't Get Without The System Prompts

The shape of each skill above is straightforward. The execution is where the value lives. Each skill needs the actual system prompt — calibrated for the specific task, with the right edge-case handling, the right format, the right tone instructions, and the right verification gates.

I've written and refined those 10 system prompts over 18 months. The version I use today is in the Claude Skills for Operators playbook ($7.99) — all 10 prompts with their input templates and the failure-mode debugging chart that took me a year to write.

If you're paying for VA work that's mostly repetitive — email, meeting notes, briefs, drafts — the math says try the skills before extending the VA contract. Total time investment to test all 10: one weekend. Total cost: $7.99 for the templates plus $20/mo for Claude Pro. The math doesn't break even with a VA; it laps it.