Spine Built My Best Proposal. Then I Copied Its Process Into a $0 Claude Skill

An AI research tool wrote a better proposal than I could. Instead of subscribing, I reverse-engineered its process — three research artifacts and one formatting device — into a Claude Skill I own, at zero marginal cost.

Spine Built My Best Proposal. Then I Copied Its Process Into a $0 Claude Skill

The most valuable thing an AI research tool gives you is not the document it produces — it is the process it follows to produce it. Last week I used Spine to build a client proposal that was better than anything I had written by hand. Instead of subscribing and moving on, I spent one afternoon reverse-engineering how it worked and encoded that process into a Claude Skill. Every proposal I write now follows the same professional method, and it costs me no additional subscription.

This is the story of that afternoon, and the thinking behind it.

What is Spine, and what does it actually do well?

Spine (getspine.ai) is an AI research platform that, in its own words, "orchestrates 300+ AI models as parallel agents to conduct deep research and produce polished reports, slides, spreadsheets, and dashboards." Instead of a chat thread, you get finished deliverables: .docx briefs, .pptx decks, Excel models with scenario tabs, interactive dashboards.

Two things stood out to me immediately:

It dispatches work in parallel, not sequentially. Multiple agents handle separate workstreams and pass structured output forward. That is an architectural choice, not a prompt trick.

It takes sourcing seriously. Spine claims 87.6% on Google DeepMind's DeepSearchQA benchmark, ahead of Perplexity (79.5%), Claude (76.1%) and ChatGPT (71.3%). Whatever you make of vendor-published benchmarks, the behaviour I observed matched the claim: every table it produced carried an evidence column.

Pricing runs from a free tier (2,000 credits/month) through Starter ($16–50/month annually) to Max ($80–299/month), with $10 top-ups for 10,000 credits. Reasonable pricing for what it does. That is not the point of this post.

The proposal that changed my mind

I ran a real lead through it: a newly opened Vietnamese restaurant in Windsor, Ontario, with essentially no digital presence. What came back was a 12-page proposal that opened with congratulations rather than a pitch, and it was better than my previous ones in ways I could name precisely.

That last part matters. When something is better in ways you cannot name, you have to keep renting it. When you can name them, you can own them.

What I actually copied: three artifacts, not one document

Here is the insight that made the whole exercise worthwhile. The proposal was not one output. It was the third output in a chain, and the first two never went to the client:

1. A Digital Footprint Audit. A fixed ten-channel checklist — Google Business Profile, Instagram, Facebook, website, TikTok, delivery apps, ads presence, reviews, CRM tooling, and name clarity in search. Each row carried a status and its evidence.

2. A Competitive Landscape table. Seven named local competitors with rating and review count, website yes/no, online ordering yes/no, differentiators — and the lead's own row at the bottom.

3. A Findings document. Strengths and weaknesses, where every single bullet cited which of the two tables it came from.

The proposal was then assembled from those three. That sequence is the entire product. A proposal written without it is a template with a name swapped in.

Why does citing your own research tables matter so much?

Because it forces a discipline that is very hard to maintain by willpower alone: if a claim cannot cite a row, it is an assumption, and it gets cut.

Three habits inside those tables did most of the work:

The one device worth stealing: the third column

If you take nothing else from this post, take this.

The audit table had a third column titled "What It Costs You Today." The pricing table had a third column titled "Gap From Section 2 It Closes."

Same device, pointed in opposite directions. Every finding has to earn a consequence. Every dollar has to earn a reason. And it has a brutal side effect: any line item that cannot name a gap it closes gets deleted from the proposal, because the empty cell makes the padding visible.

The second-best device was structural: put the client's empty row directly underneath a competitor showing "4.7 from 806 reviews," and the argument makes itself. Then find the open lane — in Windsor, only one of seven competitors even had a website — so the picture ends in opportunity rather than defeat.

How do you turn a process into a Claude Skill?

A Claude Skill is a folder with instructions that Claude Code loads on demand. Mine has five files and about 900 lines total:

Writing it down surfaced things I had been doing inconsistently for a year. The skill now refuses to invent a price, refuses to fabricate a case study, and refuses to mark a proposal as sent — a human has to click that button. Those constraints are not limitations; they are the reason I trust the output enough to put my name on it.

So what does it actually cost now?

Let me be precise, because "free" is the kind of claim I just spent 800 words arguing against.

The marginal cost of the next proposal is effectively zero: no new SaaS line item, no per-document credits, no seat I have to justify. It runs inside the Claude Code subscription I already pay for as a developer. What used to be a recurring tool decision is now a file in a folder.

But the honest version is this: I did not get the process for free. I paid for it by using a tool that had already solved the problem, then reading its output closely enough to understand why it worked. That afternoon of reverse-engineering was the price.

The real lesson: rent the output once, own the process

Tools churn. Pricing changes, companies pivot, the model behind your favourite feature gets deprecated. Process compounds.

So the next time a tool produces something genuinely better than your own work, resist the urge to just subscribe. Ask the harder question: what did it do that I was not doing? Write the answer down in enough detail that you could hand it to someone else. That document is the asset. The subscription was only ever the delivery mechanism.

I still think Spine is excellent, and I would recommend it — especially the free tier, precisely as a teacher. Use it on a real problem you care about. Then go read what it gave you like an engineer, not a customer.

FAQ

What is Spine (getspine.ai)? An AI research platform that runs 300+ AI models as parallel agents to produce finished deliverables — reports, .pptx decks, Excel models and dashboards — rather than chat responses. It has a free tier with 2,000 credits/month, with paid plans from roughly $16 to $299/month.

What is a Claude Skill? A folder of instructions, reference files and scripts that Claude Code loads on demand when a task matches. It turns a repeatable process into something you own and version-control, rather than a prompt you retype.

Is building your own skill really free? The marginal cost per document is effectively zero — it runs inside a Claude Code subscription you likely already have, with no separate per-document credits. The real cost is the upfront work of extracting and writing down the process.

What made the AI-generated proposal better than a hand-written one? Structured research before writing: a fixed-channel digital audit, a named competitor table, and a findings document where every claim cites one of those tables. Plus one formatting device — a third column linking each finding to its cost and each cost to the gap it closes.

Can this approach work for documents other than proposals? Yes. Any repeatable professional document with a research phase — audits, due-diligence memos, competitive briefs — follows the same pattern: build the research artifacts first, then assemble the document from them.

Reference: Spine — getspine.ai

#AI #ClaudeCode #Automation #Productivity #Proposals #AgencyLife #Spine

✍️ The Author: Do Ngoc Hoan Founder of CookConnects.ca & Wizy.ca. Bridging the gap between advanced algorithms and business execution. I write for technical founders looking to scale their impact with AI and robust engineering.

← Blog