An agentic harness for Claude Code
The System
The AI most operators use forgets everything the moment you close the tab. This one remembers your business, and gets sharper every time you run it.
It starts from a repo you already have, your ICP, your calls, what broke last quarter. The system reads it before it does anything, so nothing starts from a blank prompt.
00_foundation/ 01_professional/ .claude/skills/ CLAUDE.md
You’re in. Here’s what running it looks like.
You name an account. An agent pulls the company, the buying committee, and the pain signals hiding in old notes, reading your context first, so it already knows your ICP and your last calls.
It runs while you’re in your next meeting. You get back a real dossier, not a prompt you babysit. You steer it in plain English; you own every call it makes.
See how it works: the research agent · every workflow
the artifact it hands back
MEETING DOSSIER — Sales Discovery
Objective: Qualify pain, map buying committee
EXECUTIVE SUMMARY
- Prospect's team doubled in 6 months — likely
scaling pains in tooling
- Champion (Dir. RevOps) posted last week about
"outgrowing our current stack" — warm entry
ATTENDEE: Sarah Chen — Dir. Revenue Operations
Tenure: 8 months [VERIFIED — LinkedIn]
Priorities: tool consolidation, reporting [INFERRED]
Decision authority: Champion [INFERRED — title + initiator]
PREPARED QUESTIONS
1. "You posted about outgrowing your stack — what
specifically is breaking at your current scale?"
Reason: opens with their signal, not our agenda.· a real dossier, anonymized — evidence-tagged, not a summary
One system, not four tools.
Context that persists, skills that do the work, your stack wired in, and workflows that run them. Each run feeds the next, so the whole thing compounds.
what persists
00_foundation/ 01_professional/ .claude/skills/ CLAUDE.md
· a real repo you own
the tools that do the work
your stack, wired in
the loops that run them
↻ back to Context — it compounds
A skill is one move. A workflow is the play: an agent running a sequence of skills without you in the middle.
Context engineering →
One operator. A team’s worth of functions.
Prospecting, content, RevOps, competitive intel. 49 skills across the work you actually own, and every one starts from your context, not a blank prompt.
Pipeline & Prospecting
- ICP Builder
- Prospect Research
- Hypothesis Builder
- Cold Outbound Writer
- Email Sequences
- Meeting Prep
Revenue Operations
- Deal Health Scoring
- Pipeline Dashboard
- Revenue Forecast
- Rep Coaching
- Territory Analysis
Content Engine
- Content Production
- Content Editing
- Persuasive Copy
- SEO Optimization
- Topic Planning
Conversion & Messaging
- Skeptical Buyer Test
- CRO Page Audit
- Content Quality Gate
- Data Storytelling
- Legal Review
Creative Studio
- Image Generation
- Video Generation
- Slide Decks
- Document Design
- Spreadsheets
AI & Strategy
- Deep Planning
- Prompt Engineering
- Knowledge Synthesis
- Execution Engine
- Requirements Discovery
- Skill Builder
Development
- UI/UX Design
- Frontend Code
- Feature Architecture
- Spec-Driven Dev
- Test-First Code
- Git & PR Workflow
Internal Tools
- Knowledge Search
- Onboarding
- Agent Memory
- Pattern Learning
- System Install
That’s the surface area. Here’s what one of them actually looks like running.
Work you used to shelve. Now on the table.
Whole functions you kept putting off because there was no one to run them. One operator covers them now, because the system already knows your business.
A competitive-intel function
Battlecards and positioning that exist before the RFP shows up, not improvised in the room.
the pain it kills
“When a big competitor shows up in an RFP, there are no battlecards, no positioning. Every pitch is improvised.”
A content engine
Thought leadership and case studies that ship on a cadence, without a dedicated headcount.
the pain it kills
“A newsletter exists, but there’s no engine behind it — no systematic way to produce thought leadership or case studies.”
Pipeline reviews that don’t eat the week
The board reporting that took two weeks pulls together in an afternoon, from your own data.
the pain it kills
“The weekly review is still pretty suboptimal — we haven’t had easy ways to get the statistics and analysis out.”
None of this needed a new hire. It needed a system that already knows your business.
How one of them actually runs.
One command. It reads your context first, works while you don’t, and hands you something you’d walk into the room with.
- 01
One command, by name.
You type the prospect. It already knows your ICP and your last three calls with them, because it read your context before it read your request.
- 02
It works the room.
Company, attendees, the buying committee, the pain signals hiding in old notes, pulled and cross-checked while you’re in your next meeting.
- 03
You fix it in plain English.
“Wrong Renske — the CS seat is Jamie.” It re-maps the committee on the spot. No settings, no fields, no syntax to learn.
- 04
A brief you’d walk in with.
Not a summary. Attendee cards, the history, the questions worth asking, in one file you could open in the parking lot and be ready.
It compounds.
A one-off prompt forgets you the second you close the tab. A system remembers, so every run leaves the next one a little sharper. That’s the difference between a tool you use and a system that grows.
You don’t tour this graph. You feed it. Week one it’s sparse; by month two it’s dense with how you actually work, because every job you run leaves a trace the next one reads.
Memory
Persistent facts across every session. What you corrected once, it doesn't get wrong twice. The system keeps a running memory of how you work, and reads it before it starts.
Hooks
Eleven lifecycle automations that fire on their own. They load your context when a session starts, capture what changed, and checkpoint before it forgets. The plumbing that makes it feel less like a chatbot and more like a coworker.
Learnings
Every run feeds the next. The system has logged 2,321 lessons from its own work across 13,755 recorded changes, what broke and what worked, and pulls them forward so it stops repeating mistakes. The knowledge graph ties it together.
None of this shows up in a demo. All of it shows up by month two.
Week 1: 5,900 calls analyzed. Zero CLI experience.
STEEPWORKS launched January 2026. These are real results from the first cohort.
“I don't really have any questions. It's just been working. I've just been ripping in there doing stuff.”
Head of RevOps at Great Question
“For a RevOps person, two or three hours of work to make the product team happy is good.”
“I created 2 repos one for my affiliate website and one for my personal stuff. I got the personal one job hunting for me today, affiliate one found a bunch of resources incredibly fast, but I'm also kind of not sure what all to do with it, feels like someone gave me God powers and I'm telling it to get me a beer from the fridge.”
“Hour and a half after our onboarding I had a usable competitor analysis dashboard for a client. Without Claude I would have needed a week and a dev to complete this.”
“Within the first week, I built a blog, loaded our acquisition context, and started building an SDR agent that evaluates 15 months of outbound data to write better copy. It's been awesome having the system take on the organization of everything, skills to just go faster, and context that helps explain how things are interconnected.”
“This Claude setup is great. Just used it to manage some personal tax stuff for Spain. Would have cost me 3K for a lawyer to do it and many hours of manual spreadsheeting data entry. Gemini f*cked it up but our Claude's legal review and excel skills diagnosed it and got it right.”
“I am loving it man, really getting into it. Using it now for taxes, building an AI personal training app, writing resumes and cover letters for jobs, as a career coach, and to make a training plan for me to recover from some injuries and prep for a big tourney in October.”
Join the early adopters
Limited pricing available now
Get started
It’s yours. Plain files you can read.
Everything the system knows lives in plain files in a repo you own: skills, agents, the knowledge graph. Open them, read them, change them.
Your team extends it the same way. Clone it, hand it to a new hire, port it to your next company. Nothing’s trapped behind a vendor’s login.
your-system/
├─ .claude/skills/ (every skill, plain markdown) ├─ agents/ (the workflows that run them) ├─ .knowledge-graph/ (what it's learned about you) ├─ 00_foundation/ (your ICP, voice, methods) └─ CLAUDE.md (how it all fits)
· plain files, in a repo you own
Built by an operator who runs this daily, not a vendor selling it. Who’s behind it · What it costs
Strap in.
One day to live. A few weeks to compound. A system that’s yours to keep.