Most of a deal week is retrieval.
Building the list. Writing the write-up. Reading four hundred documents. Cleaning up call notes. Reformatting a memo into the house template at 11pm.
None of that is the job. The job is deciding. What fits, what the number means, who's worth a call. That part gets an afternoon. Everything around it gets the other four days.
That ratio is now a choice.
Attached to this issue: six Claude skills, one for each moment where a deal team loses hours to work that isn't judgment. Free, and built to be used the same day you read this. That's the whole newsletter. Every issue ships something you install.
What's in the kit
1. Thesis Screener. Write your thesis into a file once. Feed it a list (names, a sector, a raw Grata export) and get back a scored shortlist with rationale, signals, gaps, and lookalikes you hadn't pulled. Screens 500 names against the same bar, including the ones you'd skim past at 6pm.
2. Sourcing Memo Generator. Inputs in, partner-ready first-pass memo out. Snapshot, why-now, business, financials, draft thesis, risks, next step. You sharpen the thesis and add the thing only you know.
3. Data Room Triage. Day one, four answers: what's here, what's missing, what's concerning, what to ask for first. The gap list is the sleeper feature. What a seller leaves out is often the most important thing in the room.
4. CIM Compressor. Eighty pages to two, plus a claims-versus-evidence table that separates what's proven from what's asserted. Itemizes every addback. Ends with the ten questions the banker would rather you didn't ask.
5. Management Call Prep. A question plan built backward from what would kill the deal, phrased the way people actually talk. After the call, your messy notes become a structured record, with the deflections logged, because deflections are findings.
6. IC Memo Assistant. Assembles three months of diligence into house-format sections, with every claim traced to its source. Banker claims that were never independently verified get labeled as such. Three months in, that's the label that saves you.
They chain. The screener card feeds the memo. The memo and CIM brief feed the call plan. All of it feeds the IC memo, with a source map at the back.
The rule I'd want you to steal even if you never download this
None of these will invent a number.
Not revenue, not EBITDA, not headcount. Anything unconfirmed gets tagged [UNVERIFIED] and stays tagged, through the memo, through diligence, into the IC draft. Visible and uncomfortable, which is the point.
A model that estimates EBITDA to make a table look complete isn't being helpful. It's building a liability that walks into your investment committee wearing your name on the cover.
That's the difference between a tool a deal team can use and a demo that impresses people at a conference.
Take it
Six skills. Six template files to make them yours: your thesis, your memo format, your diligence list, your call questions, your IC template. Thirty minutes of setup and they stop sounding like tools and start sounding like your team.
Start with the screener. Run it on one sector this week. Count the names you'd never have reached.
How to set it up
Ten minutes, one time. Then every skill runs off your fund's actual criteria instead of a generic template.
1. Download and unzip the kit. Six folders, one per skill. Each has a SKILL.md and a reference/ file you'll customize.
2. Pick your setup: Claude.ai, or Claude Code / Cowork.
Claude.ai (Projects), simplest, no install:
- Create a new Project (one per skill, or one shared "Deal Process" project for all six)
- Upload the skill's
SKILL.mdand everything inreference/to Project knowledge - Paste the
SKILL.mdcontents into the Project's custom instructions - Done. Every chat in that project now runs the skill
Claude Code / Cowork, for teams running skills across tools:
- Drop the six folders into your skills directory
- Skills load automatically when a request matches, with no need to open a specific project
3. Fill in your reference file. This is the step that matters most. Each skill ships with a placeholder: your thesis, your memo format, your diligence checklist, your call questions, your IC template. Generic in, generic out. Fund-specific in, and it stops sounding like a tool.
Don't have it written down? Tell Claude: "Interview me and draft this file." Ten minutes of Q&A gets you a working first draft.
4. Run it. Start with the Thesis Screener. It's the easiest first win.
Screen these against my thesis: [paste names or a list]Here's a Grata export. Score everything, give me the top 10.Build me a target list for [sector], then screen it.
5. Verify before anything leaves your desk. Every skill in the kit tags unconfirmed numbers [UNVERIFIED]. That's not a bug. It's the whole point. Confirm those before a memo or brief goes near a partner.
BONUS!!! How Set Up Your Hermes Agent to be Private Equity Origination Associate
Meet Your New Origination Associate: Hermes Agent
What Hermes Agent is:
Hermes Agent is a free, open-source AI agent built by Nous Research, the lab behind the Hermes family of models. It was released in February 2026. You install it on your own laptop.
It is not a website you visit and it is not another chat window. It is a program that runs on your machine, remembers what it learns across sessions, and takes actions on its own: searching the web, reading files, browsing sites, and running jobs on a schedule.
Think of the difference this way. A chat window answers a question and forgets you the moment you close the tab. Hermes remembers your thesis, learns how you work, writes its own reusable playbooks from experience, and reaches you on Slack or email while you are in a meeting. One is a tool you operate. The other behaves like a hire.
One more idea to hold onto, because the setup depends on it. Hermes is model-agnostic. Hermes is the body. You choose the brain. It runs on Claude, on GPT, or on a dozen other models, and you decide which one during setup.
What you need before you start:
A Mac or Windows laptop. Ten minutes. And one decision: which brain runs it. This guide uses Claude, because the Private Equity Kit from the last issue was built and tuned on Claude, and keeping the same brain keeps your skills behaving the way they were designed to.
Step 1: Download Hermes
Go to hermes-agent.nousresearch.com and download the Hermes Desktop installer for macOS or Windows. Run it.
The installer does the technical work for you. It installs every dependency in the background, sets up the app, and makes the hermes command available on your system. You configure none of it. When it finishes, Hermes is on your machine.
Note the exact address. Several lookalike sites rank above the real one in search. The Nous Research domain above is the source of truth.
Step 2: Get your Claude API key
An API key is the credential that lets Hermes talk to Claude on your behalf. You generate it once. Here is the full path.
Go to console.anthropic.com. Sign in, or create an account with your email or a Google login. This is a developer account, and it is separate from a Claude.ai subscription. A Pro or Max plan does not include API access, so you set this up on its own.
Add billing before anything else. The key will not work until the account can be charged. Open Settings, then Billing, and add a card. Load a small prepaid credit to start. Five dollars covers a lot of memos, and you can set a monthly cap so there are no surprises. Skipping this step is the most common reason a new key fails.
Create the key. Open Settings, then API Keys, and click Create Key. Give it a name you will recognize later, such as hermes-origination.
Copy it immediately. The key begins with sk-ant- and it is shown one time only. Copy it the moment it appears and store it somewhere safe, like a password manager. If you lose it, you delete it and generate a new one.
Step 3: Add the key and choose your model
Open your terminal. That is Terminal on macOS or PowerShell on Windows. Run:
hermes model
Hermes walks you through it. Choose Anthropic from the provider list. When it asks for your key, paste the sk-ant- key you just copied. Hermes stores it privately on your machine, not in the cloud.
Then pick the Claude model. Choose Opus for the deepest reasoning on a CIM or an IC memo. Choose Sonnet for faster, cheaper passes on high-volume screening. You are not locked in. Run hermes model again anytime to switch.
Prefer speed over Claude? Run hermes setup --portal instead of the steps above. One login from Nous Research gives you a model plus web search, a browser, image generation, and speech, with no API key to manage. It is the fastest way to a working agent. For this newsletter's kit, the Claude key path keeps the exact brain your skills were built on.
Step 4: Confirm it works
Run:
hermes --tui
Type one instruction you can check by eye:
Search this week's announced middle-market software acquisitions and list the five with disclosed multiples.
If it returns a real list, the associate is live. If it stalls or errors, run hermes doctor. It tells you exactly what is missing and how to fix it.
Step 5: Give it a desk you already check
An associate you have to open a terminal to reach is not an associate. Put Hermes where you already sit. Run:
hermes gateway setup
Connect it to Slack or email. Now you brief it from your inbox, and the work comes back to the same place. The terminal was only the front door.
Step 6: Give it its standing orders
An installed agent with no instructions is an intern on day one. It has hands and no assignment. This is the assignment.
Paste the prompt below into Hermes. Fill the brackets once with your mandate. Then type run the sweep.
It does two things most tools do not. It commits your mandate to memory, so you never re-explain it. And it saves itself as a skill, so the behavior survives a restart. From then on, two words put your associate to work.
The origination prompt:
You are my private equity origination associate. Your job is to find proprietary, off-market deals that fit my firm's mandate, and to bring me a small number of high-conviction names, not a long list of noise.
Commit these standing orders to memory and save them as a reusable skill so they survive restarts. When I say "run the sweep," follow them exactly.
MY MANDATE
- Firm: [FIRM NAME]
- Thesis / sector: [e.g., vertical software for the trades: HVAC, plumbing,
electrical]
- Business model criteria: [e.g., recurring or repeat revenue, B2B, under
20% customer concentration]
- Size band: [e.g., $3M to $15M EBITDA, or $10M to $75M revenue if EBITDA is
not available]
- Geography: [e.g., US and Canada]
- Deal type: [platform / add-on to my existing platform [PLATFORM NAME] / both]
- Check size: [e.g., $15M to $60M equity]
- Hard exclusions: [e.g., no pre-revenue, no turnarounds, no pure services
with no software]
WHAT "PROPRIETARY" MEANS HERE
Prioritize companies that are:
- Founder-owned or family-owned with no prior institutional capital (no PE or VC on the cap table)
- Not currently running a broad banker auction
- Showing a credible reason it is actionable now (see signals below) Deprioritize anything already in a competitive process or picked over by sponsors. If a target is in an active auction, say so and rank it lower.
WHY-NOW SIGNALS TO HUNT FOR
- Ownership succession: founder tenure of 15 or more years, founder age
signals, retirement language, recent leadership transitions
- Fresh momentum: hiring surges, new locations, facility expansion, a recent
large contract or government award, new patent filings
- Consolidation fit: a fragmented market where this is a clean add-on to
[PLATFORM NAME]
- A single concentrated owner or decision-maker who can actually transact
WHERE TO LOOK
Use web search and the browser. Work from company databases and directories, trade association member lists, industry publications, local business journals, job boards for hiring signals, and company press releases. When a name looks strong, corroborate it against a second independent source before you surface it.
DATA INTEGRITY (NON-NEGOTIABLE)
Never invent a number. Private company financials are usually not public. If
you estimate revenue, headcount, or EBITDA, label it [UNVERIFIED] and show the basis for the estimate. Every factual claim gets a source link. Keep what is confirmed separate from what is inferred. A shorter honest list beats a longer confident one.
WHAT TO BRING ME
Return a ranked shortlist of no more than 10 names. For each:
1. Company name, location, one-line description
2. Ownership status and why it looks proprietary
3. The why-now signal, with a source link
4. Fit to mandate, and whether it is a platform or an add-on
5. Size read: confirmed figures where public, otherwise an [UNVERIFIED]
estimate with its basis
6. Decision-maker: name, role, and a likely contact path
7. A one-sentence first-touch angle specific to this company, not a template
CADENCE
When you are connected to Slack or email, run this every [weekday morning /Monday]. Only surface names you have not sent me before. Keep a running memory of every target you have flagged so you never repeat one. Open each report with the count of new names and the single strongest one.
Before your first run, ask me any clarifying question about the mandate. After that, work on your own.
Fill the brackets once. Here is a completed mandate as a model:
MY MANDATE
- Firm: Cedar Ridge Capital
- Thesis / sector: vertical software for field-service trades (HVAC, plumbing,
electrical, roofing)
- Business model criteria: recurring or repeat revenue, B2B, under 20%
customer concentration
- Size band: $3M to $12M EBITDA
- Geography: US and Canada
- Deal type: platform, plus add-ons to our platform FieldWorks
- Check size: $15M to $50M equity
- Hard exclusions: no pre-revenue, no pure services with no software, no turnarounds
Fill it once, paste the whole prompt, and answer the clarifying question it asks. After that, run the sweep is the entire interface.
Where this leaves you:
You installed the runtime and gave it a brain. You have not built the workflow yet. That is the next section: loading the six PE skills and scheduling the nightly sweep that turns this from an install into an associate that earns its seat.
The week in deals
Nestlé carves out its water business. Platinum Equity takes half. €4.9bn ($5.6bn) enterprise value. Nestlé banks €3bn in cash, keeps 50%, hands day-to-day control to a sponsor. The new entity, Peranel, houses S.Pellegrino, Perrier, Acqua Panna, and Nestlé Pure Life across 120 countries. Close expected first half of 2027. Why it matters: a clean template for corporate carve-outs. The strategic keeps the upside, the sponsor takes the operating risk.
Penske's controlling shareholders move to take it private. $210/share cash. $13.8bn implied equity value, but only $3.8bn actually changes hands, because Penske Corp and Mitsui already own 72.6%. Shares are trading above the offer, which usually means the market expects a bump. Why it matters: read the filing, not the headline. The "$3.8bn deal" is the free float, not the company.
OCS buys Mitie for £3.1bn ($4.2bn). 44.7% premium. Creates an £8.5bn revenue UK facilities management group and marks the London Stock Exchange's 11th billion-pound-plus take-private of the year. Why it matters: the delisting wave in the UK isn't slowing down.
Cinven buys Salsify. Terms undisclosed. A sponsor-to-sponsor exit. Prior backers included Permira, TPG, and Warburg Pincus. 2,000+ customers, 750M products syndicated across 2,600 commerce destinations. Why it matters: no price disclosed is its own signal. Worth a follow-up call to the banker.
Var Energi buys BlueNord for $1.3bn. Creates Europe's largest independent oil and gas company. Strategic buyer, not a sponsor. Included here because size clears the bar.
Also this week:
- CapVest completed its majority buy of TSG, a 30-country energy infrastructure services provider
- Truelink Capital bought Lyons Magnus from Paine Schwartz
- Guardian Dentistry Partners (backed by Morgan Stanley Private Credit, TPG Twin Brook, Prudential) took a majority stake in Select Dental Management: 38 locations, 130 dentists, eight states
On the LP side: GoldenTree closed its second private credit fund at a $2.75bn hard cap. Clipway, founded by ex-Ardian execs, closed a $6.4bn debut secondaries fund. Capitol Meridian doubled its flagship to $1.9bn.