How to Build Claude Skills for M&A: Do the Work Once, Reuse It on Every Deal

    How to Build Claude Skills for M&A: Do the Work Once, Reuse It on Every Deal
    Simone Vascotto··9 min read

    Key Takeaways

    • 1.A Claude Skill is a folder with a SKILL.md file. Claude reads only its name and description until a task matches, then loads the instructions and any bundled files it needs.
    • 2.The value in M&A is reuse. Every deal repeats the same tasks, and a skill turns your best version of each task into the default instead of a one-off prompt.
    • 3.Skills carry method, not deal data. The target's documents stay in the deal environment; the skill carries how your firm reads them.
    • 4.Building one is mostly writing: a precise description, short numbered steps, a fixed output format, and a test on a closed deal.
    • 5.You don't start from zero. Anthropic publishes private equity and investment banking skills, and some community diligence skills exist. None replaces a human check.

    Short answer

    Claude Skills are folders of instructions, and optionally scripts and reference files, that Claude loads only when a task calls for them. For deal teams, a skill is a way to do the hard thinking about a task once (how your firm screens a target, structures a diligence request list or writes a red-flag memo) and then run that same method on every future deal. You build one by writing a short SKILL.md file. Below is a copy-ready example for a red-flag memo, plus links to existing diligence skills you can start from.

    I've written about what Claude Skills change for deal teams, including the guardrails around confidential data. This is the practical follow-up: how to build one and what you can copy today.

    What are Claude Skills?

    Anthropic introduced Agent Skills as "folders that include instructions, scripts, and resources that Claude can load when needed." In practice, a skill is a directory with one required file, SKILL.md, and anything else the task needs: a template, a checklist, a reference document or a script.

    The SKILL.md file starts with two required fields in YAML frontmatter, name and description, followed by plain Markdown instructions. Per Anthropic's developer documentation, the name can be up to 64 characters of lowercase letters, numbers and hyphens, and the description up to 1,024 characters. The description has to say both what the skill does and when Claude should use it, because that is what Claude matches your request against.

    What separates a skill from a saved prompt is progressive disclosure, in three levels:

    • Metadata (name and description) is always loaded, at roughly 100 tokens per skill.
    • Instructions (the body of SKILL.md) load only when the skill is triggered.
    • Resources and code (bundled files and scripts) load only when the instructions reference them. Scripts run, and only their output enters the conversation.

    So a team can install dozens of skills without cluttering every chat.

    Skills work across Claude's apps, Claude Code and the API, but custom skills don't sync between them: each surface needs its own upload. Anthropic has also published Agent Skills as an open standard, so the format isn't tied to a single product.

    Why do reusable skills matter in M&A?

    Most deal work isn't new. Every deal needs a screen against the thesis, a diligence request list, a pass through the data room, a red-flag summary, a synergy case and a 100-day plan. The target changes. The method shouldn't.

    Without skills, that method lives in people. One associate has a sharp prompt in a notes app. Another improvises. The IC gets two memos built to two standards.

    A skill fixes three things.

    Consistency. The same steps, the same output format and the same checks run on every deal, regardless of who is typing. Anthropic's documentation frames the benefit simply: "Create once, use automatically."

    Institutional memory. The lessons from the last deal stop disappearing into a post-mortem deck. If a post-close surprise came from a contract clause nobody flagged, you add that check to the skill once, and every future deal gets it.

    Speed where it's safe. The structuring work is done before the deal starts. The team's time goes to judgment: what the findings mean for price, structure and the plan.

    A prompt is as good as the person who wrote it, on the day they wrote it. A skill gets better every time someone updates it after a deal.

    Under the Hood of M&A

    Get articles like this one weekly.

    Subscribe on Substack

    Where do Claude Skills fit across the deal lifecycle?

    Not every task deserves a skill. Anthropic's authoring guidance is to keep skills concise and build them around real gaps, tested with real usage. In M&A, the best candidates are the tasks you repeat on every deal.

    Deal stageSkillWhat it standardizesHuman check
    Sourcing and screeningThesis-fit screenPass/fail logic against your thesis and exclusion criteria, with explicit information gapsDeal lead confirms gaps before outreach
    Diligence kickoffRequest list builderWorkstream request lists tailored to sector and deal type, with priority and statusWorkstream leads add deal-specific items
    Data room reviewDocument triageHow documents are indexed, summarized and cited back to file and pageAssociate spot-checks every citation
    FindingsRed-flag memoSeverity scale, evidence labels and a fixed memo formatPartner decides what goes to IC
    ValuationSynergy case reviewEvery synergy assumption tied to a source, an owner and a timelineDeal and operating teams agree the case
    Integration100-day plan builderDiligence findings carried into the plan's sequencing and milestonesIntegration lead owns the plan
    Culture and ways of workingTeam-level working-style reviewQuestions on how teams decide, coordinate and handle change, summarized at team levelHR and the integration lead agree actions

    The last row is where a skill helps most, because teams most often improvise it, and where guardrails matter most. A good working-style skill looks at how teams decide and work together, never at individual scores or individual exits, and it treats what it finds as places to look rather than verdicts. For the distinction between the two lenses, see Cultural Due Diligence vs Behavioral Due Diligence.

    How do you build a Claude Skill, step by step?

    1. Pick one task you repeat on every deal. If you give Claude the same correction every time, that correction belongs in a skill.
    2. Collect your best past output. Take the memo or request list your team considers the standard and strip every target name, number and data-room detail. You want the structure, not the deal.
    3. Write the description first. Say what the skill does and when to use it, including the words your team actually types ("red flags," "issues list," "what worries us"). Anthropic's best practices recommend writing it in the third person.
    4. Write short, numbered instructions. Assume Claude already knows what EBITDA is. Spend your words on what is specific to your firm: your severity scale, your evidence rules, your output format. Anthropic advises keeping the SKILL.md body under 500 lines and moving detail into separate files.
    5. Bundle what should be identical every time. A memo template, a sector checklist or a script that totals exposure belongs in the folder. The instructions point to it, and Claude reads it only when needed.
    6. Test on a closed deal. Run the task with and without the skill, compare, and fix what it missed.
    7. Install and share. In the Claude apps, you upload a zipped skill folder under Customize > Skills, and Anthropic's skill-creator skill can walk you through building one interactively. In Claude Code, skills live in ~/.claude/skills/ for you or .claude/skills/ in a shared repository for the team. Through the API, skills are uploaded once and shared across the workspace.
    8. Give the skill an owner. One person updates it after every deal post-mortem. That is the step that turns a file into institutional memory.

    Build your own: an example red-flag memo skill

    Here is a skill you can copy. Save it as SKILL.md inside a folder named dd-red-flag-memo, add your own memo template as memo-template.md, and adjust the severity scale to your firm's. It is an example template, not a finished product: test it on a closed deal before you use it on a live one.

    ---
    name: dd-red-flag-memo
    description: Drafts a due diligence red-flag memo from findings and data-room excerpts the user provides. Rates each issue by severity, labels evidence versus inference, and cites the source document and page. Use when the user asks for a red-flag memo, issues list, key diligence findings, or "what should worry us" on an M&A target.
    ---
    
    # Due diligence red-flag memo
    
    ## Inputs to confirm before drafting
    - Deal type (platform, add-on, carve-out, merger) and the deal thesis in one or two sentences.
    - The findings or document excerpts to review. Use only what the user provides.
    - Any issues the deal team has already priced or accepted.
    
    ## Steps
    1. List every potential issue, grouped by workstream: financial, commercial, legal, tax, IT, HR and organization, operations.
    2. For each issue, record the source document and page or clause. If there is no source, label it "Gap: evidence not provided".
    3. Label every statement as Evidence (quoted or directly stated in a source) or Inference (your reading). Never present inference as fact.
    4. Rate severity:
       - Deal-breaker: could stop the deal or change the structure.
       - Price-relevant: should move valuation, the SPA or the escrow.
       - Integration-relevant: does not change the price but must reach the 100-day plan.
       - Monitor: note and revisit before signing.
    5. Link each issue to the part of the deal thesis it affects. If it affects none, say so.
    6. For people and organization issues, describe teams, functions and ways of working only. Do not assess, score or speculate about named individuals.
    7. List the questions the deal team still needs answered, with the owner who should ask them.
    
    ## Output format
    Follow memo-template.md. Open with the top five issues in a table:
    | Issue | Workstream | Severity | Evidence or inference | Source | Thesis link |
    Then one short paragraph per issue, then open questions.
    
    ## Rules
    - Do not invent figures, dates or clauses. If something is missing, flag it as a gap.
    - Keep deal data out of this skill folder. It is supplied per deal.
    - End with: "Draft for review. A named deal team member must check every source before this memo is shared."

    Three lines do most of the work. Step 2, because a memo without sources can't be checked. Step 3, because a model writes inference in the same confident tone as fact. And the closing rule, because a skill standardizes the work. It doesn't sign it off.

    Which existing M&A Claude skills can you start from?

    Here is what exists publicly. I've opened each link but haven't audited the code, so treat the community ones as starting points, not finished tools.

    From Anthropic

    From the community

    • Buy-Side Due Diligence: a skill for private company acquisitions that normalizes earnings, cross-references documents and turns red flags into valuation adjustments. Its README says it is not a replacement for a quality-of-earnings report.
    • The Deal Kit (open-deal-kit): an MIT-licensed collection of prompts, diligence skills and agents aimed at searchers, independent sponsors and small funds.
    • PE Due Diligence Synthesizer: a skill write-up for indexing a data room, cross-checking claims across documents and producing a citation-linked diligence memo.

    One gap stands out: I didn't find a public skill for cultural or behavioral diligence at team level. Most teams still run that stage on interview notes and instinct, which makes it a good candidate for your own.

    Before you install anything, follow Anthropic's own advice: use skills only from trusted sources, and audit every bundled file, because a skill can direct Claude to run code. The documentation's security section is short and worth reading before your first install.

    What should never go into a skill?

    Skills make good practice repeatable. They make bad practice repeatable too. Three things stay out: deal data (supplied per deal, inside your firm's approved environment), decisions about named individuals, and any output without a named human check.

    The firms getting value from skills don't have the cleverest prompts. Their standards stopped living in one person's head.

    About the Author: Simone Vascotto is the Founder of HumanAQ, the AI-powered platform for Cultural Due Diligence and leadership alignment in M&A.

    This piece is part of the Under the Hood of M&A newsletter.

    Frequently Asked Questions

    What is a Claude Skill?
    A folder containing a SKILL.md file with a name, a description and instructions, plus optional templates, reference files and scripts. Claude loads it only when a request matches the description.
    How is a skill different from a saved prompt?
    A saved prompt has to be pasted in and depends on who uses it. A skill is picked up automatically from its description, loads only when needed, can bundle templates and scripts, and improves each time its owner updates it.
    Do I need to code to build a Claude Skill?
    No. The core of a skill is a Markdown file. Scripts are optional, for steps that must be exact. Anthropic's skill-creator skill can guide you through building one.
    Can I use Claude Skills on confidential deal documents?
    Only within your firm's approved setup and the terms of your NDA and data room. Keep deal data out of the skill itself, check Anthropic's current data-handling terms for the surface you use, and have a person check every output against the source documents.
    Are there ready-made Claude skills for M&A due diligence?
    Yes. Anthropic's Claude for Financial Services repository includes private equity skills for screening, diligence checklists, IC memos and 100-day plans, and several community diligence skills are public on GitHub. Test any of them on a closed deal first.