Key Takeaways
- 1.A Project knows the deal. A Skill knows how your firm does this kind of work. Use both together.
- 2.A skill pays off when Claude does a task repeatedly and you want it done the same way every time, which covers most of the M&A process.
- 3.A skill standardises the work. It does not sign off on it. Every stage needs a named human check.
- 4.Keep deal data out of skills. A skill carries method, not target names, financials or data-room material.
- 5.Start with three skills, not thirty: screening, diligence and IC, each with an owner who updates it after every deal post-mortem.
Last reviewed: October 2026. Claude's features change quickly, so check the linked Anthropic documentation before you build on any product detail below.
Quick answer: Claude Skills are folders of instructions, and optionally scripts and reference files, that Claude loads only when a task matches them. For M&A teams, a skill turns your best screening memo, diligence synthesis or Day-1 brief into a repeatable workflow, so every deal gets the same standard. It doesn't depend on who wrote the prompt.
Most deal teams are past the "should we use Claude" stage. The problem now is inconsistency. One associate has a brilliant diligence prompt in a notes app. Another improvises. The IC gets two memos built to two standards and can't tell.
If you're new to Claude on deals, start with our practical guide to using Claude in M&A. This article assumes you're past that and want the good workflows to stick.
Skills, prompts and Projects: what's actually different
A prompt is a one-off instruction. It's as good as the person typing it and dies with the chat.
A Project holds background knowledge. Anthropic's Help Center describes Projects as static background knowledge that's always loaded when you chat inside them. Good for a deal: the CIM, the thesis, the model assumptions.
A Skill holds a procedure. It's a folder with a SKILL.md file: a short name and description, then the instructions. Per Anthropic's developer docs, Claude keeps only the name and description in context until a request matches, then reads the rest as needed. So you can install many skills without clogging every conversation.
The practical split: the Project knows this deal. The Skill knows how your firm does this kind of work. Use both together.
Skills by deal stage
Not every task deserves a skill. Anthropic's own guidance is that a skill pays off when Claude does a task repeatedly and you want it done the same way every time. In M&A, that's most of the process.
| Stage | Skill | Inputs | Output | Human check |
|---|---|---|---|---|
| Screening | Thesis-fit screen | Public target information, your thesis, your exclusion criteria | One-page fit assessment with explicit information gaps | Deal lead confirms the gaps before anything moves forward |
| Due diligence | Assumption stress-test | Model assumptions, diligence pack excerpts | Each assumption marked supported, thin or contradicted, with page references | Associate checks every reference against the source |
| IC memo | Sceptical IC reviewer | Draft memo, key findings | The hardest questions per assumption and the evidence each would need | Partner decides which questions the memo must answer |
| Integration | Finding-to-plan translator | Material diligence findings, draft 100-day plan | Where findings contradict the plan's sequencing or timelines | Integration lead owns the changes |
| Retention | Key-role retention brief | Role map, interview notes, assessment outputs | Roles ranked by thesis dependency and early exit signals | HR and the deal lead agree actions; no automated decisions about individuals |
The column that matters most is the last one. A skill standardises the work. It doesn't sign off on it.
Example: a people-risk skill for due diligence
People risk is where the gap between what teams believe and what they check is widest. I've written about why in People Risk in M&A: The Half Your IC Memo Misses, and the method itself is covered in What Is Behavioral Due Diligence in M&A?. I won't repeat either here.
What a skill adds is discipline. A minimal version might look like this:
---
name: people-risk-dd
description: Synthesises management interview notes and assessment outputs into a people-risk section for an IC memo. Use when the user shares interview notes, management assessments or asks about leadership, retention or coordination risk on a deal.
---
# People-risk synthesis
1. Read the deal thesis and list each synergy line.
2. For each synergy, name the roles it depends on.
3. From the interview notes, extract evidence on coordination, decision-making
and response to change. Quote the note and cite its source file.
4. Mark each finding as evidence, inference or gap. Never merge them.
5. Rank retention risk by role, not by name.
6. End every finding with a decision: mitigate, price or accept.
7. Do not draw conclusions about personality or protected characteristics.Two lines do most of the work. Step 4, because a language model writes inference in the same confident tone as fact. And step 6, because an output that doesn't end in a decision is documentation, not diligence.
Interview notes are self-report, shaped by how people present themselves. Behavioural data can be assessed alongside them and may support a more specific read on where coordination is likely to hold. That's the work we do at HumanaQ. The skill just makes sure both sources get read the same way on every deal.
Example: integration skills
A risk flagged in diligence that never reaches the integration plan gets rediscovered the hard way, months later. I made that argument in Everyone Agrees People Drive Integration Outcomes. Nobody Builds For It. Three skills help close the gap:
- Day-1 communications. Your tone, the questions employees actually ask, the commitments you will and won't make. Bundle approved past announcements as reference files so drafts start from your standard.
- Retention-risk flags. Turns the diligence people-risk output and role map into a short watch list for the operating partner, with the behaviours that would signal trouble in the first 30 days.
- 100-day plan tracking. Instead of green or red, it asks whether the people each milestone depends on are actually coordinating. A plan that says what happens without saying who does it isn't a plan. It's a schedule.
For what AI can and can't see after close, see AI in M&A: First 100 Days of Integration.
Guardrails: what never goes into a skill
Skills make good practice repeatable. They make bad practice repeatable too.
Confidentiality. A skill carries method, not deal data. Keep target names, financials and data-room material out of SKILL.md and its files. Supply them per deal, inside the environment your firm has approved.
Retention terms. Anthropic states that Agent Skills is not covered by zero data retention arrangements. If your firm relies on ZDR, confirm what that means for your setup before you build.
Third-party skills. Anthropic advises using skills only from trusted sources, because a malicious skill can direct Claude to run code or move data. Its Help Center says to review even a skill a colleague shares before enabling it.
Checking outputs. Build verification into the skill: page references, evidence/inference/gap labels, and a named human check.
What never goes in a model: anything your NDA, clean-team rules or data-room terms prohibit. Also personal data on individuals beyond what you have a lawful basis to process. And any instruction that produces automated decisions about named people.
How to set up and share skills across a deal team
Where you build a skill decides who can use it. Anthropic documents three surfaces, and custom skills don't sync between the Claude API and claude.ai:
- Claude apps. Skills need code execution turned on; you upload a zipped skill folder under Customize > Skills. On Team and Enterprise plans, the Help Center describes sharing skills with colleagues, publishing them to an organisation library, and owners provisioning them for everyone.
- Claude Code. Skills are folders on disk:
~/.claude/skills/for you,.claude/skills/in a repository for the team, or plugins and managed settings. - Claude API. Uploaded through the Skills API, shared across the workspace, run with the code execution tool and no network access.
Skills follow the Agent Skills open standard, so the format isn't locked to Claude.
A setup that works for most deal teams:
- Start with three skills, not thirty: screening, diligence, IC. Pick tasks where you give Claude the same correction every time.
- Write the description carefully. It decides when the skill fires.
- Test on a closed deal first, with and without the skill.
- Give each skill an owner who updates it after every deal post-mortem.
The firms getting value here don't have the cleverest prompts. Their standards just stopped living in one person's head.
FAQ
What are Claude Skills?
Claude Skills are folders with a SKILL.md file of instructions, plus optional scripts and reference files, that Claude loads only when a task matches. They let a team package a repeatable workflow, such as a diligence synthesis, so it runs the same way every time.
How are Claude Skills different from Projects?
A Project holds background knowledge loaded in every chat inside it, such as one deal's documents. A Skill holds a procedure that activates when relevant, such as how your firm writes a people-risk section. Use a Project per deal and Skills across deals.
Can I use Claude for due diligence on confidential deal documents?
Only within your firm's approved setup and your NDA and data-room terms. Keep deal data out of the skill itself, check Anthropic's current data-retention terms for your plan and surface, and have every output checked by a person against the source documents.
Can deal teams share Claude Skills?
Yes, depending on the surface. Anthropic describes sharing and organisation-wide publishing on Team and Enterprise plans, Claude Code teams can commit skills to a shared repository, and API skills are shared across a workspace.
Which M&A tasks should become skills first?
Tasks you repeat on every deal and correct Claude on every time: screening against your thesis, assumption stress-testing, sceptical IC review, and turning diligence findings into the integration plan.
Keep reading
- How to Use Claude in M&A: A Practical Guide for Deal Professionals
- What Claude Cannot Do in M&A: An Honest Assessment
- How to Prompt AI for M&A: Framing Beats Questions
- What Is Behavioral Due Diligence in M&A?
Join the conversation
For more on how AI is changing the way deals get done, subscribe for free to the Under the Hood of M&A newsletter on Substack.
For the live discussion, join the free AI in M&A group on LinkedIn, where practitioners share the workflows that held up in real deals, and the ones that didn't. I'd like to know which skill your team would build first.
