DEV Community: Daniel Marin

I Don't Have a Design Team. These 4 AI Skills Produce Professional Output Anyway.

Daniel Marin — Sat, 30 May 2026 17:49:07 +0000

Visual art, design systems, brand guidelines, and data visualization. From brief to mockup in minutes, no designer required.

The bottleneck is almost never ideas. It's the gap between "I know what I want this to look like" and "I have a file I can actually use." Founders need decks that don't look amateur. Solopreneurs need landing page mockups for stakeholder reviews. Marketing managers need data charts that don't embarrass anyone. And hiring a designer for every output (a $150-per-hour relationship that involves briefings, revisions, and waiting) is the wrong tool for the volume of visual work a modern one-person operation actually generates.

Most people have tried AI for design work and been disappointed. You describe what you want, you get back something generic: clip art energy with a thin layer of polish. The problem isn't the AI. It's that a general chat interface doesn't know your aesthetic, your brand, or the principles that separate intentional design from content that looks thrown together.

AI design skills (pre-configured instruction sets that tell Claude exactly how to approach visual work) solve the context problem. A skill that knows your aesthetic preferences, your brand rules, and your design philosophy produces outputs that feel considered rather than generated.

Why AI Design Fails Without a Skill

The failure mode is consistent: you ask for a design, describe it as best you can, and get something that technically matches your description but feels wrong. The proportions are off. The color choices are safe rather than intentional. The overall composition looks like someone followed instructions rather than made decisions.

This happens because design is not primarily about description. It's about judgment. When a designer works from a brief, they bring a set of aesthetic principles that the brief doesn't spell out: how much whitespace is generous versus wasteful, when a serif feels classic versus dated, what makes a color palette feel cohesive rather than collected. A general AI prompt carries none of that tacit knowledge. A skill can.

Without a design skill: Re-explain aesthetic preferences every session. Output is technically correct but feels generic. No design principles guiding the judgment calls. Brand inconsistency across outputs.

With a design skill: Design philosophy baked in from the start. Aesthetic preferences guide every decision. Consistent output across sessions and formats. First draft is already intentional, not generic.

1. Canvas Design: Museum-Quality Visual Art From a Concept

Most AI image-adjacent tools produce decoration. Canvas Design produces composition. There's a meaningful difference: decoration fills space, composition communicates.

This skill is built around a two-step process that mirrors how professional designers actually work. Define the philosophy first, then express it visually. The skill starts with a design philosophy document: a written articulation of the concept you're exploring, the aesthetic movement it references, and the principles that should guide every visual decision. Once the philosophy is established, it produces the visual artifact: a PNG or PDF composition that communicates through form, space, and color rather than text and clip art.

"Create a visual design expressing the concept of emergence: the idea that complex behavior arises from simple rules. Start with a design philosophy document, then produce the canvas."

"Design a visual philosophy inspired by Japanese minimalism (wabi-sabi, negative space, impermanence) and express it as a magazine-quality PNG for our product launch."

"Develop a canvas for our SaaS conference keynote backdrop. Concept: precision meets warmth. No text in the image. Output as a 1920x1080 PDF."

The deliberate philosophy-first step is what separates this from image generation prompts. By writing the design thinking before producing the visual, the skill forces a level of intentionality that shows in the output. Viewers may not be able to articulate why it looks considered, but they notice it.

Before: You need a hero image for your landing page. You ask an AI for "a modern, clean design that conveys innovation." You get a blue gradient with abstract geometric shapes. You use it because you have nothing better.

After: You describe the concept behind your product. The skill produces a philosophy document (precision, restraint, the tension between structure and organic growth) then a composition that expresses it. The image has a point of view. People ask who designed it.

Setup: 10 minutes. Best for: brand assets, conference visuals, portfolio work, product launch imagery, mood boards.

2. AI Design Director: A Full Design System That Knows Your Taste

One-off visuals are useful. A design system that produces consistent, on-aesthetic output across every asset you ever create is transformative. The difference between a brand that looks cohesive and one that looks assembled from stock (across landing page, deck, social posts, product UI) is whether there's a real design system behind it or just a vague color scheme someone picked.

This skill builds that system by learning from your references rather than your descriptions. Descriptions of aesthetic are imprecise: "modern but warm," "minimal but not cold." These mean something different to every designer. The skill bypasses the language problem by working from examples you actually like: 30 to 50 screenshots, URLs, or reference images. It extracts the design principles those references share (color tokens, typography scales, spacing patterns, layout logic) and documents them as a design system you own.

"Build my design reference library from these examples: [paste URLs or upload screenshots]. Extract the design principles they share: color palette, typography choices, spacing patterns, layout logic. Document them as a system I can apply to new work."

Once the design system exists, every subsequent request inherits it automatically. You don't describe your aesthetic to get a landing page. You just ask for the landing page. The skill already knows what "your style" means because it extracted it from things you actually chose.

Before: Your brand has a "look" in your head. Every freelancer interprets it differently. The deck looks different from the site, the social posts look different from both. You spend 40% of every design review explaining what's wrong before explaining what you want.

After: The system extracted your design tokens from 40 reference examples you actually love. Every new asset (landing page, icon set, presentation template) is produced from the same tokens. Cohesion is structural, not coincidental.

Setup: 2 hours. Best for: startup founders, solopreneurs with consistent brand needs, creative directors scaling output, indie makers building products.

3. Brand Guidelines Generator: Write Down What Your Brand Actually Is

Most small companies have a brand that lives in the founder's head. The logo is established, maybe a primary color, a vague sense of font preference. Every new hire asks what font to use. Every freelancer gets a different briefing from a different person. The brand drifts with each person who touches it because nothing was written down rigorously enough to constrain it.

This skill turns the implicit into explicit. Feed it what you already have (existing materials, your website copy, logos, sample outputs you approve of) and it produces a complete brand style guide: hex-code color palette, typography rules with specific font names and usage contexts, logo usage specifications, voice and tone guidelines with examples, and explicit do's and don'ts.

"Create brand guidelines for our SaaS company. We're building project management software for architecture firms. Tone: professional but not corporate, precise but not cold. Primary color: #1A2F4B. Logo attached. Include everything a freelance designer would need to work without a briefing call."

Brand guidelines are the unsexy foundational work that makes everything else consistent. Run this skill before onboarding any designer, freelancer, or agency. The two hours you spend producing a real style guide saves dozens of hours in revision cycles.

Setup: 10 minutes. Best for: growing startups, agencies creating client deliverables, marketing teams onboarding freelancers, founders scaling content production.

4. Chart Designer: Make Data Tell a Clear Story

The hardest part of data visualization isn't building the chart. It's deciding which chart to build. Bar versus line versus scatter versus heatmap is a design decision that changes what story the data tells and whether the audience can read it at a glance. Most people default to bar charts for everything, which means some of their data is being communicated poorly every time they present.

This skill handles both decisions: which chart type and how to configure it. Describe your data, your audience, and what story you're trying to tell, and the skill recommends the right visualization with a rationale, then produces the configuration for your charting library of choice: ECharts JSON, Chart.js config, or Excel setup instructions.

"Visualize our monthly revenue by product line for the last 2 years. Audience: board of directors. Story: we want to show the shift in mix toward our higher-margin product. Recommend the best chart type and give me the ECharts config."

"I have cohort retention data: 12 monthly cohorts, retention tracked at 1, 3, 6, and 12 months. I want to show which cohorts retained best and whether we've improved over time. What's the right visualization?"

"Design a dashboard layout for our weekly ops review: revenue trend, pipeline by stage, support ticket volume, and churn rate. Keep it readable for executives who will see it on a 15-inch laptop screen."

Before: You have a spreadsheet with 24 months of revenue data. You build a default bar chart in Excel. The colors are the Excel defaults. The y-axis doesn't start at zero. Three people ask questions in the meeting about what the chart is actually showing.

After: You describe your data and what story it should tell. The skill recommends a small-multiple line chart, explains why, produces the ECharts config with a deliberate color scheme and properly labeled axes. The chart communicates the story in three seconds. No questions.

Setup: 10 minutes. Best for: data analysts, product managers, founders presenting to investors, anyone building dashboards or reports.

How These Four Skills Connect

Each skill solves an independent problem. But used in sequence, they form a complete design operating system for a company without a design team:

Brand Guidelines (once, then update when brand evolves): Write down what your brand actually is before anyone else touches it.
AI Design Director (2-hour setup, then ongoing): Build a design system from references you love. Lock your aesthetic as tokens that every subsequent output inherits.
Canvas Design (per asset or campaign): Produce individual visual assets from concept through composition.
Chart Designer (per report or dashboard): Visualize data for any audience or context with chart type recommendations and export-ready configs.

The setup order matters. Brand Guidelines and AI Design Director are foundational. They establish the rules that Canvas Design and Chart Designer follow. If you run the asset-production skills without a design system behind them, the outputs are good but not yours. With the foundation in place, every output is automatically on-brand because the skill knows what on-brand means.

Practical Tips for Non-Designers

Collect references before you prompt. The clearest signal you can give a design skill is examples of work you actually like, not adjectives. Before running any design session, spend 15 minutes collecting 10 to 20 screenshots or URLs of designs that feel right. The skill can extract principles from examples that words can't fully capture.

Specify the output context, not just the output. "Design a hero image" is a weak prompt. "Design a hero image for a SaaS landing page targeting mid-market CFOs, displayed at 1440px wide, dark background, no people in the image" gives the skill enough context to make every judgment call correctly.

Iterate on the philosophy, not the pixels. When an output isn't quite right, resist the instinct to describe pixel-level changes. Instead, articulate what's wrong at the principle level: "it feels too corporate," "the spacing feels cramped," "the color choices feel random rather than deliberate." Design skills respond better to principled feedback than to specific edits.

The Brand Guidelines output is your master CLAUDE.md. Once you've generated brand guidelines, paste the core sections into the CLAUDE.md file you use for every design session. Any skill that reads it starts with your brand context already loaded. You never re-explain your color palette, typography, or voice rules again.

Getting Started

Design is the one discipline where most founders quietly accept that their output is below the standard they'd want, because hiring the alternative is expensive and slow, and general AI tools produce output that looks generated. These skills close that gap without the agency retainer. The first session is slower than asking an agency. Every session after that is instant, on-brand, and yours.

I publish all four design skills as free, downloadable templates at claudecodehq.com. Each one is a single file you drop into a folder. Start with Brand Guidelines if you've never written yours down, or AI Design Director if you already know your aesthetic and want to formalize it as a system.

Originally published on claudecodehq.com

I Run a One-Person Business With AI Handling Strategy, Sales, Ops, and Finance. Here's the Full Stack.

Daniel Marin — Fri, 29 May 2026 12:41:00 +0000

I Run a One-Person Business With AI Handling Strategy, Sales, Ops, and Finance. Here's the Full Stack.

How solopreneurs use AI skills as a lightweight operating system: strategy, business development, client operations, and finance, without hiring a team.

Running a one-person business means being the strategist, the salesperson, the account manager, the bookkeeper, and the marketer, in addition to being the person who actually does the work clients pay you for. You're expert-level at one thing and expected to be adequate at everything else.

The traditional answer is to hire for your gaps: a VA for admin, an accountant for finances, an agency for marketing. For solopreneurs without the revenue to justify those costs, or who want to stay lean by design, the answer has mostly been: do it yourself and accept that some functions will be underdone.

AI skills change this calculation. A well-built set of skills functions as a lightweight operating system for a one-person business, covering strategy, business development, client operations, and finance with enough depth to actually be useful, not just better than nothing.

The Four Functions Every Solo Business Runs

Every business, regardless of size, runs the same core functions. The difference between a 200-person company and a one-person business isn't the functions. It's who performs them and with what resources.

Strategy and Planning. Business architecture, annual planning, market positioning, growth roadmap. Replaces: business consultant at $200 to $400/hr.

Business Development. Outreach, proposals, pitch materials, pipeline management. Replaces: sales consultant or BD hire at $60 to $100k/yr.

Client Operations. Contracts, onboarding, reporting, relationship management. Replaces: account manager or ops VA at $30 to $60k/yr.

Finance. Budget tracking, invoicing, cash flow management, tax prep support. Replaces: part-time bookkeeper at $500 to $1,500/mo.

The skills don't eliminate the need for judgment. They eliminate the need for starting from scratch every time a function needs exercising. The business plan still requires your strategic thinking. The proposal still requires your knowledge of the client. The invoice still requires your pricing decisions. The skills handle the structure, the format, the first draft, and the analysis layer.

Function 1: Strategy and Planning

Most solopreneurs are reactive rather than strategic. Not by choice, but because carving out time for business-level thinking is hard when client work fills the calendar. The result: you get busy, then you get quiet, then you scramble for new clients, then you get busy again. The feast-and-famine cycle that characterizes most solo practices is almost always a planning problem dressed up as a pipeline problem.

Business Architect: build the plan you never had time to write. This is the highest-leverage skill in the stack for most solopreneurs because it addresses the function that gets most consistently skipped. It runs a structured business architecture session: clarifying your positioning (who you serve and why you specifically), identifying your most profitable service lines, building a 90-day growth plan with specific actions, and mapping the bottlenecks that are limiting revenue.

"I've been running my [type] practice for [X years]. Revenue is [range] and it's been flat for [period]. Run a business architecture session. Diagnose why growth has stalled, identify my most defensible positioning, and give me a 90-day plan with specific weekly actions."

"I want to niche down from [broad service] to [specific niche]. Pressure-test this decision: is the market large enough, what's the competitive landscape, and what's the transition plan that doesn't destroy current revenue?"

The second prompt (niching down) is one of the most common strategic decisions solopreneurs face and one of the hardest to think through alone because of the revenue risk. The skill will model the decision honestly: the upside, the downside, and what would need to be true for it to work.

When you need the formal document (for a bank loan, investor conversation, or grant application), a business plan skill produces the complete structured output: executive summary, market analysis, service offering, competitive positioning, financial projections, and operating plan.

Function 2: Business Development

Solopreneurs who are excellent at their craft often have a difficult relationship with selling. The activities that fill the pipeline (cold outreach, proposals, pitching) feel different from the actual work, require a different mindset, and produce rejection that the actual work rarely does. The result is that business development happens inconsistently: intensely when revenue drops, not at all when revenue is comfortable.

Cold Email: outreach that doesn't feel like outreach. The fastest way to kill a solopreneur's willingness to do outreach is to write cold emails that feel like cold emails: template-obvious, transactional, easy to ignore. A cold email skill solves the personalization problem. Give it information about the prospect and your service, and it writes an email that opens with something genuinely specific to them before making a brief, concrete ask.

"Write a cold email to [prospect role] at [company type]. What I know about them: [research notes]. My service: [describe]. Their most likely pain point: [describe]. Open with something specific to them, not generic flattery. Keep it under 120 words. End with a single low-friction ask (15-min call, not 'let's schedule a meeting')."

The "under 120 words" constraint is worth keeping. Longer cold emails have lower reply rates almost universally. The skill will push back if you give it too much to say, a useful constraint for solopreneurs who tend toward over-explaining.

Once a prospect responds, a proposal skill builds the full pitch document: problem statement, proposed approach, deliverables, timeline, and investment, in a format calibrated to your service type.

Function 3: Client Operations

The operational layer of a solo practice (contracts, onboarding, reporting, renewals) is where most solopreneurs are inconsistent. Contracts get sent late or not at all. Onboarding is improvised every time. Client reports are either too detailed (take too long to write) or too sparse (don't convey enough value).

Contract Review: protect yourself before you sign. Solopreneurs often sign client contracts with minimal review, either because they don't have legal counsel, because they don't want to seem difficult, or because the document looks standard. A contract review skill reads any agreement and surfaces what matters: liability clauses that are unusually one-sided, IP assignment language that could claim ownership of work you'd expect to reuse, payment terms with hidden gotchas, and auto-renewal or termination provisions that could create problems later.

"Review this client contract before I sign it. I'm a freelance [role]. Flag: any IP assignment that claims more than this specific project, payment terms that deviate from net-30, liability caps that are asymmetric, and anything that would prevent me from working with similar clients in the future."

Most solopreneurs who use this skill discover that "standard" contracts aren't standard at all. The IP and non-compete clauses in particular vary widely and the defaults often favor the client.

Client Report: demonstrate value without spending hours writing. Regular client reports are one of the highest-leverage retention tools available. They make invisible work visible, justify the engagement, and remind the client why they hired you. Most solopreneurs know this and still don't do them consistently, because writing a good report takes two hours they don't have.

A client report skill produces a polished, structured report from bullet points: what was done this period, key results with metrics, what's coming next, and any decisions or approvals needed. Input ten minutes of notes. Get a professional report that reads like it took an hour to write.

"Write a monthly client report for [Client]. This month: [bullet points of work done and results]. Next month: [planned activities]. Client is [describe their communication style]. Keep it professional but not stiff."

Function 4: Finance

Solo business finance is genuinely simple by enterprise standards, but it's the function most solopreneurs manage worst, because it's the one with the least immediate feedback. You can ignore your finances for months with no immediate consequence, and then hit a tax bill or a cash flow gap that required months of preparation to avoid.

Budget Analyzer: understand your actual financial picture. Most solopreneurs have a vague sense of their finances. They know roughly what they earn and roughly what they spend, but the actual picture (profitability by client type, effective hourly rate, subscription creep, tax liability tracking) stays blurry.

"Analyze my last 3 months of business transactions: [paste CSV or data]. Show revenue by client, expenses by category, net profit, and flag every recurring charge so I can decide what to keep."

"I earned [amount] from [client types] this quarter. My fixed costs are [list]. What's my effective hourly rate by client type, and which clients are most profitable per hour invested?"

The effective hourly rate by client prompt is one most solopreneurs have never run, and the answer is often surprising. The client who seems most valuable by absolute revenue frequently has the worst effective rate once you account for revision cycles, communication overhead, and scope creep.

Pair the budget analyzer with an invoice generator for the invoicing side: produce professional, legally complete invoices from project notes in minutes, without paying for invoicing software.

The Solopreneur's Master CLAUDE.md

All of these skills become significantly more useful when they share a common context file:

# My Business Context

## What I Do
[Service description: specific, not generic. Who you serve, what outcome
you deliver, how you're different from alternatives]

## Current Clients
[Active client list with one-line context on each relationship]

## Revenue & Goals
[Current monthly/annual revenue, target, timeline]

## Rates & Pricing
[Your standard rates, how you price projects, what you won't do below]

## Voice & Communication Style
[How you write to clients: formal/warm, concise/detailed, phrases you use or avoid]

## Current Constraints
[Capacity situation, tools you use, things you can't change right now]

## This Quarter's Focus
[The 1-2 things that matter most for business growth in the next 90 days]

Every skill reads this file before producing output. The business architect knows your revenue goal before it suggests growth strategies. The cold email skill knows your positioning before it writes outreach. The budget analyzer knows your target rate before it calculates profitability. The context makes every output immediately applicable rather than requiring editing to fit your situation.

Keep this file updated monthly. Ten minutes to refresh the current clients list, revenue numbers, and quarterly focus.

The Complete Solo Business Cycle

Quarterly strategy: Business architect reviews positioning, identifies highest-leverage actions, builds 90-day plan.
Business development: Cold email writes personalized outreach to 10 to 20 ideal prospects per month.
Proposal: Convert interested prospects to clients with a structured, professional proposal.
Contract: Review client contract before signing. Flag IP, liability, and payment terms.
Client delivery: Monthly client report from bullet-point notes. Demonstrate value, prompt renewals.
Financial review: Monthly financial picture. Revenue by client, expenses, effective rate, projections.
Invoicing: Professional invoices from project notes. Consistent, fast, no software subscription.

Getting Started

The best place to start: build your master CLAUDE.md this week (20 minutes), then use the business architect skill for a quarterly review. The strategic clarity you get from one serious session is usually worth more than any individual automation. Build the operating system from the top down. Strategy first, execution layer second.

I publish all of these skills as free, downloadable templates at claudecodehq.com. Each one is a single file you drop into a folder. No coding, no subscription, no team required. Pick the function that's most underdone in your business right now and set up that skill first.

Originally published on claudecodehq.com

I Built a Personal LLM Wiki. Now Every AI Session Starts Like It Already Knows Me.

Daniel Marin — Thu, 28 May 2026 20:06:15 +0000

What an LLM Wiki is, why it's different from Obsidian or Notion, and how to build one so your AI assistant always has the context it needs to give you useful answers.

Most people who use AI heavily notice the same failure pattern: they get a great answer in one session, close the tab, and have to re-explain the same context next time. The AI is stateless between sessions. Every conversation starts cold. The output quality is directly proportional to how well you brief it, and briefing from scratch every time is exhausting.

The fix that's emerging in technically curious circles is the personal LLM Wiki: a structured, AI-readable knowledge base about you, your work, and your domain that an AI can load as context at the start of a session. Not a notes app. Not a second brain. Something specifically designed to make an LLM immediately useful without a briefing.

This post explains what an LLM Wiki is, why it's different from the knowledge management tools you already use, and how to build one, including how to populate it from the knowledge you've already accumulated in notes, conversations, and documents.

What Is an LLM Wiki?

A traditional wiki is a collection of hyperlinked pages designed for humans to navigate. You browse a hierarchy, follow links, search for terms. The structure serves the human reader's way of moving through information.

An LLM Wiki is organized around a different reader: the AI. It's a structured document (or set of documents) written to give an LLM the context it needs to reason about your situation, answer questions about your domain, and make decisions consistent with your values and constraints. The human rarely reads it directly. You maintain it and the AI uses it.

Traditional notes / wiki:

Organized for human navigation (hierarchy, links)
Written in personal shorthand or stream-of-consciousness
Comprehensive: captures everything
Retrieval: you search for it
Value: a record of what you thought

LLM Wiki:

Organized for AI consumption (structured sections)
Written explicitly, no implied context
Curated: contains what the AI needs to know
Retrieval: AI reads it automatically
Value: makes every AI session immediately useful

The practical implication: your Obsidian vault or Notion workspace is probably not already an LLM Wiki, even if it's comprehensive and well-organized. Notes written for yourself are full of implied context that an AI can't infer: shorthand, references to things you understand but never wrote down, emotional subtext, and gaps that would be obvious to you but aren't on the page.

An LLM Wiki is explicit by design. It assumes the reader has never met you.

How It Differs From a Second Brain

The "second brain" concept (popularized by Tiago Forte and implemented in tools like Obsidian, Notion, and Roam) is about externalizing your thinking so you can retrieve and recombine it later. The goal is your future self: building a system your future self can search through and find useful.

An LLM Wiki has a different goal: making the AI immediately useful in the current session. It's not about archiving your thinking. It's about curating the context an AI needs to reason well on your behalf.

Second brain: comprehensive capture. You want to capture everything: ideas, quotes, meeting notes, random connections. More is better. The system is a long-term archive you query over years.

LLM Wiki: curated context. You want the AI to know the things that make the most difference to its output quality: your background, your current priorities, your domain knowledge, your constraints and preferences. Less is more. A focused, well-maintained wiki outperforms an enormous unstructured dump of notes.

The best setup for heavy AI users: a second brain (notes system) for comprehensive capture, and an LLM Wiki for curated AI context. They serve different purposes and work best together.

What Goes In an LLM Wiki

The contents depend on how you use AI. For most people, the high-value sections cluster into five categories:

1. Who you are. Background, current role, domain expertise, career arc. Not a CV. The specific things that change how an AI should frame its answers to you. A cardiologist and a graphic designer both asking "explain oxidative stress" want different depth and vocabulary. The wiki tells the AI which one you are.

2. Current context and priorities. What you're working on right now, what matters this quarter, open decisions, active projects. This is the section that needs the most frequent updating (monthly or when major context changes) and the section with the highest immediate impact on output quality.

3. Domain knowledge and opinions. Your considered views on topics in your field. Not Wikipedia-level facts, but your specific perspective. "I think the consensus on X is wrong because Y." "My framework for evaluating Z is..." This makes AI responses align with your actual thinking rather than the generic consensus.

4. Preferences and constraints. How you like to communicate, what format you want responses in, what you consider a good answer vs. a great one. Practical constraints: tools you use, things you can't change, resources you have or don't have. Eliminates suggestions that are technically correct but don't apply to your situation.

5. Reference materials for specific tasks. Definitions that matter in your field, templates you use repeatedly, standards you work to, terminology that has specific meaning in your context. The things you'd have to explain if a new colleague started tomorrow.

Building It: The Guided Interview Approach

Rather than staring at a blank document and wondering what to include, the best approach is a guided interview that extracts the right information in the right structure:

"Help me build my personal LLM Wiki from scratch. Interview me section by section: background and expertise, current work and priorities, domain knowledge and opinions, preferences and constraints, and key reference material. Ask follow-up questions to make my answers specific. Push back on vague answers. After each section, show me the wiki text you're building so I can confirm it's accurate."

The "push back on vague answers" instruction matters. The most common mistake in building an LLM Wiki is writing in the same shorthand you use in personal notes. "I work on AI stuff" is useless context. "I'm a machine learning engineer at a Series B startup building recommendation systems for e-commerce, with five years of production ML experience" is the context that changes output quality.

Budget 45 to 60 minutes for the initial build. The output is a Markdown file (your LLM Wiki) that you put in every project folder alongside your skill files.

What using it looks like afterward:

Before wiki: "I'm a [role] at a [company] and I'm trying to [explain context for 5 minutes]. What's your recommendation?"

After wiki: "What's your recommendation?" Claude has already read the wiki. It knows who you are, what you're working on, and your constraints. The answer is immediately calibrated.

Populating From Knowledge You Already Have

Most people with years of experience have the raw material for a great LLM Wiki scattered across notes apps, documents, and past conversations. The gap is synthesis: extracting the structured, explicit knowledge from the unstructured accumulation.

From your notes. If you have an Obsidian vault, Notion workspace, or folder of markdown notes, AI can read through them and extract the content most relevant to your wiki: your recurring opinions on domain topics, the frameworks you use repeatedly, the decisions and their reasoning, the terminology that has specific meaning in your context.

"Read through my notes from the last 12 months. Extract: my recurring opinions and frameworks on [domain], terminology I use with specific meaning, patterns in how I think about [topic], and decisions I made with reasoning that might inform future decisions. Format the output as sections I can paste into my LLM Wiki."

From your AI conversations. Heavy AI users have a goldmine of knowledge buried in their conversation history: problems worked through, decisions made, frameworks developed through back-and-forth. AI can extract structured intelligence from conversation exports: the decisions and their reasoning, recurring themes across sessions, open questions, and the patterns in how you think and work.

"Here are my exported AI conversations from the last 6 months. Extract the content that belongs in my LLM Wiki: what do I keep coming back to, what frameworks do I seem to use repeatedly, what opinions have I developed and stated explicitly, what constraints and preferences come up across sessions? Format as wiki sections."

Run both approaches once when you first build your wiki. They bootstrap a knowledge base from material you've already produced without requiring you to reconstruct it from memory.

What a Well-Built LLM Wiki Looks Like

Here's the structure, annotated so you understand why each section is there:

# Personal LLM Wiki — [Your Name]
Last updated: [Date]

## Identity & Background
[Role, domain, years of experience, relevant credentials.
Written explicitly: assumes the reader knows nothing about you.]

## Current Context
### Active Projects
[What you're working on right now with enough detail to matter]

### Open Decisions
[Decisions you're actively wrestling with]

### This Quarter's Priorities
[What success looks like in the next 90 days]

## Domain Knowledge & Opinions
### [Topic Area 1]
[Your actual views, not the consensus. What you think and why.]

### [Topic Area 2]
[Frameworks you use, approaches you favor, things you've found don't work]

## Preferences & Constraints
### Communication Style
[How you want responses: length, format, tone, what to avoid]

### Tools & Environment
[What you use, what you don't, what you can't change]

### Decision-Making Style
[How you like to evaluate options, what you over/underweight]

## Reference Material
### Terminology
[Terms that mean specific things in your context]

### Templates & Formats
[Structures you use repeatedly]

### Standards & Criteria
[What a good X looks like in your world]

The wiki lives as a Markdown file. It goes in the same folder as your project or task, alongside your CLAUDE.md skill file. Claude reads both: the wiki provides who you are, the skill file provides what Claude should do. Together they produce output calibrated to both the task and the person.

Maintaining Your LLM Wiki

The most common failure mode: building a great wiki and never updating it. Within three months it's stale. Your priorities have shifted, you've developed new views, the projects in your current context section are done. Stale context is worse than no context in some ways, because it produces confidently wrong outputs.

Monthly. Update the Current Context section. Replace completed projects with new ones, update open decisions, refresh this quarter's priorities. Takes 10 to 15 minutes.

Quarterly. Review the Domain Knowledge section. Have any of your views evolved? Are there new frameworks you've developed? Add what's new, remove what's no longer accurate.

Ad hoc. When you have an insight, develop a new framework, or have a conversation that produces something worth keeping, add it to the relevant section.

Annually. Full review. Rewrite sections that have grown stale, consolidate overlapping entries, remove things that no longer apply.

What Changes When You Have One

The difference isn't subtle. With a well-maintained LLM Wiki, questions that previously required a paragraph of setup become one-liners. Recommendations stop being generic and start being specific to your situation. Suggestions that don't apply to you (wrong tool, wrong scale, wrong context) stop appearing because the wiki has already communicated your actual constraints.

The deeper shift: AI stops being a tool you use for isolated tasks and starts functioning more like a well-briefed collaborator. One that knows your background, understands your priorities, and doesn't need to be re-introduced every session. The briefing overhead disappears. The context compounds.

That's what distinguishes heavy AI users who get dramatically better results from those who plateau at "it's useful sometimes." The difference is almost never model quality. It's almost always context quality.

Getting Started

The LLM Wiki is probably the highest-leverage single thing you can do to improve AI output quality. Not by changing the model, but by changing the quality of context it works from. Build it once, maintain it in 15 minutes a month, and the compounding benefits show up in every session afterward.

I publish free playbooks for building and maintaining your LLM Wiki at claudecodehq.com: the guided wiki builder, a second brain synthesis tool for extracting wiki content from your existing notes, and a conversation history mapper for pulling structured knowledge from past AI sessions. Each one is a single file you drop into a folder and start using immediately.

Originally published on claudecodehq.com

I Audited My Week and Found 7 Hours of Wasted Time. Here's the Checklist I Used.

Daniel Marin — Tue, 26 May 2026 17:44:00 +0000

A domain-by-domain audit for finding the 5+ hours a week hiding in your repetitive tasks, and the AI skills to automate each one.

Most people don't lose time in big obvious chunks. They lose it in five-minute increments, twenty times a day: searching for a file, writing the same kind of email for the fourth time this week, transferring meeting notes into a format someone can actually use, scheduling a call that required six back-and-forth messages to land.

Those increments add up. Research consistently shows knowledge workers spend 20 to 30% of their time on tasks that could be automated or dramatically compressed. Administrative overhead that feels like work but doesn't actually produce anything. For most people, that's 8 to 12 hours a week. Finding 5 of those hours isn't ambitious. It's conservative.

The reason most people don't find them: they don't audit. They know abstractly that they're spending time on repetitive things, but they don't look systematically at where those things are concentrated or what it would take to fix them.

This post is that audit.

How to Run the Audit

The audit has two modes. You can run it yourself using the domain checklists below (work through each section, check the items that apply to you, and total up the estimated hours). Or you can run it with AI using a guided audit skill, which walks you through a structured conversation about your daily and weekly routines and produces a prioritized automation plan tailored to your specific situation.

Either way works. The checklist version is faster and produces a rough estimate. The AI-guided version is more thorough. It asks follow-up questions, surfaces automations you wouldn't have thought of, and produces a ranked list of where to start based on effort vs. time recovered.

If using the AI-guided audit:

"I want to run a full life automation audit. Walk me through my typical day and week. Ask me about each area one at a time. For each thing I describe, flag whether it's automatable, partially automatable, or genuinely requires my judgment. At the end, give me a ranked list of the top 5 automations by estimated time recovered vs. effort to set up."

Before starting either version, think about the last two weeks. Not your ideal week. Your actual week. The audit finds real friction, not imagined inefficiency.

Domain 1: Email and Communications

The highest-leverage domain for most people. Email isn't just time-consuming. It's cognitively expensive. Each context switch to process a message costs attention that compounds across the day.

Check if these apply to you:

I spend more than 45 minutes/day in my inbox (3 to 4 hrs/wk)
I write similar replies multiple times a week: status updates, follow-ups, intro emails (1 to 2 hrs/wk)
I manually sort or file emails rather than having a system (1 hr/wk)
I read emails in the moment rather than in batched processing windows (1 to 2 hrs/wk)
I spend time deciding which emails actually need a response from me vs. FYI (30 to 60 min/wk)

Typical recoverable time: 2 to 5 hours/week.

An email triage skill categorizes, prioritizes, and drafts replies for a full inbox in minutes. A full inbox processing workflow handles actions, delegates, defers, and done items.

Domain 2: Calendar and Scheduling

Scheduling friction is insidious because no single instance feels significant, but the aggregate is real. Back-and-forth scheduling emails, manual calendar management, and unplanned context-switching between tasks accumulate quietly.

Check if these apply:

I send 3+ emails to schedule a single meeting (30 to 60 min/wk)
My calendar doesn't reflect my actual priorities: meetings crowd out focused work (1 to 2 hrs/wk lost focus)
I don't have protected time blocks for my highest-leverage work (1 hr/wk lost)
I manually plan my week rather than having a structured weekly planning process (30 to 45 min/wk)
I frequently context-switch mid-task because meetings aren't clustered efficiently (1 to 2 hrs/wk in recovery time)

Typical recoverable time: 1.5 to 3 hours/week.

A calendar automation skill restructures your week around your actual priorities: protected deep work, clustered meetings, and a realistic weekly plan.

Domain 3: Files, Folders, and Information Chaos

Information retrieval is an invisible time drain. You don't think of "searching for a file" as a task. But if you do it ten times a day for 3 to 5 minutes each, that's 30 to 50 minutes gone.

Check if these apply:

I spend time searching for files I know I have but can't find quickly (30 to 60 min/wk)
My Downloads folder is chaotic: screenshots, installers, documents all mixed (15 to 30 min/wk in friction)
I maintain duplicate files across multiple folders or devices (15 to 30 min/wk)
I have no consistent naming or organization system for project files (30 to 60 min/wk)
I scroll through long documents to find specific sections I need (20 to 30 min/wk)

Typical recoverable time: 1 to 2 hours/week.

A file organizer skill sorts thousands of files by type, archives old ones, and flags duplicates in one session.

Domain 4: Meeting Overhead

The meeting itself is often the smaller time cost. The overhead (prep, note-taking, processing, follow-up, tracking action items) can equal or exceed the meeting duration. For someone in 5 to 8 meetings a week, this adds up fast.

Check if these apply:

I don't have a consistent pre-meeting prep process (15 to 30 min/wk in wasted meeting time)
I take raw notes during meetings and never process them into structured actions (30 to 60 min/wk in lost accountability)
Action items from meetings get lost: no system to track commitments (30 to 60 min/wk in rework and follow-up)
I write meeting follow-up emails manually rather than using a consistent format (20 to 40 min/wk)
I attend meetings I don't need to be in but haven't developed a polite exit system (30 to 90 min/wk)

Typical recoverable time: 1.5 to 3 hours/week.

A meeting notes skill pastes raw notes and returns decisions, owners, deadlines, and a follow-up email in under a minute.

Domain 5: Daily Startup and Weekly Admin

The first 30 to 60 minutes of each day are disproportionately important. They set the frame for everything that follows. Most people spend that window reactively (inbox, Slack, whatever appeared overnight) rather than from a considered starting position.

Check if these apply:

My morning starts reactively: inbox and messages before I've identified my top priority (1 to 2 hrs/wk in misallocated morning time)
I don't have a consistent weekly planning ritual: the week just happens (1 to 2 hrs/wk in lost direction)
I track to-dos across 3+ places: notes app, email flags, sticky notes, memory (30 to 60 min/wk in mental overhead)
I don't do a weekly review: things slip, commitments get forgotten (30 to 60 min/wk in rework)
Administrative tasks (expenses, reports, status updates) pile up and require catch-up (1 hr/wk)

Typical recoverable time: 1.5 to 3 hours/week.

A morning brief skill starts every day with a 5-minute structured brief: priorities, pending decisions, and one highest-leverage task.

Tallying Your Results

If you worked through the checklists honestly, you likely found more than five hours. Most people do. The question is which ones to fix first.

High first: Items where the automation is simple to implement and the time recovered is high. Email triage and file organization typically live here. One-time setup, immediate daily payback.

Second: Items where the time recovered is significant but setup requires a few more minutes. Calendar restructuring, meeting processing habits, weekly review ritual.

Later: Items that require behavior change rather than just a tool. Stopping reactive mornings, declining unnecessary meetings. High value, but not a quick win. Work on these after the easier automations are running and you have momentum.

A practical rule: implement one automation per week for the next four weeks. By week five you have four working automations compounding, and the time you recovered from the first ones funds the energy to set up the next ones.

After the Audit: Three-Step Implementation

1. Pick your highest-leverage automation. The one where the checklist indicated the most time and the setup is under ten minutes. Email triage is the most common first choice. It produces visible results from the first session and compounds daily.

2. Set it up and use it for two weeks before adding another. The temptation is to set up everything at once. Resist it. One automation that becomes a habit is worth more than four that get used twice and forgotten. Two weeks is enough time for a new workflow to become automatic.

3. Re-run the audit in 90 days. Your workflow will have shifted. New friction will have appeared. Some things you thought were problems will have resolved or turned out not to matter. The audit is a recurring practice, not a one-time fix.

Once a Year: The Bigger Picture

The five-domain audit above focuses on tactical time: the recurring friction in your daily and weekly workflow. Once a year, it's worth running a more fundamental review. Not just "how do I do what I do more efficiently" but "am I spending my time on the right things at all."

The distinction: the weekly automation audit recovers hours. The annual life audit redirects them toward what they should have been going to in the first place. Both matter, but in sequence. Fix the efficiency layer first, then direct the recovered capacity toward something worth having.

Getting Started

The five hours are there. They're in the same places for almost everyone: email, scheduling, files, meeting overhead, and reactive mornings. The audit takes 20 minutes. The first automation takes 10 minutes to set up. By the end of the week, you've already started getting them back.

I publish free playbooks for every automation mentioned in this audit at claudecodehq.com: the guided life automation audit, calendar restructuring, file organization, meeting processing, morning briefings, and more. Each one is a single file you drop into a folder. No coding, no subscription. Start with the one that matches your biggest time drain.

Originally published on claudecodehq.com

I Built an AI Chief of Staff. It Briefs Me Every Morning and Challenges My Assumptions Before I Commit.

Daniel Marin — Mon, 25 May 2026 16:29:39 +0000

How to use AI skills to get daily briefings, weekly accountability, priority triage, and a strategic sounding board that has no agenda. Setup takes 30 minutes.

Your inbox lives in Gmail. Your calendar is in Google Calendar. Your tasks are split between a notes app and a tool your team uses that you check reluctantly. Slack has notifications you're ignoring. Your actual priorities, the ones that determine whether this quarter succeeds, live exclusively in your head.

No tool currently has the full picture. And the cognitive overhead of maintaining it mentally, session after session, meeting after meeting, compounds into a kind of strategic fog: you know what matters, but you spend most of your time on what's in front of you rather than what's most important.

A real chief of staff solves this. They hold context across everything: your priorities, your key decisions, your commitments, the things you said you'd follow up on three weeks ago. They brief you before you walk into the room. They ask the question that challenges your assumption before you make the call. They don't let important things disappear into the operational noise.

AI skills can replicate this function. Not the relationship, not the presence, not the judgment that comes from a decade of working alongside you. But the context-holding, the briefing, the decision support, and the accountability structure. Here's how to build one.

Why This Is Different From Asking AI for Help

The failure mode of executives who try to use AI and conclude it's not useful for them: they open a chat window, explain their situation from scratch, get a generic response, and close the tab. The problem isn't the AI. It's that a cold session with no context produces cold output.

A skill works differently. The CLAUDE.md file holds your standing context permanently: your priorities for the quarter, your decision-making style, your key stakeholders, your constraints, your current open questions. Every session starts with Claude already knowing what you're working on and what matters to you. You don't brief it. It briefs you.

Generic AI chat: "I'm a CEO of a 60-person B2B SaaS company. We have a board meeting next week and I'm trying to decide whether to raise a bridge round or push for profitability. Here's the context: [ten minutes of explaining]..."

With an AI Chief of Staff skill: "Board meeting is Thursday. Help me think through the bridge vs. profitability decision." Claude already knows: company stage, burn rate, board dynamics, your stated Q3 priorities, and that you've been leaning toward profitability for two months. The answer is immediately useful.

The setup takes 30 to 45 minutes once. After that, every session starts from a complete picture.

Skill 1: Executive Chief of Staff (The Daily Operating Layer)

This is the skill you open every morning and every time you need to think through something quickly. It knows your role, your company context, your team, your current priorities, and your preferred communication style. It functions as a persistent operating layer, not a one-off assistant.

Morning briefing. Start the day with a structured brief: what's on today's calendar, what decisions are pending, what you said you'd follow up on, and what the one highest-leverage thing is to accomplish before anything else. Takes two minutes. Replaces the fifteen minutes of context-assembly that used to happen in your head while reading emails.

Pre-meeting prep. Before any significant meeting, ask for a quick brief: the person's background, the relevant context from previous interactions, what you want to get out of it, and what you should be careful about.

Priority triage. When the inbox is full and everything appears urgent, paste in the list of demands on your attention and ask for a prioritized view: what genuinely moves the needle vs. what feels urgent but isn't, what you should do, delegate, defer, or decline.

Communication drafting. Board updates, all-hands memos, investor emails, difficult conversations with direct reports. The skill knows your voice and your context. It drafts in your register, not a generic corporate register.

"Brief me for today. What's on my calendar, what decisions have I been deferring, and what's the one thing I should make sure happens today?"

"I have 12 things in my head that need attention this week. Here they are: [list]. Prioritize them and tell me what the top three actually are, and why."

"Draft my Friday all-hands update. Key news this week: [bullets]. Tone: direct and honest, no corporate softening. Under 300 words."

Skill 2: Chief of Staff Check-In (The Weekly Accountability Rhythm)

The daily skill handles operations. This skill handles the weekly strategic rhythm: the structured review that most executives know they should do and rarely do consistently.

A real chief of staff runs this meeting with you. They bring the list of commitments you made last week, hold you accountable to what didn't get done, surface the pattern of what keeps slipping, and ask the uncomfortable question about whether the current week's plan is actually aligned with the quarter's priorities.

The skill replicates this structure: a weekly session that reviews what you committed to, what actually happened, what's blocking progress on what matters most, and what the adjusted plan is for the coming week. It asks the hard questions. It doesn't let you reframe a missed commitment as a strategic pivot without justification.

"Weekly check-in. Last week's commitments: [list]. What happened: [list]. Run the review. What got done, what slipped, what's the honest explanation, and what am I going to stop saying I'll do if I keep not doing it?"

"I've now deferred [specific initiative] three weeks in a row. What are the possible explanations for that, and what should I actually do about it?"

The second prompt (asking directly about a pattern of deferral) is where this skill produces its highest-value output. Most executives have a persistent item they keep moving. The skill will name the possible explanations honestly: it's not actually a priority, you're avoiding a difficult conversation it requires, you don't have enough information to decide, or it's someone else's job and you haven't delegated it clearly. That kind of direct challenge is hard to get from most people in your organization.

Skill 3: CEO Advisor (A Strategic Sounding Board Without an Agenda)

The loneliness-at-the-top problem is real and structural. Everyone in your organization has a stake in your decisions. Your board has its own interests. Your investors are aligned in some ways and misaligned in others. The person who can tell you you're wrong, without any skin in the game, is genuinely rare.

This skill fills a specific gap: a sounding board that knows your company deeply, has no stake in your decisions, and is designed to challenge your assumptions before you commit. It doesn't tell you what you want to hear. It applies structured strategic frameworks (pre-mortem analysis, second-order thinking, devil's advocate) and pushes back on reasoning that isn't solid.

"I'm leaning toward [strategic decision]. Challenge my reasoning. What am I not seeing, what assumptions am I making that could be wrong, and what does the bear case look like if I'm wrong about the key variable?"

"Run a pre-mortem on this decision. Assume it's two years from now and this choice turned out to be a significant mistake. What are the three most likely explanations for why it went wrong?"

"I'm about to have a difficult conversation with [person/board/investor]. Steel-man their position for me. What's the strongest version of the argument I'm going to hear, and what are the points I genuinely can't counter?"

The pre-mortem and steel-man prompts are the highest-leverage uses. They produce the kind of thinking that usually requires an experienced board member or advisor who knows your situation well enough to push on the specific weak points, not generic strategic advice.

Building Your Executive CLAUDE.md

The quality of your AI Chief of Staff is directly proportional to the quality of the context you give it. Here's the template:

# Executive Context

## Role & Company
[Your title, company, stage, industry, headcount, key metrics]

## Current Quarter Priorities (3 max)
1. [Priority: what success looks like in one sentence]
2. [Priority: what success looks like in one sentence]
3. [Priority: what success looks like in one sentence]

## Key Decisions Currently Open
- [Decision: what it is, what I'm leaning toward, what's blocking resolution]

## Key People & Relationships
[Direct reports, board members, key investors, critical external relationships:
one line each on context that matters]

## Communication Style
[How I write and speak: direct/warm/formal/conversational, anything I always/never do]

## My Decision-Making Tendencies (honest version)
[What I tend to over-weight, what I tend to under-weight, biases I'm aware of]

## Current Constraints
[Budget, headcount freeze, board-level mandates, anything I'm working around]

## Last Updated
[Date: review monthly or when major context changes]

Two sections deserve particular attention. Key Decisions Currently Open is what most executives forget to include and what produces the most immediate value. When your CoS skill already knows what you're wrestling with, every relevant conversation and piece of information gets filtered through that lens automatically.

My Decision-Making Tendencies is uncomfortable to write but disproportionately valuable. If you know you tend to over-weight recency bias in people decisions, or that you historically under-invest in operational details when you're excited about strategy, encoding that gives your advisor skill the specific angles to challenge you on. A generic AI will give you generic pushback. One that knows your specific failure modes will ask the right questions.

A Practical Daily and Weekly Routine

Daily (10 minutes):

Morning: open the Chief of Staff skill. Get the day's brief: priorities, pending decisions, calendar context, one highest-leverage task.
Before any high-stakes meeting: ask for a quick context brief.
When inbox overflows: paste threads in for triage and draft replies.
After any significant meeting: paste notes for structured decisions, owners, next actions, and a follow-up draft.
Before major decisions: open the CEO Advisor. Run a pre-mortem, steel-man the opposition, and have your assumptions challenged before you commit.

Weekly (30 minutes):

Open the Chief of Staff Check-In. Review last week's commitments, hold yourself accountable on what slipped, set the top three for the coming week.

What an AI Chief of Staff Doesn't Replace

Being precise about the limits matters.

Real relationships and trust. A human CoS builds relationships with your team, reads the room in a board meeting, and earns political capital on your behalf. The AI skill does none of this. It's a cognitive tool, not an organizational actor.

Real-time information. Claude Code runs locally and reads files you provide. It doesn't monitor your inbox, listen to your meetings, or pull live data. You bring information to it. It doesn't gather it autonomously.

Execution and follow-through. The AI Chief of Staff advises, drafts, and organizes. It doesn't send emails, book meetings, or ensure your direct reports actually do what was decided. Execution still requires humans.

Within those limits, the AI Chief of Staff is most powerful for executives who have the judgment and capacity to act on good input, but who are currently constrained by the friction of context-assembly and the scarcity of advisors who can give them genuinely independent pushback. That's most senior leaders, most of the time.

Getting Started

Build your CLAUDE.md this week. Thirty minutes, no coding, just writing down what's already in your head. Use the Executive Chief of Staff skill for one morning brief. If it produces something useful from your first prompt, you'll know within five minutes whether this is worth building out further. Most executives who try it do.

I publish all three core skills (plus supporting skills for inbox triage, meeting intelligence, decision matrices, and executive dashboards) as free, downloadable templates at claudecodehq.com. Each one is a single file you drop into a folder. No coding, no subscription, no IT ticket.

Originally published on claudecodehq.com

The AI Toolkit Our PE Fund Uses Across the Entire Deal Cycle

Daniel Marin — Sat, 23 May 2026 16:42:09 +0000

Deal sourcing, screening, CIM production, IC memos, returns analysis, and portfolio monitoring. Built for practitioners who are tired of document production eating their analyst hours.

The private equity and investment banking workflows that eat the most analyst time are almost always document production and data synthesis, not judgment. Screening a deal against fund criteria, drafting a CIM from a data room, building an IC memo structure from diligence notes, monitoring a portfolio across twelve companies. The judgment calls require senior expertise. The production work mostly requires endurance.

AI skills flip that ratio. Because a skill holds your firm's criteria, templates, and context permanently, every document starts from your framework rather than a blank page. The analyst writes the sections that require genuine synthesis. The skill handles the structure, the formatting, and the initial pass that used to take half a day.

This guide covers the full toolkit, organized by deal phase, with separate tracks for IB and PE workflows.

Why AI Skills Beat Financial Templates

Most finance teams have templates: pitch deck shells, CIM structures, IC memo formats inherited from whoever set them up years ago. Templates solve the blank-page problem but create a different one. They produce outputs that are structurally correct but substantively thin. Analysts fill in the boxes without necessarily building the analysis the boxes are supposed to represent.

An AI skill is different from a template in three ways that matter for deal work:

It knows your criteria. A PE deal screening skill that holds your fund's actual investment criteria (sector focus, revenue thresholds, EBITDA minimums, geographic restrictions, deal size range) flags mismatches immediately rather than after an analyst spends four hours building a model on an ineligible target.

It synthesizes, not just formats. Given a company's data room documents, a CIM-building skill doesn't just populate a structure. It extracts and synthesizes the relevant information, flags gaps that need clarification, and surfaces inconsistencies between the management presentation and the financial statements.

It improves with your feedback. A template is static. A CLAUDE.md skill evolves. When a senior banker or partner says "we always lead with market size before business model," you add that instruction once and it applies to every subsequent output. Institutional knowledge becomes executable, not just tribal.

Investment Banking Track

Pitch Deck: Win the Mandate Before the Process Starts

A pitch deck for a sell-side mandate needs to accomplish three things quickly: demonstrate you understand the business better than the company thinks you do, establish a credible valuation range that passes the smell test, and show a process architecture that gives sellers confidence they'll get the best outcome. Most pitches fail on the first point. The analysis is generic enough that it could apply to any company in the sector.

An IB pitch deck skill builds the analytical backbone of a sell-side pitch: market positioning analysis, comparable transaction set with relevant multiples, preliminary valuation range (EV/EBITDA, EV/Revenue, precedent transactions), buyer universe thesis, and process timeline.

"Build a sell-side pitch for [Company], a $45M revenue B2B SaaS business in facilities management software. LTM EBITDA: $8.2M, growing 22% YoY. Comparable transactions: [list]. Produce: market positioning, valuation range with comp set, strategic buyer universe rationale, and a 12-week process timeline."

The skill is most effective when your CLAUDE.md holds your bank's pitch format preferences, sector coverage focus, and any house views on valuation methodology for specific verticals.

CIM Builder: First-Draft the Book in Hours, Not Days

The Confidential Information Memorandum is the single most labor-intensive document in a sell-side process. A thorough CIM requires synthesizing management presentations, audited financials, market research, competitive positioning, customer concentration analysis, and growth initiatives into a coherent 40 to 80 page narrative. Analysts historically spend two to three weeks on a first draft. Partners spend another week rewriting it.

A CIM builder skill compresses the first-draft timeline by doing the synthesis layer: extracting key facts from data room documents, building the financial summary tables in the right format, drafting the business description and market overview sections, and flagging information gaps that need management follow-up.

"Build the business overview section of the CIM using this management presentation: [paste or attach]. Flag any claims that need verification against the financials, and note anywhere the narrative contradicts the historical growth rate."

"Here are three years of audited financials. Build the financial summary section: income statement bridge, EBITDA reconciliation, revenue breakdown by segment, and a normalized EBITDA calculation with addback justifications."

Private Equity Track

Deal Sourcing: Build Proprietary Pipeline Before the Bankers Call

The best PE deals are the ones where you find the company before it runs a process. Where the relationship is established, the founder trusts you, and you're not competing with seventeen other funds in a banker-run auction.

A deal sourcing skill systematizes this work. Feed it your fund's investment criteria and sector thesis, and it generates a structured target identification framework: the specific company characteristics that signal fit, the data sources for finding them (industry databases, conference attendee lists, trade publication coverage), and outreach sequencing that doesn't feel like cold selling to founder-operators who didn't ask to be acquired.

"Build a deal sourcing system for our thesis: lower-middle-market B2B software businesses, $5 to $25M revenue, founder-owned, serving regulated industries. What are the best identification sources, what signals should we screen for, and what does a 6-month outreach sequence look like that doesn't feel transactional?"

Deal Screening: Kill Bad Deals Before You Build a Model

The purpose of deal screening is to fail fast. Most deals that cross a PE firm's desk don't fit the fund's criteria. Wrong sector, wrong size, wrong business quality, wrong management team, wrong competitive dynamics. The screening phase should surface these mismatches in hours, not after a week of analyst time building a model on a company that never had a chance.

A deal screening skill runs a structured first-pass evaluation against your fund's criteria. Feed it a CIM, teaser, or basic company information, and it produces: a criteria match/mismatch table, preliminary business quality assessment across the key dimensions (recurring revenue, customer concentration, competitive moat, management depth), initial red flags with specific evidence, and a recommendation on whether to advance to a full diligence process.

"Screen this CIM against our fund criteria: [paste CIM summary]. Our parameters: $10 to $50M EBITDA, less than 30% customer concentration in top 5, recurring revenue above 60%, defensible niche, no turnarounds. Score each criterion, flag red flags with citations from the document, and give me a pass/advance/conditional-advance recommendation with reasoning."

What goes in your screening CLAUDE.md:

# Fund Investment Criteria

## Hard Criteria (automatic pass if failed)
- Target EBITDA range: $[X]M to $[Y]M
- Revenue model: [recurring % minimum]
- Geography: [restrictions]
- Sector: [focus / exclusions]
- Deal type: [control buyout / growth equity / etc.]

## Soft Criteria (scored, not automatic)
- Customer concentration
- Management quality signals
- Competitive moat assessment
- Growth quality (organic vs. acquired)

## Automatic Red Flags
- [List specific patterns that are instant disqualifiers]

IC Memo: Build the Investment Case That Survives Partner Scrutiny

The Investment Committee memo is where deals are made or broken internally. An IC memo that presents the thesis clearly, anticipates the objections, and addresses the key risks head-on moves efficiently through approval. One that buries the downside, overweights management optimism, or fails to present the bear case gives the IC committee no choice but to slow the process down with questions that should have been answered in the document.

An IC memo skill builds the full structure from your diligence notes and financial model: investment thesis (3 to 4 sentences, not a paragraph), business overview calibrated to what the IC needs to know rather than everything you learned, market and competitive analysis, financial summary with entry assumptions, returns analysis across scenarios, key risks with explicit mitigants, and the monitoring framework for post-close.

"Draft the IC memo thesis section for [Company]. The investment case in one sentence: [describe]. Key support points from diligence: [list]. Lead with the strongest evidence for the thesis, then address the bear case proactively."

"Write the risk section of the IC memo. Identified risks from DD: [list]. For each risk: quantify the potential impact on EBITDA or exit multiple, identify the specific mitigant, and note how we'll monitor post-close."

The risk section prompt is worth running separately. It's often the weakest part of analyst-drafted memos because analysts are (understandably) invested in the deal advancing. A skill with explicit instructions to "quantify the downside and be specific about mitigants" produces more rigorous risk analysis than most first drafts.

Returns Analysis: Model the Full Exit Spectrum Before You Commit

Returns analysis in PE is about more than the base case. The base case always looks fine. That's why you're bringing the deal to IC. What matters is the distribution of outcomes: how does the return profile hold up under a revenue miss, a multiple compression, a slower-than-expected exit timeline, or a combination?

A returns analysis skill builds the full scenario matrix (base, bull, and bear cases) across the relevant return dimensions: MOIC, IRR, and equity value at exit under each scenario. It flags the specific assumptions the returns are most sensitive to, which tells you where to focus diligence effort and which deal terms matter most in negotiation.

"Build a returns analysis for this deal. Entry: $85M EV, 7.5x EBITDA. LTM EBITDA: $11.3M. Debt: $45M at close. Hold period: 4 to 6 years. Base/bull/bear assumptions: [revenue growth, margin expansion, exit multiple]. Show MOIC and IRR for each scenario, sensitivity table on exit multiple vs. EBITDA growth, and flag the two assumptions we should pressure-test hardest."

The "flag the two assumptions we should pressure-test hardest" instruction is where the skill earns its keep. A returns model that tells you the bear case IRR is 8% is less useful than one that tells you your returns are almost entirely driven by exit multiple assumptions, and that you should therefore spend diligence time on comparable transaction precedents, not revenue forecasting.

Portfolio Monitoring and Value Creation

The deal sourcing and execution skills cover origination through close. The work that drives actual fund performance happens after close: identifying value creation opportunities, tracking portco performance against the investment thesis, and escalating problems before they become permanent impairments.

"Here are monthly KPIs for our five portcos vs. budget: [paste data]. Flag any metrics more than 10% off budget, identify which variances are one-time vs. structural, and tell me which boards need an agenda item on this at the next meeting."

"Build a value creation plan for [Portco]. Investment thesis: [describe]. Current EBITDA: $8.1M. Target at exit: $14M. Generate the initiative list with estimated EBITDA impact, execution owner, and 12-month milestones for each lever."

Due Diligence: The Connecting Tissue

Across both the IB and PE workflows, due diligence is the phase where the most document synthesis happens, and where AI skills provide the clearest time savings. A due diligence skill processes data room documents systematically: extracting key terms from contracts, flagging change-of-control provisions, summarizing customer agreements, identifying representations and warranties exposure, and producing a structured DD findings document mapped to the IC memo risk section.

The specific value here is completeness. Manual data room review is subject to fatigue and attention drift. The clause buried on page 40 of a customer contract that limits assignment rights gets missed. A skill processing documents systematically doesn't get tired of reading the fifteenth contract.

Setting Up Skills at the Firm Level

The highest-leverage implementation is a shared firm CLAUDE.md that every analyst and associate pulls into their project folders. This file holds:

Fund investment criteria. The specific parameters that define a fundable deal. Not the marketing version in your deck, but the actual thresholds that would disqualify a company. Hard criteria and soft criteria separated clearly.

Document format preferences. The section order and content expectations for your IC memo, CIM, and pitch deck. Senior partners have strong preferences. Encoding them prevents the back-and-forth of "we always put market sizing before the business model overview."

Sector thesis and market views. Your fund's current sector focus, thematic investment hypotheses, and any specific market dynamics you're tracking. This prevents analysts from writing market overviews that contradict your partners' published views.

House style for financial analysis. Preferred valuation methodologies by sector, how you define normalized EBITDA, standard addback policy, return threshold expectations. The methodological consistency that currently lives only in the heads of your most senior people.

A deal-specific file sits alongside the firm CLAUDE.md: company name, deal stage, entry assumptions, current open questions. The skill reads both (firm context plus deal context) for outputs that are simultaneously on-standard and deal-specific.

Getting Started

I publish the full PE and IB toolkit as free, downloadable playbooks at claudecodehq.com. The collection includes IB pitch decks, CIM builder, deal screening, IC memos, returns analysis, deal sourcing, due diligence automation, value creation planning, portfolio monitoring, teasers, process letters, buyer lists, merger models, and deal trackers. Each one is a single CLAUDE.md file you drop into a project folder. Start with whichever stage of the deal cycle currently costs your team the most analyst hours.

Originally published on claudecodehq.com

I Cut My Job Application Time From 4 Hours to 90 Minutes. Here Are the 6 AI Skills I Used.

Daniel Marin — Fri, 22 May 2026 10:45:42 +0000

Tailoring resumes, writing cover letters that get read, researching companies, benchmarking salary, and preparing for interviews. No coding required.

Job searching is a full-time job. The cliché exists because it's accurate. Each application demands a tailored resume, a custom cover letter, company research, salary benchmarking, and interview preparation. Multiply that by twenty or thirty applications and the repetitive work alone becomes overwhelming.

The frustrating part is that most of the work is the same every time. The research follows the same pattern. The cover letter hits the same beats. The interview questions are largely predictable. You're not doing creative work. You're doing high-stakes documentation that just happens to feel personal.

AI skills are well-matched to exactly this kind of problem. A skill that knows your background, your target role, your experience level, and your preferred voice can handle the repetitive layer of every application, so you spend your energy on the parts that actually require human judgment: the conversations, the relationships, the decisions.

Setting Up Your Job Search Folder

Before getting into the individual skills, the most effective setup is a dedicated job search folder with a master CLAUDE.md that holds your core context (background, target role, constraints) so every session starts from a complete picture rather than a blank slate.

# Job Search Context

## Background
[Your current or most recent role, industry, years of experience]

## Target Roles
[Job title(s) you're applying for, seniority level]

## Key Strengths
[3-5 accomplishments with numbers: results you've produced]

## Hard Constraints
[Location, remote/hybrid requirement, min salary, industries to avoid]

## Voice & Style
[How you write: formal/conversational, concise/detailed]

## Current Resume
[Paste your base resume text here]

Every skill in this guide builds on this foundation. The more specific your context block, the less editing every output needs. A resume tailored by a skill that knows your background is substantially better than one produced by asking a general AI from scratch.

Per-application sub-folders. For each company you seriously pursue, create a sub-folder (for example, job-search/acme-corp/) and paste the job description and any research notes into a text file there. Skills that read local files can then work with both your master context and the company-specific material simultaneously.

1. Resume Tailoring: Match Every Application Without Rewriting From Scratch

Sending the same resume to every job is a fast path to silence. ATS systems filter for keyword matches before a human ever sees your application, and recruiters who do read resumes spend seven seconds on average deciding whether to continue. Generic resumes fail both tests.

A resume tailoring skill takes your base resume and a specific job description and produces a tailored version: reordered and reworded to surface the experience most relevant to this role, with the exact keywords from the posting woven in naturally.

"Here's the job description: [paste JD]. Tailor my resume for this role. Surface the most relevant experience, match the keywords they're using, and reorder bullet points to lead with what they care about most. Keep everything factually accurate."

The key instruction in that prompt ("keep everything factually accurate") is worth including explicitly. You want the skill to reframe and emphasize, not invent. Claude will honor this constraint reliably when it's stated clearly.

Expect to spend about five minutes editing the output. The skill handles the heavy lifting. You make the judgment calls about what actually sounds like you.

2. Cover Letter Writing: Say Something Specific, Not Something Polished

The worst cover letters are polished but generic. They hit the expected beats ("I'm excited about this opportunity," a paragraph about experience, a closing paragraph) but say nothing that couldn't appear on any other application. Hiring managers read them in two seconds and move on.

The best cover letters are specific: they name something real about the company or role, connect it directly to something real in the applicant's background, and make a concrete case for fit. They read like they were written by a person who did their homework, not by someone filling in a template.

"Write a cover letter for this role: [paste JD]. Open with something specific about what this company is working on, not a generic statement about being excited. Connect my background to their actual problem. Keep it under 300 words."

The "under 300 words" constraint matters. Shorter cover letters get read. Longer ones get skimmed or ignored. If you can't make your case in 250 to 300 words, the case probably isn't clear enough yet.

3. Applicant Screening: See Your Application the Way the ATS Does

Most job seekers optimize their resumes for human readers without thinking about the automated screening layer. At companies that receive hundreds of applications, ATS software filters candidates before a human ever reviews them, matching against required skills, keywords, and experience levels defined in the job description.

An applicant screening skill flips the perspective. Instead of helping you write your application, it evaluates your application the way a screener would: identifying where you match the stated requirements, where you have gaps, and what keywords are present or absent. It tells you whether you'd make it past the first filter before you submit.

"Act as an ATS and recruiter screening this application. Here's the job description: [paste JD]. Here's my resume: [paste resume]. Score my match, identify missing keywords, flag any gaps in requirements, and tell me what I should fix before submitting."

Use this skill after tailoring your resume. It's the QA step before you hit submit. Run it on every application where you're not obviously qualified on paper. The ten minutes it takes can be the difference between getting screened out and getting a call.

4. Background Research: Know the Company Before Anyone Else Does

Arriving at an interview without company research is a disqualifying signal. It tells the interviewer you didn't care enough to spend an hour learning about them. But company research done well (understanding the business model, recent news, competitive position, and what the interviewer's team is actually working on) takes more than casual Googling.

A background research skill structures this work. Give it everything you've gathered (the company website, recent news articles, the job description, the interviewer's LinkedIn profile) and it synthesizes a focused brief: what the company does and how it makes money, recent developments worth knowing, the team's likely priorities, and specific questions you should ask.

"I have an interview at [Company] for a [Role] role on [Date]. Here's what I've found: [paste research]. Produce a pre-interview brief: business model, recent news I should reference, what the team is likely focused on, and 5 smart questions I should ask my interviewer."

The five questions at the end of an interview aren't just courtesy. They're part of the evaluation. Questions that demonstrate you understand the company's actual situation (not just the surface-level pitch) tell interviewers you do your homework. This skill produces exactly those questions.

5. Compensation Benchmarker: Know Your Number Before They Ask

"What are your salary expectations?" is one of the most consequential questions in the job search process, and most candidates answer it with insufficient information. They either anchor too low (leaving money on the table) or too high (pricing themselves out before demonstrating value).

A compensation benchmarking skill takes your role, level, location, industry, and years of experience and produces a structured salary analysis: market range for the role, total compensation breakdown (base, bonus, equity norms), how your specific profile maps to the range, and a recommended ask based on where you sit.

"Benchmark compensation for a [Job Title] with [X years] experience in [city/remote], [industry]. I have [specific skills/credentials]. What's the realistic range, what's a strong ask, and what comp components should I be negotiating beyond base?"

Run this skill before your first conversation with a recruiter, not after. Once you anchor to a number, it's hard to reset. Knowing your market value beforehand also changes how you carry yourself in early conversations: you're evaluating the offer, not hoping to be chosen.

6. Interview Prep: Practice the Real Questions, Not Generic Ones

Generic interview prep ("tell me about yourself," "what's your greatest weakness") produces generic answers. Interviewers hear them dozens of times a week and they register as exactly what they are: rehearsed responses to expected questions.

An interview prep skill takes the job description and your background and generates the specific questions this role is likely to ask: behavioral questions tailored to the stated responsibilities, technical questions matched to the required skills, and culture-fit questions aligned with the company's evident values. Then it helps you build STAR-format answers from your actual experience rather than generic templates.

"Generate the 10 most likely interview questions for a [Role] at [Company type], based on this JD: [paste JD]. For each question, tell me what the interviewer is really trying to assess."

"Help me build a STAR answer for this question: [paste question]. My relevant experience: [describe situation]. Push back if my answer is vague or if I'm not being specific enough about results."

The second prompt ("push back if I'm not being specific enough") is the important one. The most common interview answer failure is vagueness: "I improved team communication" instead of "I introduced a weekly async update that reduced meeting time by 40%." A skill with that instruction will drill you until your answers have real numbers.

The Full Application Sequence

Used together, these skills cover every stage of the application process in a logical order:

Find a role. Save the job description to your application folder.
Resume Tailor. Tailor your base resume to this specific JD: keywords, reordering, emphasis.
Applicant Screening. QA the tailored resume against ATS criteria before submitting.
Cover Letter. Write a specific, under-300-word cover letter that opens with a real hook.
Background Research. Build your company brief: business model, recent news, smart questions.
Compensation Benchmarker. Know your number before the recruiter screen.
Interview Prep. Practice the real questions with STAR answers that have actual numbers.

The whole sequence takes 60 to 90 minutes per serious application, compared to 3 to 4 hours doing it manually. More importantly, the quality at each stage is higher because you have a tool doing the structured work while you focus on the substance.

Common Questions

"Will a cover letter written with AI be obvious to recruiters?"

Only if you use it without editing. The skill produces a draft. The specificity and voice come from the context you provide and the edits you make afterward. A cover letter that mentions a real company initiative and connects it to a real story from your background reads like a human wrote it, because a human (you) did the substantive work. The skill structured it.

"Do I need technical skills to use these?"

No. You create a folder, download a CLAUDE.md file, open Claude Code in that folder, and type your request. The hardest part is having your base resume and target job description ready to paste in, which you already have.

"Can I use these if I'm changing careers?"

Yes, and career changers arguably benefit more than lateral movers. The resume tailoring and cover letter skills are specifically good at identifying transferable skills and reframing experience in the language of a new field. The key is being honest in your master context about both where you're coming from and where you're trying to go.

"How much editing does the output need?"

For most people: 10 to 20 minutes per document. You'll find phrases that don't quite sound like you, facts that need slight adjustment for accuracy, and tone choices you want to nudge. That's normal and expected. The skill saves you from the blank page problem and the structural thinking. You edit, you don't rewrite.

Getting Started

The job search is hard enough without spending four hours manually customizing every application. These skills handle the structural, repetitive work so the time you do spend is on the conversations that actually move things forward.

I publish all six of these skills as free, downloadable templates at claudecodehq.com. Each one is a single file you drop into a folder. No coding, no API keys, no subscription. Start with the resume tailor on your next real application and see how much time the first one saves you.

Originally published on claudecodehq.com

Our Marketing Team Stopped Re-Explaining the Brand to AI Every Session. Here's the System We Built Instead.

Daniel Marin — Thu, 21 May 2026 18:04:06 +0000

From ad copy at scale to campaign analysis: how to make AI remember your brand voice, audience, and strategy across every session.

Most marketing teams are already using AI. But using it badly. The workflow is the same everywhere: open a chat window, explain the brand from scratch, describe the audience, paste in some context, generate output, close the tab. Next session: start over. The brand voice that took years to build gets re-explained in two sentences every time.

Claude Skills fix the fundamental problem. A skill that holds your brand guidelines, target personas, campaign context, and tone rules doesn't need a briefing. Every session starts from a complete picture. The first output is on-brand because the skill already knows what "on-brand" means for you specifically.

This guide covers five skills that span the full marketing workflow (from brand understanding through campaign planning, creative production, content multiplication, and competitive intelligence) and how they work together as a coherent system rather than isolated tools.

Why Skills Beat Chat for Marketing Work

The core advantage is persistent, shared context.

Without a skill: Re-explain brand voice every session. Re-paste persona descriptions per prompt. Output quality varies with who wrote the prompt. No institutional memory between team members. Off-brand outputs require extensive editing.

With a skill: Brand voice baked into every output. Personas, guidelines, tone rules always present. Consistent output regardless of who runs the session. Shareable file across the team. First draft is already on-brand.

For teams, the shared CLAUDE.md is particularly valuable. Drop the skill file into a shared folder (Notion, Drive, Dropbox) and every team member works from the same foundation. The junior copywriter's first ad draft and the senior strategist's campaign brief both start from the same brand context.

1. Brand Analyzer: Understand Your Brand Before You Create Anything

The most common source of inconsistent marketing output is unclear brand definition. "Friendly but professional" means something different to every person on the team. "We talk to small business owners" leaves the audience underspecified. When the brief is vague, the output is unpredictable.

A brand analyzer skill starts by diagnosing your brand before producing anything. Feed it your existing marketing materials (website copy, social posts, ad examples, any brand guide you have) and it produces a structured brand analysis: what you actually communicate (versus what you think you communicate), your current voice characteristics, audience assumptions baked into your messaging, and where your brand positioning is ambiguous or inconsistent.

"Analyze our brand based on these materials: [paste website copy, recent ads, social posts]. What voice characteristics come through consistently? Where are we inconsistent? What does our messaging imply about who we think our audience is?"

Run this first, before building any other skill. The output becomes the brand section of your master marketing CLAUDE.md: the document that every other skill reads to understand who you are before producing output.

Use the output to build your master brand CLAUDE.md:

# Brand Context

## Voice
[Paste the voice characteristics Brand Analyzer identified]

## Audience
[Primary persona: who they are, what they care about, what they fear]

## Messaging Hierarchy
[Core message → Supporting claims → Proof points]

## Tone Rules
[What we always do / never do in copy]

## Examples of on-brand copy
[2-3 samples that represent us at our best]

2. Campaign Planner: Strategy and Architecture Before the Creative Starts

The most expensive mistake in marketing is executing the wrong campaign well. Creative production, ad spend, and distribution costs are all sunk once a campaign launches, which makes pre-production strategy disproportionately valuable. Most teams skip it because it takes time they feel they don't have.

A campaign planner skill compresses campaign strategy into a structured session. Describe your goal, budget, timeline, and audience, and it produces: campaign objective and success metrics, channel mix recommendation with rationale, content pillars and message hierarchy, a week-by-week execution calendar, and the measurement framework to assess whether it worked.

"Plan a campaign to launch [product/feature] in Q3. Goal: [specific outcome with number]. Budget: $[X]. Primary audience: [persona]. Timeline: 8 weeks. Give me the full campaign architecture: channels, messaging pillars, weekly calendar, and what we measure to know if it worked."

The output isn't a document to file. It's the brief that other skills work from. Once the campaign planner has defined the messaging pillars, the ad copy generator knows exactly what angles to explore. Once the channel mix is decided, the content repurposer knows where each asset needs to live.

3. Ad Copy Generator: High-Volume Creative Without Brief Fatigue

Paid advertising at any meaningful scale requires volume. Testing 20 headline variations, 5 angle hypotheses, 3 audience framings: the math produces a lot of copy fast. Writing each piece from scratch is slow. Briefing an agency for each variant is expensive. Asking a general AI without context produces off-brand output you spend more time editing than you saved generating.

An ad copy skill is built for production volume. Because it reads your brand CLAUDE.md, every variant starts on-brand without a briefing. Specify the platform (Meta, Google, LinkedIn), the ad format (headline, body, CTA), the angle (problem-agitate-solve, social proof, curiosity hook), and the quantity. You get a complete test matrix back.

"Write 10 Meta ad headlines for our [product] targeting [persona]. Test 3 angles: time savings, cost reduction, and ease of use. Keep each under 40 characters. Mark which angle each headline uses."

"Generate a full Meta ad set: 5 primary text variants, 3 headline variants, 3 CTA options. Campaign goal: trial signups. Angle: [describe]. Format the output as a table I can paste into our ads manager."

The table format request matters. Most teams waste time reformatting AI output before it can be used. Asking for a specific format upfront means the output goes directly into your workflow. Copy-paste, not copy-reformat-paste.

4. Content Repurposer: Multiply Every Asset Across Every Channel

Most marketing teams produce content at one-tenth the rate their distribution channels can absorb. A blog post gets published once, maybe shared once on LinkedIn, and then it's done. The research, interviews, and thinking that went into it sit unused. Meanwhile the team is staring at a blank content calendar for next month.

A content repurposer skill extracts maximum reach from every piece of content you already have. Feed it a blog post, a podcast transcript, a webinar recording summary, or a case study, and it produces platform-native versions for every channel in your distribution mix: an X thread, a LinkedIn article, an email newsletter excerpt, short-form video script points, and pull-quote graphics text.

"Repurpose this blog post across our distribution channels: [paste post]. We publish on LinkedIn, X, and email. Produce: a 10-tweet thread, a LinkedIn article version (800 words), a 150-word email newsletter excerpt, and 5 pull quotes for social graphics. Match our brand voice throughout."

The repurposing habit is where the ROI math gets compelling. A 2,000-word post that took 4 hours to write becomes 6 pieces of platform-native content in 20 minutes. The team appears to be everywhere without producing proportionally more original work.

5. Competitive Ads Extractor: Mine Competitor Creative for Strategic Intelligence

Competitive ad analysis is one of the highest-leverage research activities in marketing, and one of the most neglected. Competitors' ads tell you what messages they've found worth paying to distribute, which audiences they're targeting, and which angles are getting enough engagement to keep running. That's valuable signal.

A competitive ads skill structures this intelligence. Feed it competitor ads you've collected from Meta Ad Library, Google Ads Transparency, or manual observation, and it extracts: the core message of each ad, the audience it's targeting based on copy signals, the emotional hook being used, how long it appears to have been running (a proxy for performance), and what the aggregate pattern tells you about their current strategy.

"Analyze these competitor ads: [paste ad copy and descriptions]. For each: identify the core message, target audience signals, emotional hook, and angle. Then give me a summary: what messages are they betting on, where are the gaps we could own, and what should we test based on what they're not saying?"

The "what they're not saying" prompt is particularly useful. Competitive gaps (the angles your competitors have left open) are often better creative territory than going head-to-head on the messages they're already spending to amplify.

Closing the Loop: Campaign Performance Analysis

The five skills above cover strategy, production, and intelligence. The final piece is measurement: turning raw campaign data into the insight that improves the next cycle.

A marketing performance reporter skill takes your campaign metrics (CTR, CPC, conversion rate, ROAS, channel breakdown) and produces a structured performance report: what worked, what didn't, what to scale, what to cut, and what hypotheses to test next.

"Here are our Q2 campaign results: [paste metrics by channel and creative]. Write a performance report with headline numbers, channel-level analysis, which creatives won, what we should scale in Q3, and three specific hypotheses to test next quarter."

The Full Marketing Workflow

Used in sequence, the six skills form a complete campaign cycle:

Brand Analyzer (once, then update quarterly): Define your brand voice, audience, messaging hierarchy, and lock it into a master CLAUDE.md.
Competitive Ads Extractor (pre-campaign or monthly): Research competitor creative before planning. Understand the landscape and the gaps.
Campaign Planner (per campaign): Build campaign architecture: objective, channel mix, messaging pillars, calendar, measurement plan.
Ad Copy Generator (per campaign, ongoing): Produce on-brand creative at volume. Full test matrices for every channel and format.
Content Repurposer (every time you publish): Multiply every asset across your distribution channels. 1 piece of content becomes 6+ formats.
Marketing Performance Reporter (post-campaign or monthly): Analyze results, identify what to scale and cut, generate hypotheses for the next cycle.

Setting Up Skills for a Team

Individual skills are powerful. Team-level skills are a force multiplier.

Share the master brand CLAUDE.md. Store it in a shared folder (Notion, Dropbox, Drive). Every team member downloads it into their local project folder. Everyone works from the same brand foundation. No diverging interpretations of "our voice."

Create campaign-specific sub-files. For each campaign, create a campaign context file (campaign goal, messaging pillars, audience, constraints). Team members add this to their project folder alongside the brand CLAUDE.md. The skill then reads both: brand context plus campaign context, for outputs that are simultaneously on-brand and on-brief.

Version and update systematically. Add a version date to your CLAUDE.md header. When the brand evolves, update the shared file and note what changed. Team members know they're working from the current version, not a six-month-old brand brief that was never updated.

Getting Started

Marketing is already a domain where AI adoption is ahead of most fields. But most teams are using it as a faster chat tool rather than a persistent, brand-aware system. The shift from "AI that needs a brief every time" to "AI that already knows the brief" is where the compounding advantage kicks in.

I publish all six of these skills as free, downloadable templates at claudecodehq.com. Each one is a single file you drop into a folder. Start with the Brand Analyzer to build your master CLAUDE.md, then layer on the rest as your workflow demands.

Originally published on claudecodehq.com

I Cancelled $900/Month in SaaS Subscriptions. These 10 AI Skills Replaced Them.

Daniel Marin — Tue, 19 May 2026 18:31:18 +0000

Budget tracking, contract review, CRM, SEO, invoicing, and more. Honest cost comparisons included.

The average solopreneur or small team runs 10 to 15 SaaS subscriptions. Most of them made sense individually when they were added. Collectively, they add up to $600, $900, $1,200 a month, and half of them get used once a week at best.

Claude Skills don't replace everything. Tools with complex databases, real-time integrations, or team collaboration features are still worth what they cost. But a surprising number of SaaS tools are really just document-generation and analysis engines: good at producing a specific type of structured output from structured input. That's exactly what a well-built AI skill does, permanently, for the cost of your existing Claude subscription.

Here are ten categories where the substitution is clean, the output is comparable, and the math works strongly in your favor.

1. Budget Analyzer

Replaces: YNAB, Monarch Money, Copilot ($13 to $17/mo each)

Categorize transactions, surface subscription creep, and produce a realistic monthly budget from your actual bank export. Not an idealized spreadsheet.

Budget apps require you to manually categorize transactions, reconcile accounts, and check in daily to stay on top of things. An AI skill takes your CSV export, does the categorization automatically, flags every subscription charge by service and cost, and hands you a summary. You run it when you need it, not as a daily ritual.

"Analyze my last 3 months of transactions. Show spending by category, flag every subscription charge with the exact amount, and suggest a realistic budget based on my actual habits."

2. Contract Review

Replaces: ContractSafe, LegalZoom, ad-hoc lawyer review ($50 to $300+/review)

Flag risky clauses, summarize obligations on both sides, and redline against your standard terms. In minutes rather than days.

For solopreneurs and small businesses, contract review is either skipped entirely (risky) or sent to a lawyer at $250 to $400/hour (expensive). An AI contract review skill reads agreements and surfaces the specific clauses worth scrutinizing: liability caps, IP assignment, termination triggers, auto-renewal traps. It doesn't replace legal counsel for high-stakes deals, but it means you arrive at that conversation knowing exactly what to ask about.

"Review this service agreement. Flag any unusual liability clauses, one-sided termination rights, or IP assignments I should be aware of before signing."

3. Brand Guidelines

Replaces: Frontify, Bynder, brand consultant ($50 to $200/mo or $1,500+ for a consultant)

Build and maintain a comprehensive brand guide (voice, visual identity, messaging hierarchy, usage rules) from a conversational brief.

Brand management platforms charge enterprise prices for what is fundamentally a structured document. An AI skill produces the same artifact: a complete brand guide covering logo usage, color palette, typography, tone of voice, and messaging do's and don'ts, through a guided session. The output is a file you own and can update yourself.

"Build a brand guidelines document for my business. I'll describe what we do, who we serve, and what we stand for. You produce the full guide including voice, visual rules, and messaging framework."

4. Calendar Automation

Replaces: Motion, Reclaim.ai, Calendly Pro ($15 to $34/mo)

Intelligently schedule your week, protect deep work blocks, and handle meeting logistics without paying for an AI calendar app.

AI calendar tools are genuinely useful, but they require ongoing subscription fees for functionality AI can replicate on demand. A calendar skill takes your current calendar state and priorities, then produces a structured weekly plan: deep work blocks, meeting consolidation, buffer time, and daily intentions. You run it Sunday evening or Monday morning and have a clear week mapped out.

"Here's my week: [paste events]. I need to protect 4 hours of deep work daily and prep for the Thursday client call. Reorganize my schedule to make this work."

5. Business Analytics Reporter

Replaces: Databox, Klipfolio, Looker Studio builds ($47 to $200/mo)

Turn raw business data into executive-ready reports with trend analysis, anomaly detection, and plain-English commentary. Without a BI platform.

Business intelligence tools are powerful but complex, and most small teams use 10% of their capability. A business analytics skill takes your raw numbers (revenue, traffic, conversion rates, whatever you track) and produces a structured report with period-over-period comparisons, highlights, lowlights, and the narrative explanation of what's happening and why. No dashboards to build or maintain.

"Here's our metrics for April vs. March: [paste data]. Write a monthly business performance report with headline numbers, trend analysis, what's working, what needs attention, and three action items."

6. CRM Automation

Replaces: HubSpot Starter, Pipedrive, Close CRM ($20 to $65/mo)

Maintain deal notes, draft follow-ups, update pipeline status, and surface next actions from conversation logs. Without paying for a CRM seat.

For solo operators managing under 50 active relationships, most CRM features go unused. What actually matters: knowing where each deal stands, what was said last, and what needs to happen next. A CRM skill takes your notes or call summaries and produces structured updates (next action, deal status, key context) that slot directly into whatever lightweight system you already use.

"Here are my notes from yesterday's sales calls: [paste notes]. For each deal, extract: current stage, what they said about timeline and budget, my committed next action, and a follow-up email draft."

7. SEO Audit

Replaces: Ahrefs, Semrush, Moz Pro ($99 to $129/mo)

Audit a page or site for technical SEO issues, on-page optimization gaps, and content opportunities. Without a $100+/month tool subscription.

Most SEO platform features are overkill for solopreneurs and small teams who just want to know: is this page optimized, what's wrong with it, and what should I fix first? An SEO audit skill runs a structured analysis of any page (checking title tags, meta descriptions, heading hierarchy, content depth, internal linking, and keyword alignment) and produces a prioritized fix list. Pair it with free tools like Google Search Console for keyword data and you have 80% of the functionality at none of the cost.

"Audit this page for SEO: [paste URL and content]. Check on-page optimization, heading structure, content gaps, and internal linking. Give me a prioritized list of what to fix."

8. Invoice Generator

Replaces: FreshBooks, Harvest, Wave Pro ($16 to $55/mo)

Generate professional, legally complete invoices from project notes or a brief description. Instantly, in any format, with no invoicing software required.

Invoicing tools charge monthly fees for what is fundamentally a document generation job. An invoice skill produces professional invoices from minimal input: client name, line items, rates, payment terms. It handles the formatting, calculates totals, adds the right legal language for your jurisdiction, and outputs a file you can send directly. For freelancers sending fewer than 20 invoices a month, this replaces the invoicing platform entirely.

"Generate an invoice for [Client Name]. Project: website redesign. Hours: 24 at $125/hr. Expenses: $340 for stock photos. Payment due 30 days. Include late fee terms."

9. Meeting Notes to Actions

Replaces: Otter.ai, Fireflies.ai, Notion AI meeting summaries ($17 to $25/mo)

Turn raw meeting notes or a transcript into structured action items, decision logs, and follow-up drafts. In under a minute.

Meeting transcription tools are useful for capturing raw text, but the real value is in extraction: who owns what, what was decided, what still needs to be resolved. A meeting notes skill takes your transcript or rough notes and produces: a decision log, a numbered action list with owners and deadlines, open questions still requiring resolution, and a follow-up email draft ready to send. It takes 30 seconds and means nothing falls through the cracks.

"Here are my notes from today's project sync: [paste notes]. Extract all decisions made, action items with owners, open questions, and draft a follow-up email summarizing next steps."

10. Cold Email

Replaces: Lemlist, Apollo.io, Instantly ($49 to $97/mo)

Write, personalize, and sequence cold outreach emails that actually get replies. Without paying for an outreach platform.

Cold email platforms charge for deliverability infrastructure (worth it at scale) and sequence builders (not worth it for most solopreneurs). The actual hard part, writing emails that resonate, is what they don't solve. A cold email skill focuses on the writing: it researches the prospect from the context you provide, writes a personalized opener, constructs a compelling offer paragraph, and ends with a frictionless CTA. Pair it with your existing email account and a free deliverability tool and you've covered the full workflow.

"Write a cold outreach email to [prospect role] at [company type]. My offer: [describe]. Their likely pain point: [describe]. Keep it under 120 words. End with a single low-friction ask."

Where AI Skills Don't Replace SaaS

This post would be dishonest without the counterpoint. There are categories where SaaS tools are genuinely better and the substitution doesn't work cleanly:

Real-time data and integrations. Tools like Datadog, PagerDuty, or Stripe need to ingest live event streams and trigger instant alerts. AI skills work on data you bring to the session. They can't poll your infrastructure or fire webhooks.

Multi-user collaboration. Notion, Linear, and Figma are collaboration surfaces. AI skills are single-user by nature. They augment your work, not your team's shared workspace.

Email and calendar infrastructure. The calendar skill plans your week. It doesn't send calendar invites or sync across devices. For the actual sending and syncing, you still need your existing email/calendar provider.

High-volume automation at scale. If you're sending 10,000 cold emails a month, a platform with deliverability infrastructure and sequence automation is still the right call. AI skills shine at the writing and strategy layer, not mass send infrastructure.

The substitution is most effective for solopreneurs and teams under 10 people, where the marginal cost of SaaS is high relative to usage, and the flexibility of an AI skill often produces better results than a rigid template-driven tool anyway.

How to Actually Make the Switch

The practical move isn't to cancel everything at once. It's to run the AI skill in parallel with your current tool for one billing cycle, then decide.

Pick one tool from the list above that you rarely use but keep paying for.
Set up the corresponding skill. Takes under five minutes.
Use it for your next real task in that category. Don't test it with toy data. Give it something you actually need.
If the output is good enough, cancel the subscription at the end of the month. If it isn't, you've lost nothing.

Most people find that two or three substitutions cover their cost concerns entirely. The rest is just bonus. Start with whichever one stings most on your credit card statement.

Getting Started

I publish all ten of these skills as free, downloadable templates at claudecodehq.com. Each one is a single file you drop into a folder. No coding, no configuration, no new subscription. Pick the one that matches your most expensive underused tool and try it on real work this week.

Originally published on claudecodehq.com

I Built 5 Custom AI Skills Without Writing a Single Line of Code. Here's Exactly How.

Daniel Marin — Sun, 17 May 2026 18:33:32 +0000

If you can write an email, you can build a Claude Skill. Five real examples with copy-paste snippets you can customize today.

Every guide to "building Claude Skills" eventually gets to a line like: "then edit your configuration file" or "paste this into the terminal." And for a lot of people, that's where the tutorial ends. Not because they can't do it, but because nobody explained what any of it actually means.

Here's what nobody tells you: building a custom Claude Skill is just writing a text file. It doesn't require a programming language. It doesn't require a terminal. It doesn't require any technical background at all. If you can write an email, you can build a Claude Skill.

What "Building a Claude Skill" Actually Means

A Claude Skill is a set of instructions Claude reads before you start talking to it. These instructions tell Claude: who it's acting as, what its job is, how to format its answers, and how to handle edge cases. Everything Claude needs to be immediately useful, without you having to explain it every single session.

The instructions live in a file called CLAUDE.md. That's the entire technical mechanism. One file, in a folder, written in plain English. When you open Claude Code in that folder, it reads the file automatically and the skill activates.

The mental model: Think of it like a job briefing document. A new employee (even an incredibly capable one) needs to know: what's my role here, what am I supposed to produce, and what are the rules? The CLAUDE.md is that briefing. You write it once, and Claude shows up to every session already knowing the answers.

"Building a skill" sounds like software development. It isn't. It's closer to writing a really good job description, which is a skill most people already have.

The Anatomy of a CLAUDE.md File

You don't need to follow any particular format to build a working skill. But the most effective files tend to answer the same five questions:

1. Who is Claude in this context? A role or persona that frames every response. "You are a senior financial analyst" or "You are a plain-language editor who rewrites jargon into clear prose." This isn't decoration. It shapes how Claude reasons, what vocabulary it uses, and what it prioritizes.

2. What is Claude's specific job? The task scope: what it should do and, just as importantly, what it should not do. Defining scope prevents Claude from wandering into adjacent territory you didn't ask for.

3. What does a good response look like? Output format instructions: numbered list, markdown table, two-paragraph summary, bullet points, JSON, a specific structure you've seen work well. Claude follows these exactly.

4. What context does Claude need? Facts about you, your business, your audience, your constraints. "My audience is non-technical C-suite executives." "I run a solo consulting practice, not a large team." This is what makes the skill feel personal.

5. What are the edge cases? Anticipate the messy situations: "If I give you incomplete data, ask me for the missing fields before proceeding." Skills that handle edge cases feel polished and professional.

You don't need all five to have a working skill. Even a two-paragraph CLAUDE.md that establishes a role and an output format is dramatically better than starting from scratch every session.

Five Custom Skills You Can Build Today

The following examples include real CLAUDE.md snippets. Each one is short enough to read in a minute and customize in five. Open any text editor, paste the text, edit it for your situation, save it as CLAUDE.md in a project folder. That's the whole build process.

1. A Personal Writing Voice Coach

Most people write differently than they speak: more formal, more hedged, more jargon-heavy. A writing voice coach skill locks in your natural voice and applies it consistently across everything you write.

# Writing Voice Coach

## Your Role
You are my personal writing coach and editor. Your job is to help me write and edit
content that sounds exactly like me, not a generic AI, not a corporate press release.

## My Voice
- Direct, warm, and slightly conversational
- Short sentences. No filler phrases like "In conclusion" or "It's worth noting"
- I use contractions. I avoid passive voice
- I never use the words: leverage, synergy, holistic, transformative, or robust

## What You Do
When I give you a draft: edit it to match my voice without changing my meaning.
When I give you a topic: write a first draft in my voice, then ask if I want changes.
When I give you a sentence that's too formal: rewrite it the way I'd actually say it.

## Output Format
Always show: [Your edited version] then a one-line note on the biggest change you made.

To customize it: replace the bullet points under "My Voice" with words and phrases that describe how you actually write. Look at your best emails and pull the patterns out of them. The more specific you are, the more accurate the output.

2. A Meeting Prep Assistant

The five minutes before a high-stakes meeting are usually spent skimming your own notes in a panic. A meeting prep skill changes that.

# Meeting Prep Assistant

## Your Role
You are my pre-meeting briefing assistant. I'll give you context about an upcoming
meeting and you'll prepare me to walk in sharp.

## My Context
I'm a [your role] at a [your company type]. My meetings typically involve [clients /
internal teams / vendors / investors].

## What You Produce
For every meeting brief, produce exactly:
1. **3-bullet situation summary**: what's happening and why this meeting matters
2. **5 questions I should ask**: ordered from most to least important
3. **Watch-outs**: 2-3 things that might come up I should be ready for
4. **My one-line goal**: what a successful outcome looks like for me

## Edge Cases
If I give you an agenda with no context: ask me two clarifying questions before
proceeding. If I only have 2 minutes: give me just the situation summary and top question.

Fill in the "My Context" section with your actual role and who you typically meet with. The specificity there is what makes the watch-outs genuinely useful rather than generic.

3. A Daily Life Automation Auditor

Most people have more automatable friction in their day than they realize: repetitive tasks they've just normalized because they've been doing them forever.

# Automation Auditor

## Your Role
You are a workflow efficiency consultant. I'll describe my daily or weekly routines
and you'll identify which tasks I'm doing manually that could be automated, delegated,
or eliminated entirely.

## Output Format
For each area I describe, produce:
- **Manual tasks found**: list of things I'm doing by hand
- **Automation potential**: High / Medium / Low with one-line reason
- **Quickest win**: the single easiest thing to automate first, with a concrete
  tool recommendation

## Rules
Don't recommend enterprise software I'd need IT to approve. Focus on tools I can
set up myself in under an hour: Zapier, Apple Shortcuts, Claude Code playbooks,
IFTTT, Notion automations, calendar rules, email filters.

4. A Custom AI Persona for a Specific Domain

General-purpose Claude is good at everything but expert at nothing specific to you. A custom persona skill lets you build a specialized version.

# [Domain] Expert

## Your Persona
You are a [role] with [X years] of experience in [specific domain]. You have deep
knowledge of [specific subfields]. You communicate like a senior practitioner talking
to a capable colleague, not a professor lecturing a student.

## What You Know
- [Industry-specific fact 1]
- [Industry-specific terminology or standards]
- [Common mistakes you've seen and how to avoid them]

## How You Respond
- Lead with the direct answer, then explain the reasoning
- Flag when something is genuinely uncertain vs. when you're confident
- If I ask something outside your domain: say so and redirect me

## Tone
[Formal / conversational / technical / plain-language]

Replace the brackets with real content from your domain. The more specific the "What You Know" section, the better Claude will perform on domain-specific questions that a generalist would get wrong or hedge excessively.

5. An App Connector and Cross-Tool Workflow Builder

One of the most powerful uses of a custom skill is building a workflow that ties together multiple apps: your calendar, your task manager, your email, your notes.

# Workflow Builder

## My Tool Stack
- Task manager: [Notion / Todoist / Linear]
- Calendar: [Google Calendar / Apple Calendar]
- Email: [Gmail / Outlook]
- Notes: [Obsidian / Notion / Apple Notes]
- Automation: [Zapier / Make / Apple Shortcuts]

## Your Job
When I describe a workflow I want to automate, produce:
1. A plain-English description of what should trigger it and what should happen
2. The exact steps to build it in [my automation tool]
3. Any gotchas I might hit and how to handle them

## Output Rules
Be specific about field names and app settings, not generic. If you're not sure
what a field is called in a specific app, say so rather than guessing.

How to Make Your Custom Skill Better Over Time

The first version of any CLAUDE.md you write will be good. The fifth version will be noticeably better. That's normal, and it doesn't require any technical work, just observation and iteration.

After each session using your skill, ask yourself one question: "What did Claude do that I had to correct or rephrase?" The answer is almost always a missing instruction. Add it to the CLAUDE.md, save the file, and it won't happen again next session.

The iteration pattern:

Use the skill on a real task
Note one thing that wasn't quite right
Add one instruction to CLAUDE.md that would have prevented it
Repeat next session

Most people get to a stable, highly-personalized skill file after 4 to 6 sessions. After that, the skill rarely needs changes. It's tuned to exactly how you work, and every session starts from that baseline.

Common Mistakes in Custom Skills

Being too vague about output format. "Give me a summary" can mean a paragraph, a bullet list, a table, or ten things. If you have a format that works for you, describe it explicitly. Claude will follow it exactly.

Writing instructions that conflict. "Be thorough" and "be brief" in the same file will produce inconsistent results. Pick one. You can always ask for more in the conversation if you need it.

Forgetting to add personal context. The thing that makes a skill feel personal rather than generic is the context section. Who you are, what you're trying to accomplish, what constraints you're working under. Don't skip this.

Writing it once and never touching it again. The CLAUDE.md is a living document. The best skills evolve over time as you notice what works and what doesn't. Budget five minutes after your first few sessions to add one new instruction based on what you saw.

Getting Started

The best first skill to build is the one that solves the thing you find most repetitive right now. Not the most impressive use case, not the most technically sophisticated. The one where you currently spend the most time re-explaining the same context to Claude every session.

Write three sentences: who Claude is in this context, what its job is, and what a good response looks like. Save that as CLAUDE.md. You now have a working skill. Refine from there.

If you'd rather start from something pre-built and adapt it, I publish a full library of downloadable skills at claudecodehq.com, each one a fully-crafted CLAUDE.md you can customize. The pattern is always the same: create a folder, drop in the file, open Claude Code, start working. Building your own custom skills follows exactly the same pattern. You just write the CLAUDE.md yourself instead of downloading it. There's no technical step between those two things.

Originally published on claudecodehq.com

I'm a Freelancer. These 5 AI Skills Save Me 15 Hours a Month on Work That Doesn't Pay.

Daniel Marin — Sat, 16 May 2026 16:02:39 +0000

Content, cold email, proposals, contract review, and invoicing. The admin work that eats your billable hours, automated.

Freelancing sounds like freedom: pick your projects, set your rates, work when you want. And it is, until you account for the invisible second job that comes with it. Writing proposals. Chasing invoices. Reviewing contracts. Producing content to stay visible. Cold-emailing prospects who never reply. None of that work is billable. All of it eats time.

The average freelancer spends 15 to 20 hours a month on admin work that generates zero revenue. That's two to three billable days lost every month to the business of running a business.

Claude Skills (purpose-built instruction files that tell Claude Code exactly how to handle a specific task) can automate most of it. Here are the five that matter most.

1. Blog Post Writer: Stay Visible Without Spending Hours Writing

The pain: You know thought leadership brings in clients. You never have time to write it.

Freelancers who publish consistently attract better-quality inbound leads. The problem isn't knowing what to write. It's the activation energy of sitting down and writing it. Researching, drafting, editing, formatting, and publishing a single post can eat half a day.

A blog writing skill turns a rough idea or bullet list into a fully structured, on-brand post ready to publish.

"Write a 900-word blog post for freelance UX designers. Topic: 'Why your portfolio is losing you clients (and what to fix this weekend)'. Audience: mid-level freelancers looking to level up their rates. Tone: direct, practical, no fluff. Include a clear intro hook, three actionable sections, and a CTA at the end."

What you get: A ready-to-publish post with a strong hook, structured sections, smooth transitions, and a CTA, written in the tone you specified. You edit for voice, add any personal examples, and publish. Total time: under 20 minutes instead of 3 hours.

Time saved: 2 to 3 hours per post.

2. Cold Email Personalizer: Land Clients Without the Dread

The pain: Cold outreach works. Personalized cold outreach really works. Writing 30 personalized emails does not.

Generic cold emails get ignored. Personalized ones (referencing the prospect's recent work, a specific problem their company is facing, a detail from their LinkedIn) get replies. But personalizing at scale is the task that every freelancer knows they should do and almost nobody actually does, because it takes forever.

A cold email skill takes a prospect list and produces a tailored email for each one.

"Write a cold email to Sarah Chen, Head of Design at Fintech startup Lumio. I'm a freelance product designer. They just raised a Series A and their website still looks pre-seed. Reference that the funding announcement mentioned plans to triple the team. They'll need a design system to scale. Keep it under 100 words, no fluff, clear ask."

What you get: A tight, specific email that opens with an insight about their situation, explains the gap it creates, and makes a clear, low-friction ask. All in under 100 words. Run the same prompt for each prospect on your list, adjusting the input details. Twenty emails in twenty minutes.

Time saved: 4 to 6 hours per outreach batch.

3. Proposal Generator: Pitch Without Starting From Scratch

The pain: Every proposal feels like it should be custom. Writing from a blank page every time is exhausting.

Whether you're responding to an Upwork brief, a referral intro, or an RFP from a mid-market company, the proposal is where deals are won or lost. It's also where most freelancers spend disproportionate time.

A proposal generator skill produces a tailored pitch from your core credentials and the job details. You add the human touches. It handles the structure and the boilerplate.

"Write a proposal for a freelance content strategy contract. The client is a B2B SaaS company in the HR tech space. They need a 6-month content strategy, monthly editorial calendar, and 4 long-form articles/month. My background: 6 years in B2B content, worked with 3 HR tech companies. Emphasize ROI and measurable outcomes. Keep it under 400 words."

What you get: A structured proposal with a punchy opener, relevant credential highlights, a clear scope summary, and a confident close. Paste it into your proposal template, adjust the specifics, attach your rate, send. First draft in under two minutes.

Time saved: 1 to 2 hours per proposal.

4. Contract Review: Catch Bad Terms Before You Sign

The pain: You can't afford a lawyer for every contract. You also can't afford to sign one with a trap in it.

Freelancers sign contracts that disadvantage them all the time: IP assignment clauses that hand over work you never intended to sell, non-compete terms that block future clients in your niche, payment clauses with 90-day net terms buried in paragraph 11.

A contract review skill reads a contract and flags the clauses that actually matter: payment terms, IP ownership, kill fees, non-solicitation, and liability caps. It tells you what to push back on and how.

"Review this freelance services agreement. Flag anything unusual in the IP assignment, non-compete, payment terms, and liability sections. Tell me what's standard, what's negotiable, and what I should push back on before signing. Give me specific redline language for the two or three issues that matter most."

What you get: A prioritized list of flagged clauses with plain-English explanations of what each one means for you practically, which are standard and not worth fighting, and which are legitimately problematic. Plus specific redline language you can paste into your response email, without paying $400/hour for the same output.

Time saved: 2 to 3 hours (or $300 to $500 in legal fees).

5. Invoice Generator: Get Paid Faster, With Less Friction

The pain: Generating and sending invoices is the last thing you want to do after finishing a project.

Invoice generation sounds trivial until you're doing it for 15 clients a month, with different rates, different currencies, different tax requirements, and different payment methods.

An invoice skill produces a properly formatted, branded invoice PDF from a one-line prompt, with sequential numbering, correct tax calculation, and your payment details in the format the client expects. No more copying last month's invoice and forgetting to change the date.

"Invoice for Bright Agency. 32 hours of brand strategy work at $150/hr. Plus $600 for the brand audit deck. They're UK-based so GBP and 20% VAT. Net 30. Their billing contact is accounts@brightagency.co."

What you get: A fully formatted PDF with correct VAT calculation, your bank details, payment terms, and due date. Saved to your invoices folder with a consistent naming convention. Ready to email in under 60 seconds.

Time saved: 15 to 20 minutes per invoice.

The Compounding Effect: Running Them Together

Each skill saves time on its own. Together, they cover the full arc of winning and running freelance work, from landing a client to getting paid:

Attract inbound. Blog writing produces consistent thought leadership that brings warm leads to you, so cold outreach isn't the only source of new business.

Run targeted outreach. Cold email sends tailored pitches to the prospects you want, at a volume that would otherwise be impossible to personalize.

Win the pitch. The proposal generator produces a tailored pitch in two minutes, fast enough to respond while the client is still warm, strong enough to stand out.

Sign safely. Contract review catches the clause that would have assigned your IP or locked you into 90-day net terms before you agree to it.

Get paid cleanly. The invoice generator creates and sends a professional invoice the day the project closes. No delay, no formatting errors, no awkward follow-up.

Total time saved across the five skills: roughly 12 to 18 hours per month. For a freelancer billing at $100/hour, that's $1,200 to $1,800 in recovered capacity every month. At higher rates, the math gets more compelling fast.

Before You Set These Up: Do a 15-Minute Time Audit

Log where your non-billable hours went last month. Most freelancers are surprised how quickly "just a couple hours on admin" adds up. The skill that saves the most time is always the one that addresses your biggest actual drain, not the one that sounds most impressive. Start there.

Getting Started

The goal isn't to automate your work. It's to automate the work around your work. The part clients pay for is your judgment, your craft, your relationships. Everything else is overhead. These skills handle the overhead so you can do more of the former.

I publish all five of these skills (and dozens more) as free, downloadable templates at claudecodehq.com. Each one is a single file you drop into a folder. No coding, no API keys, no subscription. Pick the one that addresses your most painful non-billable task and set it up today. It takes under ten minutes. Most freelancers who try one set up all five by the end of the week.

Originally published on claudecodehq.com

I Lost Code Claude Wrote for Me. Then I Found Out Every Session Is Saved. Here's How to Find It.

Daniel Marin — Tue, 12 May 2026 17:15:05 +0000

Where Claude Code stores session history, how to search past sessions by date or keyword, how to recover lost code, and four workflows that make session management effortless.

Claude Code automatically saves the full history of every session: the code it wrote, files it changed, commands it ran, and every message exchanged. This history lives on your machine in a specific directory and is fully searchable.

If you've ever closed a session and realized you needed something from it, the good news is: it's almost certainly still there.

Where Claude Code Stores Session History

All Claude Code session data is stored in the ~/.claude/ directory on your machine. This is your home directory's hidden .claude folder, present on macOS, Linux, and WSL on Windows. Inside it you'll find:

~/.claude/
├── projects/          # One subfolder per project, keyed by directory path
│   └── <project-id>/
│       └── <session-id>.jsonl  # One file per session
├── settings.json      # Your global Claude Code preferences
└── CLAUDE.md          # Your global CLAUDE.md (if set)

Session files use the .jsonl (JSON Lines) format: one JSON object per line, each representing a message, tool call, file change, or command execution in the session. Every session gets its own file, timestamped in the filename.

Windows users: If you're running Claude Code in WSL, the ~/.claude/ directory is inside your WSL home (\\wsl$\Ubuntu\home\yourname\.claude\). If running natively on Windows, check %USERPROFILE%\.claude\.

Viewing Session History With the CLI

The fastest way to view or resume a past session is with the built-in claude CLI. Two flags cover the most common needs:

List recent sessions

# Resume the most recent session in the current project
claude --continue

# Or pick from a list of recent sessions
claude --resume

claude --resume opens an interactive picker showing your recent sessions with timestamps and the first message of each session as a preview. Use arrow keys to navigate, Enter to resume. Claude reloads the full conversation context from the session file. It can see everything that happened, not just a summary.

Continue the most recent session instantly

claude --continue

This picks up exactly where you left off without the picker. Useful when you closed the terminal by accident and want to get back immediately. Claude resumes mid-thought. If it was in the middle of a task, it can see that and pick back up.

Searching Session History Manually

When you know roughly what you're looking for (a specific function name, an error message, a timestamp), you can search the raw session files directly. They're plain text (JSON Lines), so standard tools work:

Search all sessions for a keyword

# Search all sessions for a function name or string
grep -r "generateInvoice" ~/.claude/projects/

# Show the surrounding context (5 lines before and after)
grep -r -C 5 "database migration" ~/.claude/projects/

Find sessions from a specific date

# List session files modified in the last 7 days
find ~/.claude/projects/ -name "*.jsonl" -mtime -7 -ls

# List sessions from a specific project directory
ls -lt ~/.claude/projects/*/*.jsonl | head -20

Extract code blocks from a session

cat ~/.claude/projects/<id>/<session>.jsonl | python3 -c "
import sys, json
for line in sys.stdin:
    obj = json.loads(line)
    if obj.get('type') == 'assistant' and obj.get('content'):
        for block in obj['content']:
            if isinstance(block, dict) and block.get('type') == 'text':
                print(block['text'])
"

This works but gets tedious fast for anything beyond a one-off lookup. The four workflows below automate this into repeatable processes.

Workflow 1: Find and Recover Lost Code

The most common session history need: you closed a session and want something back. Maybe Claude wrote a migration script you never committed. Maybe it refactored a function in a way you want to reference. Maybe the terminal crashed mid-session. The code is in ~/.claude/. You just need to find it.

Describe what you're looking for (the approximate date, a file name, a keyword from the conversation, a function name) and a recovery workflow scans the ~/.claude/ directory, extracts matching sessions, reconstructs the relevant code blocks with their original file paths and context, and presents them ready to copy or re-apply.

"Find the database migration code Claude wrote for me last Tuesday. It was for the users table, adding the email_verified column."

The output: scans session files from that timeframe, finds the migration code block, extracts it with the original file path and surrounding context, and offers to re-apply the change to the current project.

Workflow 2: Log Decisions as They Happen

Session recovery solves the problem of finding past work reactively. A session logging workflow solves it proactively: it automatically detects architectural decision points during a development session and logs them (the decision made, alternatives considered, and the rationale) into a structured, searchable log file alongside your code.

Six months later, when you or a teammate wonders why a certain design choice was made, the answer is in a log file rather than lost with the session.

"Start the session logging protocol for this refactoring session. Log all architectural decisions, including why we chose to extract the payment module separately instead of keeping it inline."

The output: a structured decision log at docs/decisions.md with timestamped entries for each key decision point, alternatives considered, and rationale, updated throughout the session.

Workflow 3: Capture Context for the Next Session

Every time you start a new Claude Code session, you lose the context of the previous one. The new session doesn't know what decisions were made yesterday, what blockers are open, or what was left mid-task. The first ten minutes of every session become re-explaining.

A session capture workflow runs at session end and produces a structured handoff note: key decisions made, open questions with context, completed vs. remaining tasks, and a file change log with rationale. The next session loads this note as its starting context.

"Capture this session for handoff. I'm stopping for the day and want to continue tomorrow without losing context. Summarize decisions, open blockers, and what's left on the auth refactor."

The output: a SESSION_HANDOFF.md with decisions, blockers, completed/remaining tasks, and file change log. Load it in the next session: "Read SESSION_HANDOFF.md and continue from there."

Workflow 4: Visualize Months of Session History

If you use Claude Code heavily, you accumulate months of session history. That history contains a lot of accumulated problem-solving: approaches that worked, dead ends, architectural insights, debugging patterns. It's institutional knowledge that's effectively invisible because it's buried in hundreds of session files.

A mind mapping workflow exports your session history, chunks it into summaries, and generates a visual mind map organized by topic cluster, surfacing recurring themes, key insights, and patterns across sessions. The output is importable into NotebookLM or Obsidian, turning scattered session history into a searchable second brain.

"Map my last 6 months of Claude Code sessions. Cluster by topic area (auth, database, API layer, devops) and surface the key decisions and solutions from each cluster."

Which Approach to Use

I need to get back to where I just left off: Use claude --continue or claude --resume. Built-in, instant, no extra setup needed.

I lost code Claude wrote in a past session: Use the recovery workflow. Describe what you're looking for. It scans ~/.claude/ and extracts the relevant code.

I want to document decisions during a session: Use session logging. Activate it at session start. It auto-logs key decisions throughout.

I'm stopping for the day and want to continue seamlessly tomorrow: Use session capture at session end. Load the handoff note at the start of tomorrow's session.

I want to extract knowledge from months of session history: Use mind mapping. Export, cluster, and import into NotebookLM or Obsidian.

Common Questions

"How long does Claude Code keep session history?"

Session files in ~/.claude/ are kept indefinitely. They don't expire or auto-delete. They only disappear if you manually delete them or clear the directory. On a typical development machine, the ~/.claude/ directory stays manageable in size (a few hundred MB) even after years of use.

"Is my session history private?"

Session files stay on your local machine. They're not synced to Anthropic or any cloud service by default. The content of your sessions (code, file paths, conversation) lives only on the machine where Claude Code is installed.

"Can I view session history on a different machine?"

Sessions are local, so they don't automatically transfer between machines. You can copy the ~/.claude/projects/ directory manually to another machine and Claude Code will recognize the session files there. The session capture workflow is a cleaner alternative for multi-machine setups. It produces a portable markdown file instead of depending on raw session files.

"Can I search sessions for a specific file that was changed?"

Yes. File modification events are logged as tool calls in the session JSONL. The recovery workflow can search specifically for file-change events, making it easy to find "what did Claude do to auth.ts in the session last Thursday" without sifting through conversation text manually.

Getting Started

The ~/.claude/ directory is one of the most underutilized features of Claude Code. Most developers discover it only after losing code they wish they'd saved, and then realize the history was there all along. The built-in --resume and --continue flags cover the simple cases.

I publish free playbooks for all four of the advanced workflows (recovery, logging, handoff, and mind mapping) at claudecodehq.com. Each one is a single file you drop into a project folder. No coding, no configuration.

Originally published on claudecodehq.com