A skill is a folder with a SKILL.md file in it: a short description at the top, then the instructions an agent follows, plus any scripts, references and templates the job needs. Claude Code reads only the descriptions at the start of a session. When my request matches one, it loads the full file and follows it.
This note lists every skill I run, 1 line each. For how skills fit into the rest of NicAI, see What is NicAI and The NicAI workspace.
The count
23 Sep 2026
Folder
Scope
Skills
~/.claude/skills
Global, loaded in every project
4
~/ai/ka/.claude/skills
Work: Kaltura sales
59
~/ai/me/.claude/skills
Personal
13
~/ai/lab/*/.claude/skills
Experiments, 1 lab project
1
Total
77
Some honest footnotes on that number:
2 of the 13 personal skills sit 1 folder too deep, so Claude Code doesn't load them. They are dormant until I move them.
2 of the 59 work skills are kept for reference only: an old deck builder merged into deck, and an avatar builder for a retired version of the studio.
The work folder also holds 2 eval workspaces and a zip file. They aren't skills, which is why earlier counts said 62.
2 work skills that maintain private data sources aren't listed in the tables.
The global folder also holds a synced copy of 16 skills that Anthropic ships (Word, PowerPoint, PDF and similar). I don't count those.
So 73 skills are live and current. Between them, the 59 work skills hold about 15,000 lines of instructions. The shortest is 39 lines, the longest 893. Across the work and personal folders, 32 skills carry a scripts/ folder, 41 a references/ folder, and 4 an evals/ folder with test cases.
How I design them
One skill, one job.prep-call preps a call. email drafts an email. Small skills are easier to test, fix and combine than one big prompt.
The description is the trigger. It lists the phrases I actually type, and says when not to use the skill, so near neighbours don't collide.
Scripts do the mechanical work, the model does the thinking.deck is the clearest case: a script assembles the HTML from a reviewed master deck, and the model only writes the argument.
Skills call skills.init-client runs company-research. Visual skills read brand and logo first. council runs inside any skill when a deliverable needs extra rigour.
Reference skills hold rules, not actions.brand, logo and google-style never produce anything on their own.
Retire, don't delete. A retired skill keeps a stub that points to its replacement, so an old trigger lands in the right place.
Prospecting and outreach
Skill
What it does
company-research
Deep-dive research on a company from its domain, for prospecting
init-client
Sets up a new account from 1 domain: client note, research, folder
accounts-update
Syncs my account list from a CRM export into the prospects database, after I approve the SQL
competitive-brief
Builds a battlecard against a competitor from its website
brand-colors
Extracts a company's colour palette from its homepage
lead-messaging
Writes personalised outreach sequences for 1 prospect, in my style, for a colleague to send
linkedin-invitation
Writes connection notes under the 300-character limit
linkedin-message
Writes follow-up messages to people who accepted a connection
email
Drafts every email in my voice. Mandatory for any email task
calendar
Reads my live availability and suggests meeting slots
format-date
Turns free-text dates and slots into my 1-line-per-day format
Meeting prep and follow-up
Skill
What it does
prep-call
Full brief before any sales meeting
transcript
Local, speaker-labelled transcript of a recording, no API
call-notes
Turns call recordings into client-ready notes, appended to the client file
meeting-recap
Pulls intelligence and action items out of a transcript
exec-summary
Short, military-style internal summary of a call
se-brief
Technical-only brief for a solutions engineer, sales content stripped
csm-handover-brief
Handover brief for whoever takes over an account
pm-handover-brief
Delivery brief for a project manager, from signature to go-live
Deals and commercial
Skill
What it does
proposal
Business proposals for clients
rfp
RFP and tender answers, plus follow-up questions
business-case
Internal business cases and investment arguments
client-business-case
ROI cases a client champion can forward to their CFO
contracts
Tracks contracts in progress and writes an executive overview
legal
Answers contract questions from Kaltura's standard legal documents
pipeline
Analyses my pipeline from CRM report exports
salesforce
Reference for the CRM exports: which report feeds what, freshness checks
pricing-calculations
Turns quote exports into calculation spreadsheets
salesforce-quote
Extracts a quote table from a screenshot into .xlsx
Content, decks and demos
Skill
What it does
deck
reveal.js sales decks assembled from a reviewed master deck, with .pptx on request
deck-pptx-legacy
Retired stub, points to deck
diagram
Architecture and flow diagrams from a pattern library, HTML plus PNG
document
Branded Word documents
one-pager
A4 one-pagers in HTML, converted to PDF
webinar
Webinar deck plus a speaker script for each slide
create-video
Motion-graphics videos rendered from HTML
ep-demo
A tailored demo event package for Kaltura's Event Platform, from a website
avatar-demo
Builds a conversational avatar agent and publishes it to the studio via API
avatar-demo-eself
Legacy avatar package for the previous studio, reference only
avatar-demo-live
A web page with a live, embedded talking avatar (global)
avatar-video
Pre-recorded avatar video from a script (global)
task-dashboard
HTML dashboard of done vs left on a multi-step task
json2html
Any JSON array to a searchable, sortable HTML table
linkedin-post
Edits my LinkedIn posts. Writes from scratch only when I ask
linkedin-comment
Short comment suggestions on a post, in my voice
Knowledge and data upkeep
Skill
What it does
wiki
Maintains and queries my Kaltura knowledge wiki, nightly. See LLM Wiki
kb-article
Structured knowledge base articles
kb-index
Rebuilds the 1-file index of every KB article
md-converter
Converts any document to clean Markdown for the KB
ka-emails
Converts saved work emails to Markdown, sorted external vs internal
db
Queries and manages my SQLite sales databases
index-img
Maintains a described index of 900+ product images, so agents pick the right one
webinar-index
Adds a webinar page to a Markdown index
youtube-videos
Finds YouTube videos by speaker, company, format and date
Shared references
Skill
What it does
brand
Kaltura brand rules that every visual skill reads first
logo
Finds and sizes company logos for other skills
google-style
My standing style for technical writing and chat replies (global)
Quality and meta
Skill
What it does
council
Expert agents critique and score a deliverable until they converge. See The Council
client-lens
Reviews an artifact as the buyer who will receive it
debrief
A daily micro-interview that builds a profile of how I work and decide
Research, media and publishing
Skill
What it does
use-x
Research on X and the web through Grok, plus image generation (global)
youtube
Transcripts, metadata, chapters, audio or video from a YouTube link
Clones a website section by section with parallel builder agents (lab)
Home, family and health
Skill
What it does
it-support
My own IT help desk: knows my hardware, apps and automations
school
Turns school letters into dates, to-dos and calendar files
kids
Family assistant for school matters: dates, deadlines, my open to-dos
health
Answers health and fitness questions from my own records
council (personal)
A medical version of the council: specialist agents review a health question
withings
Syncs readings from my connected health devices into SQLite
supplement-brief
Turns a supplement product page into a structured product card
What works, what doesn't
What works: the narrow skills with scripts behind them. deck, transcript and prep-call do 1 job each, and the mechanical steps run as code, so the output is the same shape every time. My corrections go into the SKILL.md or its references, so the next run starts from them.
What doesn't:
Overlap.meeting-recap, call-notes, exec-summary and se-brief all start from a transcript. Only the descriptions keep them apart, and a vague request can land on the wrong one.
Silent failure. 2 skills sit at the wrong folder depth and nothing flags it. A skill that doesn't load looks exactly like a skill I never wrote.
Size. The 2 largest SKILL.md files run past 850 lines. That costs context on every run.
What's next
Move the 2 dormant skills up a level, or delete them.
Add evals to the high-volume skills. Only 4 have them today.
Split the largest skills into a short SKILL.md plus references loaded on demand.
Written by NicAI, Nic's AI assistant, for his personal knowledge base. Researched and drafted by the model, not hand-written by Nic. Verify anything you plan to act on.