Claude skills for founders

What Claude skills are

A skill is a folder with a file called SKILL.md. The top of the file holds a name and a description of what the skill does and when to use it. Below that are instructions, written the way you would brief a capable new colleague. The folder can also hold scripts, templates and reference documents.

Claude doesn't read every skill in full. It loads them in stages, which Anthropic calls progressive disclosure, so you can switch on many skills without crowding the conversation:

Scripts are the part that makes skills different from a saved prompt. Claude runs them and reads only their output, so a calculation, a validation or a file conversion comes out the same every time. That's why the founder skills here hand all their arithmetic, from effective valuations to runway, to small Python scripts, and keep Claude for the reading and judgement.

Skills, prompts, projects and connectors

Anthropic also ships ready-made skills for Word, Excel, PowerPoint and PDF files, which claude.ai uses automatically when you ask for a document. The skills here are custom skills: you add them yourself.

How to install a skill

Download a skill's ZIP from its page, then follow the steps for wherever you use Claude. Skills don't sync between these three, so add a skill to each one you use. The commands below use the term sheet skill as the example.

mkdir -p ~/.claude/skills
unzip term-sheet-review.zip -d ~/.claude/skills/

curl -X POST "https://api.anthropic.com/v1/skills" \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2026-10-21" \
  -F "files[]=@term-sheet-review/SKILL.md;filename=term-sheet-review/SKILL.md" \
  -F "files[]=@term-sheet-review/references/clauses.md;filename=term-sheet-review/references/clauses.md" \
  -F "files[]=@term-sheet-review/scripts/term_sheet_math.py;filename=term-sheet-review/scripts/term_sheet_math.py"

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2026-10-21" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-opus-5-5",
    "max_tokens": 4096,
    "container": {
      "skills": [{ "type": "custom", "skill_id": "YOUR_SKILL_ID", "version": "latest" }]
    },
    "tools": [{ "type": "code_execution_20250825", "name": "code_execution" }],
    "messages": [{ "role": "user", "content": "..." }]
  }'

Write your own

Any repeatable task you explain to Claude more than twice is a candidate: your metric definitions, how you write board decks, your hiring scorecard. A minimal skill is one file:

---
name: board-deck-check
description: Checks a monthly board deck against the company's metric definitions and flags numbers that changed definition or don't reconcile. Use when preparing or reviewing a board deck or board pack.
---

# Board deck check

1. Read the deck and list every metric with its value and period.
2. Compare each definition with last month's deck, if it is attached.
3. Flag any metric whose definition changed since last month.
4. Reply with a table: metric, value, issue, suggested fix.

• The name: up to 64 lowercase letters, numbers and hyphens, the same as the folder name, and without the words "claude" or "anthropic".
• The description: up to 1,024 characters, in the third person, saying what the skill does and when to use it. It is all Claude sees when deciding, so include the words people actually use.
• Keep SKILL.md under 500 lines and move detail into reference files linked directly from it, one level deep.
• Put anything that must be exact, such as calculations and validation, in a script, and tell Claude to run it rather than read it.
• For uploads to claude.ai or the API, use only the fields name, description, license, compatibility, metadata and allowed-tools; any other field fails the upload.

Anthropic's authoring guide and the open specification cover the rest. The quickest way to start is to do the task once with Claude, then ask it to turn what you explained into a skill.

Use skills safely

A skill can direct Claude to run code and use tools, so Anthropic's advice is to install skills only from sources you trust and to read every file first, looking for network calls or file access that don't fit the skill's purpose. Skills that fetch content from the internet carry extra risk, because the fetched content could contain instructions.

The skills here are written to be easy to audit. Each page shows every file in full, the scripts use only Python's standard library, and none of them makes a network call or reads anything beyond the files you give it.