Claude Code Skills by smorrel1

Claude Code Skills

A collection of custom skills for Claude Code, Anthropic’s CLI tool for AI-assisted software development.

What are Skills?

Skills are reusable modules that extend Claude Code’s capabilities. Each skill provides:

A SKILL.md file with instructions Claude reads to understand how to use the skill
Scripts and utilities the skill uses to perform tasks
Documentation for setup and usage

For content creators

One recording becomes a week of posts, locally, no subscription:

interview-transcript: YouTube or X video in, cleaned speaker-labelled transcript out (Kindle delivery optional)
repurpose: transcript in, X thread + LinkedIn post + newsletter section out
humanizer: every draft checked against 400+ AI-writing tells before you see it

curl -fsSL https://smorrel1.github.io/claude-code-skills/install.sh | bash -s -- interview-transcript repurpose humanizer

See it on a real episode: repurpose/examples holds the unedited thread, LinkedIn post and newsletter section produced from the David Senra x Travis Kalanick interview.

Install in one command

Works with skills.sh tooling in Claude Code, Codex CLI and compatible agents:

npx skills add smorrel1/claude-code-skills@ce-to-510k     # FDA 510(k) porting, written from a real clearance
npx skills add smorrel1/claude-code-skills@after-meeting  # transcript -> summary -> follow-up drafts
npx skills add smorrel1/claude-code-skills@email          # multi-account Gmail with draft-safety guards

Or install everything with one command (non-destructive, symlinks into ~/.claude/skills):

curl -fsSL https://smorrel1.github.io/claude-code-skills/install.sh | bash

These skills come from daily use running a regulated medical-AI company solo: the ce-to-510k skill was written from an actual FDA clearance, the email and meeting skills run a real CEO workload every day. Feedback and issues welcome.

Available Skills

Skill	Description	Setup Required
apple-notes	Export Apple Notes to markdown	None (macOS only)
gmail	Read, search, draft Gmail emails	OAuth setup
gsheet	Read/write Google Sheets	OAuth setup (uses Gmail creds)
gcal	Google Calendar CRUD operations	OAuth setup (uses Gmail creds)
docx-editor	Word document character normalization	None
interview-transcript	YouTube/X.com transcript to Kindle EPUB	yt-dlp + Calibre
monthly-report	Monthly board report generation	Custom paths + OAuth
ce-to-510k	Port CE Mark / UKCA technical files to FDA 510(k) for AI/ML SaMD	None
docx-delta	Compare two Word documents, redline with tracked changes	python-docx
cmux	Manage many parallel Claude Code sessions in cmux workspaces	cmux
todo-agent	Two-way channel between an Apple Notes To Do list and your agents	macOS + cmux

Installation

1. Clone or Copy Skills

# Option A: Clone the repo
git clone https://github.com/YOUR_USERNAME/claude-code-skills.git ~/.claude/skills

# Option B: Copy individual skills
cp -r path/to/skill ~/.claude/skills/

2. Install Python Dependencies

Most skills auto-install dependencies on first run, but you can pre-install:

pip install google-auth google-auth-oauthlib google-api-python-client

3. Configure OAuth (for Google skills)

For Gmail, Google Sheets, and Google Calendar skills:

Go to Google Cloud Console
Create a new project or select existing
Enable the required APIs:
- Gmail API
- Google Sheets API
- Google Calendar API
Create OAuth 2.0 credentials (Desktop app type)
Download credentials.json and place in ~/.claude/skills/gmail/
First run will open browser for authorization

Skill Details

Apple Notes

Export all Apple Notes to markdown files. Fast Python-based exporter that reads directly from the Notes SQLite database.

Requirements: macOS with Apple Notes

Usage:

python3 ~/.claude/skills/apple-notes/scripts/export_notes.py
python3 ~/.claude/skills/apple-notes/scripts/export_notes.py --output ~/Documents/notes

Output: Organized markdown files by folder with date-prefixed filenames.

Gmail

Read, search, and draft emails using the Gmail API.

Setup:

Place credentials.json in ~/.claude/skills/gmail/
Run any command to trigger OAuth flow

Usage:

# Search emails
python3 ~/.claude/skills/gmail/scripts/gmail_utils.py search --query "from:john@example.com" --max 5

# Read specific email
python3 ~/.claude/skills/gmail/scripts/gmail_utils.py read --id "message_id"

# Create draft
python3 ~/.claude/skills/gmail/scripts/gmail_utils.py draft --to "user@example.com" --subject "Hello" --body "Message body"

# Reply to thread
python3 ~/.claude/skills/gmail/scripts/gmail_utils.py draft --reply-to "message_id" --body "Reply text"

Google Sheets

Read and write Google Sheets data.

Setup: Uses credentials from Gmail skill. Requires separate OAuth authorization for Sheets scope.

Usage:

# List sheets
python3 ~/.claude/skills/gsheet/scripts/read_gsheet.py "SPREADSHEET_URL" --list-sheets

# Read data
python3 ~/.claude/skills/gsheet/scripts/read_gsheet.py "SPREADSHEET_URL" --sheet "Sheet1" --rows 10

# Update cell
python3 ~/.claude/skills/gsheet/scripts/read_gsheet.py "SPREADSHEET_URL" --sheet "Sheet1" --edit "A1" --value "New Value"

Google Calendar

CRUD operations for Google Calendar.

Setup: Uses credentials from Gmail skill. Requires separate OAuth authorization for Calendar scope.

Usage:

# List calendars
python3 ~/.claude/skills/gcal/scripts/cal_utils.py calendars

# List upcoming events
python3 ~/.claude/skills/gcal/scripts/cal_utils.py list --days 7

# Create event
python3 ~/.claude/skills/gcal/scripts/cal_utils.py create --summary "Meeting" --start "2025-01-15 10:00"

# Check availability
python3 ~/.claude/skills/gcal/scripts/cal_utils.py availability 2025-01-15

Interview Transcript

Download and transcribe interviews/talks from YouTube and X.com (Twitter), clean them up, and convert to Kindle-ready EPUBs with cover images and clickable table of contents.

Features:

Auto-detects YouTube vs X.com URLs
YouTube-first search for X.com posts (faster subtitle extraction vs whisper transcription)
Mechanical filler word removal + LLM-based transcription fixing and section headers
EPUB with cover image and clickable TOC
Send to Kindle, upload to Google Drive for shareable link

Setup:

brew install yt-dlp calibre
pip install jinja2
# Optional (X.com fallback): brew install whisper-cpp ffmpeg

Usage:

# Download transcript
python3 ~/.claude/skills/interview-transcript/scripts/youtube_transcript.py --url "https://youtube.com/watch?v=VIDEO_ID"

# Clean filler words
python3 ~/.claude/skills/interview-transcript/scripts/youtube_transcript.py --clean transcript.txt

# Convert to EPUB with cover
python3 ~/.claude/skills/interview-transcript/scripts/youtube_transcript.py --to-epub structured.txt \
  --title "Interview Title" --author "Speaker" --youtube-cover "https://youtube.com/watch?v=VIDEO_ID"

Or just use it as a Claude Code skill: /interview-transcript <URL>

DOCX Editor

Utilities for working with Word documents, particularly fixing character encoding issues that break find-and-replace.

Usage:

# Normalize special characters
python3 ~/.claude/skills/docx-editor/scripts/fix_docx_chars.py "Text with "smart quotes""

Monthly Report

Note: This is a personal workflow skill. Fork and customize paths in config.py for your own use.

Generates monthly board reports by:

Exporting Apple Notes
Consolidating meeting minutes and transcripts
Processing emails
Synthesizing into report format

External Dependency: This skill expects Fireflies.ai meeting transcripts to be downloaded to a local directory. Set up a cron job to periodically sync transcripts from Fireflies to your configured FIREFLIES_DIR path.

Example cron setup:

# Edit crontab
crontab -e

# Add jobs to download and format transcripts every 5 minutes
*/5 * * * * /path/to/fireflies_downloader/run_script.sh fireflies_downloader.py
7,12,17,22,27,32,37,42,47,52,57 * * * * /path/to/fireflies_downloader/run_script.sh format_transcripts.py

Example run_script.sh wrapper (handles conda/venv activation):

#!/bin/bash
echo "=== Job started at $(date) ===" >> "$HOME/fireflies_downloader/cron_log.txt"
source "$HOME/.bash_profile"
conda activate your-env  # or: source venv/bin/activate
cd "$HOME/fireflies_downloader"
python "$1" >> "$HOME/fireflies_downloader/cron_log.txt" 2>&1
echo "=== Job ended at $(date) with status $? ===" >> "$HOME/fireflies_downloader/cron_log.txt"

See monthly-report/SKILL.md for detailed workflow documentation.

ce-to-510k

Port CE Mark / UKCA medical device technical documentation into an FDA 510(k) submission package for AI/ML SaMD (Software as a Medical Device). Distilled from a real 2023 AI/ML SaMD 510(k) clearance including the full Additional Information Request and Interactive Feedback cycle. Generalises across imaging, signal-processing, clinical decision support and other AI/ML domains.

What it produces:

Planning document for the 510(k) campaign
Predicate device research and analysis (using the FDA 510(k) database)
Pre-Submission (Q-Sub) request when appropriate
Mapped 510(k) eCopy package structure (sections 00 to 21) with content gap list
Strategic recommendation on whether to pursue a Pre-Sub

Highest-value content: references/06-fda-ai-ml-deficiencies.md, a categorised catalogue of the AI/ML deficiency question patterns FDA uses during AINN and Interactive Feedback rounds, with a 16-item pre-empt checklist you can paste into your draft.

Usage: invoke via the Skill tool or by mentioning any of: 510(k), FDA submission, Q-Sub, pre-sub, predicate device, substantial equivalence, CE to FDA, MDR to FDA, AI/ML 510k. Give Claude the path to your CE Mark / UKCA technical file root and your device details; the skill works through eight phases and produces deliverables on disk.

Not suitable for: De Novo, PMA, IVD CLIA, EUA, IDE, drug applications.

Disclaimer: procedural guidance only, not legal or regulatory advice. Verify all FDA guidance citations against current FDA publications before submission.

See ce-to-510k/README.md and ce-to-510k/SKILL.md for details.

Creating Your Own Skills

Create a folder under ~/.claude/skills/your-skill-name/
Add a SKILL.md file with:
- YAML frontmatter with name and description
- Usage instructions for Claude
Add scripts in a scripts/ subdirectory
Claude will automatically discover and use skills based on context

SKILL.md Template

---
name: my-skill
description: Brief description that helps Claude know when to use this skill.
---

# My Skill

## Usage

\`\`\`bash
python3 scripts/my_script.py [options]
\`\`\`

## Examples

...

Security Notes

Never commit credentials: The .gitignore excludes credentials.json, token.json, and similar files
OAuth tokens expire: Tokens auto-refresh, but if you see auth errors, delete token.json and re-authenticate
Scope permissions: Each Google skill requests only the scopes it needs

License

MIT License - feel free to use, modify, and share.

Local overrides (private data stays out of git)

Skills that carry user-specific configuration (accounts, addresses, personal file paths, private workflow rules) keep the public SKILL.md generic. The private version lives in SKILL.local.md in the same skill directory, which is gitignored (*.local.md). Each participating SKILL.md opens with a pointer instructing the agent to read SKILL.local.md INSTEAD when it exists, so the private version takes full precedence at load time while never reaching the repository. To personalise a skill: cp SKILL.md SKILL.local.md inside the skill folder and edit the local copy.

docx-delta

Compare two Word documents and get a delta view. Replaces Word’s Review Compare feature and LibreOffice’s headless compare, which is unreliable to automate.

D=~/.claude/skills/docx-delta/scripts/docx_delta.py
python3 $D old.docx new.docx                  # counts only
python3 $D old.docx new.docx -o delta.docx    # redline with real tracked changes
python3 $D old.docx new.docx -o delta.html    # side by side, for reading
python3 $D old.docx new.docx --md             # unified text diff

The .docx output carries genuine w:ins and w:del markup, so Word’s Review pane shows the changes and Accept and Reject work.

Paragraphs are aligned first, then words are compared inside each changed paragraph, so a single edited figure is visible in a long paragraph.

What it compares: body paragraphs and table cell text. Footnotes, endnotes, comments, headers and footers are counted and named in the output, not compared. Images and styling are not compared.

Requires: python-docx (pip install python-docx). No Word, no LibreOffice.

cmux

Manage dozens of parallel Claude Code sessions running in cmux workspaces: find which session owns a topic, hand work to another session and prove it started, restart every session by its own id after an update, and roll up what each one is doing.

python3 ~/.claude/skills/cmux/scripts/send_task.py --workspace "grant applications" "task text"
python3 ~/.claude/skills/cmux/scripts/restart_sessions.py           # verify, by session id
python3 ~/.claude/skills/cmux/scripts/refresh_registry.py           # regenerate the registry

send_task.py exists because cmux send fills the input box without submitting it: a multi-line paste arrives with a “paste again to expand” hint, so the first Enter only expands it and the task stays there looking delivered. It sends one short line pointing at a spooled brief, presses Enter until the target is running, and exits non-zero if it never started. Workspaces are addressed by NAME; numbers drift on every reorder and have misrouted briefs.

Copy references/registry-durable.example.md to registry-durable.md and fill in your own workspaces. The generated registry.md and the task spool are gitignored.

todo-agent

Turns an Apple Notes To Do list into a two-way channel with your agents. Any bullet that opens with claude (the @ is optional) is picked up within five minutes, handed to a chief-of-staff session, and ticked with the outcome. Agents write follow-ups back near the TOP of the note, because a long list is only read down to the first page.

python3 ~/.claude/skills/todo-agent/scripts/todo_agent.py scan --dry-run
python3 ~/.claude/skills/todo-agent/scripts/todo_agent.py prepend --heading "Call follow-ups" --items items.json
python3 ~/.claude/skills/todo-agent/tests/test_trigger.py

Most of the code exists to handle undocumented Apple Notes behaviour: every whole-body write multiplies font sizes by 1.2 and it compounds, a note’s name comes from its first line so copies shadow the original, and Notes fragments your markup so markers must be matched against stripped text. Writes go to a note pinned by Core Data id and are backed up first.

Requires macOS (Apple Notes) and cmux. Note bodies, delivery state and logs are gitignored.