v23cc ·

A set of Claude Code slash commands that offload repetitive tasks to a local LLM — no cloud API calls, no tokens spent on boilerplate.

Instead of asking Claude to summarize a video or draft a commit message (burning context and API quota), v23cc pipes the work to a locally running model (e.g. Gemma via mlx-lm). Claude stays focused on what it's good at; the local model handles the grunt work.

What's included:

• /v23cc:model — manage which local model preset is active
• /v23cc:config — show config list
• /v23cc:youtube — fetch subtitles from a YouTube video and summarize them in any language
• /v23cc:commit — generate a git commit message from staged changes
• /v23cc:sync-docs — update README.md and CLAUDE.md using local LLM
• /v23cc:branch — create, switch, rename, track, or list git branches
• /v23cc:pr — generate a PR title and description from branch diff
• Atlassian MCP — search Jira and Confluence directly in Claude Code via a local MCP server

Requirements

• Node.js >= 18
• Python 3 — for the local LLM caller script (installed automatically to ~/.v23cc/)
• Local LLM server — an OpenAI-compatible endpoint running locally (e.g. mlx-lm on port 9000)
• yt-dlp — required for /v23cc:youtube (pip install yt-dlp)
• gh — required for /v23cc:pr (GitHub CLI)

Install

# Interactive (prompts global or local)
npx v23cc@latest

# Global (works in all projects)
npx v23cc@latest --global

# Local (current project only)
npx v23cc@latest --local

# Custom namespace (commands become /wp:commit, /wp:pr, etc.)
npx v23cc@latest --local --namespace wp

The --namespace flag lets you rename the command prefix to anything you like. The internal tool home (~/.v23cc/) stays the same regardless of namespace. Re-installing with a different namespace automatically removes the old one.

Notes:

• --local requires a .claude/ directory in the current directory. If one is not found, the install exits with an error — you are likely in the wrong directory, or should use --global.
• Running --local from $HOME is also an error, since that path resolves to the same location as --global.
• Multiple installs (e.g. one global, one or more local) are tracked in ~/.v23cc/config.json. Shared scripts and MCP files in ~/.v23cc/ are only removed when the last install scope is uninstalled.

Commands

Command	Description
/v23cc:model [list|use|add|remove]	Manage local LLM model presets
/v23cc:config	Show config list
/v23cc:youtube <URL> [--lang ko] [--percent 20]	Summarize a YouTube video using local LLM
/v23cc:commit [--max 72] [--no-prefix] [--only-msg] [--all]	Generate and commit using local LLM
/v23cc:sync-docs [--lines 100] [--keep "section name"]	Update README.md and CLAUDE.md using local LLM
/v23cc:branch [create|switch|rename|track|list|current]	Create, switch, rename, track, or list git branches
/v23cc:pr [--only-msg]	Generate a PR title and description using the local LLM
/v23cc:atlassian [init|status]	Set up Jira & Confluence credentials
/v23cc:jira <ISSUE-KEY> [--deep] [--note "..."]	Analyze a Jira issue and suggest implementation approach

Workflow

# Add a local LLM model preset
/v23cc:model add gemma mlx-community/gemma-4-e4b-it-4bit 9000

# Switch active model
/v23cc:model use gemma

# Summarize a YouTube video in Korean (default)
/v23cc:youtube https://youtube.com/watch?v=...

# Summarize in English, shorter output
/v23cc:youtube https://youtube.com/watch?v=... --lang en --percent 10

# Generate and commit (conventional prefix style by default)
/v23cc:commit

# Commit without conventional prefix
/v23cc:commit --no-prefix

# Print message only, no commit
/v23cc:commit --only-msg

# Stage all files then commit
/v23cc:commit --all

# Update README.md and CLAUDE.md
/v23cc:sync-docs

# Create a new branch and switch to it
/v23cc:branch create feature/my-feature

# Rename current branch
/v23cc:branch rename feature/new-name

# Rename another branch
/v23cc:branch rename old-name new-name

# Track and checkout a remote branch (auto-detects remote)
/v23cc:branch track feature/my-feature

# Track a branch from a specific remote
/v23cc:branch track upstream/feature/my-feature

# Switch to an existing local branch
/v23cc:branch switch main

# List local branches
/v23cc:branch list

# List all branches including remotes
/v23cc:branch list --all

# Print current branch name
/v23cc:branch current

# Generate PR title and description
/v23cc:pr

# Print PR message only, no PR creation
/v23cc:pr --only-msg

# Analyze a Jira issue for implementation approach
/v23cc:jira WPN-123

# Deeper analysis with source snippets
/v23cc:jira WPN-123 --deep

# Add extra context (use @file for file content)
/v23cc:jira WPN-123 --note "related to WPN-100, must avoid breaking the auth flow"

Direct shell usage

When your Claude token budget is low, run the scripts directly:

# Manage model presets
bash ~/.v23cc/bin/model.sh list
bash ~/.v23cc/bin/model.sh use gemma
bash ~/.v23cc/bin/model.sh add gemma mlx-community/gemma-4-e4b-it-4bit 9000
bash ~/.v23cc/bin/model.sh remove gemma

# Summarize a YouTube video
bash ~/.v23cc/bin/youtube.sh https://youtube.com/watch?v=...
bash ~/.v23cc/bin/youtube.sh https://youtube.com/watch?v=... --lang en --percent 10

# Generate and commit
bash ~/.v23cc/bin/commit.sh
bash ~/.v23cc/bin/commit.sh --no-prefix
bash ~/.v23cc/bin/commit.sh --only-msg
bash ~/.v23cc/bin/commit.sh --all

# Update README.md and CLAUDE.md
bash ~/.v23cc/bin/sync-docs.sh
bash ~/.v23cc/bin/sync-docs.sh --lines 80 --keep "Architecture"

# Create, switch, rename, track, or list branches
bash ~/.v23cc/bin/branch.sh create feature/my-feature
bash ~/.v23cc/bin/branch.sh switch main
bash ~/.v23cc/bin/branch.sh rename new-name
bash ~/.v23cc/bin/branch.sh rename old-name new-name
bash ~/.v23cc/bin/branch.sh track feature/my-feature
bash ~/.v23cc/bin/branch.sh track upstream/feature/my-feature
bash ~/.v23cc/bin/branch.sh list
bash ~/.v23cc/bin/branch.sh list --all
bash ~/.v23cc/bin/branch.sh current

# Generate PR title/description
bash ~/.v23cc/bin/pr.sh

# Analyze a Jira issue
bash ~/.v23cc/bin/jira.sh WPN-123
bash ~/.v23cc/bin/jira.sh WPN-123 --deep
bash ~/.v23cc/bin/jira.sh WPN-123 --note "extra context here"

# Show config
bash ~/.v23cc/bin/config.sh

Atlassian MCP (Jira & Confluence)

v23cc includes a local MCP server that integrates Jira and Confluence directly into Claude Code. No slash commands needed — just talk to Claude naturally.

Setup

1. Install and restart Claude Code:

   npx v23cc@latest --global
   # Restart Claude Code to load the MCP server

2. Configure your Atlassian credentials once:

   /v23cc:atlassian init
   

   Claude will ask for your domain, email, and API token one at a time.

   Generate an API token at id.atlassian.com/manage-profile/security/api-tokens.

Usage

Just ask Claude naturally — no commands to remember:

What you say	What happens
"search jira for ocr"	Searches Jira, groups results by Bug/Story/Task, writes v23cc/jira/<timestamp>.md
"search jira for payment bugs in project CORE"	Filtered by project and type
"search confluence for deployment guide"	Fetches pages, summarizes each via local LLM, writes v23cc/confluence/<timestamp>.md

Results are written as markdown files to v23cc/jira/ and v23cc/confluence/ in your project (git-ignored automatically).

Jira tool parameters

Parameter	Description	Default
query	Search text	required
project	Project key filter (e.g. CORE)	—
max	Max results	20
type	Issue type: Bug, Story, Task	—

Confluence tool parameters

Parameter	Description	Default
query	Search text	required
space	Space key filter	—
max	Max results	10

Note: Confluence summaries require a local LLM server running (configured via /v23cc:model). If no model is active, raw content is shown instead.

Uninstall

npx v23cc@latest --global --uninstall
npx v23cc@latest --local --uninstall

Legal

This tool uses yt-dlp to download subtitles from YouTube. Users are responsible for complying with YouTube's Terms of Service and applicable copyright laws in their jurisdiction.

License

MIT