vivadicta cli skills

Agent skills for VivaDicta for Mac's command-line interface (vivadicta). Drop them into any stdio agent - Claude Code, Cursor, Codex, Goose, Warp etc - so the agent knows when and how to call the CLI for the user's dictation workflows.

Official pack maintained alongside the vivadicta CLI.

Install the full pack globally (recommended):

npx skills add n0an/vivadicta-cli-skills --global

Install a single skill for a specific agent:

npx skills add n0an/vivadicta-cli-skills --skill vivadicta-transcribe-flow --agent claude-code

Preview what's in the pack without installing:

npx skills add n0an/vivadicta-cli-skills --list

Manage installed skills:

skills list             # see what's installed
skills update           # refresh to latest
skills remove           # remove a skill

The pack installs prose-only SKILL.md files; there's no runtime. Updating to a new skill-pack version is as lightweight as git pull on a dotfile.

Before these skills are useful, make sure the CLI itself is installed and reachable:

# Install the VivaDicta app (notarized DMG) from https://vivadicta.com first, then:
brew install n0an/tap/vivadicta
vivadicta recent 5

Write skills (vivadicta-transcribe-flow, vivadicta-search-and-rewrite) also require the VivaDicta UI app to be running, since writes route through a Unix socket into the running app.

Reference skill: canonical flags, output formats, env vars, exit codes, TTY behavior, and a map of every subcommand.

Use when:

• You need the correct vivadicta command, flag, or output format.
• You need to interpret an exit code or error payload.
• You want the discovery cheat sheet for the 11 subcommands.

Example:

What vivadicta command lists my last 20 transcriptions as JSON and pipes the titles through jq?

End-to-end workflow for transcribing a local audio file or a YouTube URL into the user's history. Handles input-shape detection, blocking vs --async, progress output, and error recovery.

Use when:

• The user wants to transcribe an audio file (m4a, mp3, wav, etc.).
• The user wants to transcribe a YouTube URL.
• You need to decide between blocking and async, or recover from a timed-out / failed transcription.

Example:

Transcribe ~/Downloads/standup.m4a and give me the result.

Find a past transcription via natural-language query or the latest keyword, then rewrite it with an AI preset (summary, action points, email, translation, etc.).

Use when:

• The user wants to turn an existing dictation into action items, an email, a summary, or any other preset output.
• You need to resolve natural-language queries ("yesterday's call with sam") to a specific transcription.
• The user asks about available presets or gets an ambiguous-preset error.

Example:

Rewrite yesterday's call with Sam as action items.

Export a transcription to Obsidian-ready markdown (YAML frontmatter + body) and pipe it into a vault, daily note, Raycast action, or cron digest.

Use when:

• The user wants a dictation record as a markdown file in their Obsidian vault.
• You need to append a transcript to a daily note or copy it to the clipboard.
• You're building a shell / Raycast / cron automation that consumes transcripts as markdown.

Example:

Export my last dictation to ~/vault/transcripts as a dated markdown file.

Tagged milestones (v1.0.0, v1.1.0, …) for users who want to pin. The default branch (main) always reflects the latest guidance; day-to-day edits land without a tag bump.

To pin to a specific version:

npx skills add n0an/vivadicta-cli-skills@v1.0.0

• VivaDicta for Mac - the app itself, required for the CLI to do anything meaningful.
• n0an/vivadicta-cli - the thin bash shim distributed via Homebrew that becomes vivadicta on $PATH.
• n0an/vivadicta-mcp - the parallel shim for the MCP server (vivadicta-mcp), a separate install path for MCP clients.
• n0an/homebrew-tap - formula source for both shims.

MIT. See LICENSE.