Skill Detail

Pronounce Developer Jargon

Answers short pronunciation questions about developer tools, AI models, acronyms, and project names by using the say-it CLI to play local audio and return IPA, a plain-English respelling, source evidence, and contested alternatives.

Developer ToolsMulti-Framework
Developer Tools Multi-Framework Published
⭐ 39 GitHub stars
COPY SKILL INSTRUCTIONS (OPTIONAL)
npx skills add agentskillexchange/skills --skill pronounce-developer-jargon Copy
Uses the third-party skills CLI, not an ASE-owned installer. Check your agent’s compatibility. This copies instructions; complete the upstream tool setup below separately.
At a glance
Last updated
Sep 2, 2026
Quick brief

# Pronounce Developer Jargon

How it works

What this skill actually does

Use this skill when a user asks how to pronounce one developer term, project,
product, AI model, acronym, or researcher nameβ€”for example `kubectl`, `nginx`,
`Qwen`, `JEPA`, `GIF`, or `PostgreSQL`. It wraps the open-source `say-it` CLI and
its 1,900+ entry TSV dictionary. The tool plays the intended local pronunciation
through macOS `say`, Linux `espeak-ng`/`espeak`, or Windows PowerShell
`System.Speech`, rather than asking a generic speech engine to guess from the raw
spelling. Dictionary records include General American IPA, an English-like TTS
respelling, alternate readings, confidence labels, editorial notes, and a source
URL when an official FAQ, creator statement, or other useful citation exists.

Do not use this skill to narrate sentences or paragraphs. It is deliberately
scoped to one short technical name. Quote the target when invoking the CLI so
punctuation in names such as `C++` is passed literally.

## Instructions

1. Run `say-it –json “”` to inspect the canonical dictionary record
without playing audio. Never invent a source when `source_url` is empty.
2. Run `say-it “”` to play the primary reading three times. The default
also speaks recorded alternatives after an audible β€œor”.
3. Reply with the IPA, a readable stressed respelling, the source URL when
present, and a brief note when the confidence is `contested`.
4. If the user asks for the rival reading, run `say-it –alt “”`. Use
`–solo` when the user wants only the primary and `–why` for a text record.
5. When `in_dict` is false, say that the local dictionary has no curated record;
do not present the speech engine guess as creator-verified.

## Installation

Install the upstream CLI and dictionary. Its installer also places the upstream
`pronounce-word` skill into detected Claude Code, Codex, and Kiro skill folders:

git clone https://github.com/anzy-renlab-ai/pronounce.git
cd pronounce
./install.sh

GitHub CLI can install and pin the upstream skill directly for Copilot, Claude
Code, Cursor, Codex, Gemini CLI, or Antigravity:

gh skill install anzy-renlab-ai/pronounce pronounce-word --pin v2.28.1

To install this Agent Skill Exchange copy manually, clone the catalog and copy
the directory into the skill folder used by the agent runtime:

git clone https://github.com/agentskillexchange/skills.git
cp -R skills/skills/pronounce-developer-jargon ~/.agent-skills/pronounce-developer-jargon

### Optional Third-Party Installer

The `skills` npm package is maintained by a third party. Pin its version when
using it:

npm exec --package=skills@1.5.7 -- skills add agentskillexchange/skills --skill pronounce-developer-jargon