Recipe · Power users

Turn the app into an AI prompt expander.

Type a few Vietnamese keywords describing a coding task. Press your hotkey. Get a complete, structured English prompt pasted into Claude Code, Codex, ChatGPT, or Claude Desktop — no manual prompt-writing, no copy-paste-translate dance.

Why this works

Contextual MT's outbound translation bindings let you pin a custom style instruction per hotkey. The LLM treats that instruction as authoritative — so instead of asking it to translate keywords literally, you can ask it to expand them into a full English prompt. Same hotkey UX as any other translation, just a different shape on the way out.

Works against both Contextual MT Cloud and any self-hosted backend you configure. No app update required.

Setup (3 minutes)

  1. Open Settings → Translation bindings → Outbound and add a new binding.
  2. Set Target language to English (en).
  3. Pick any modifier-rich hotkey that does not collide with your existing bindings. ⌥⇧P works (P for Prompt).
  4. Open the Custom style instruction field and paste the meta-prompt below.
  5. Save. The hotkey is live the instant Settings closes — try it from any text field.

Meta-prompt to paste

Tune the bullet list to match your stack. The shorter and more opinionated it is, the more consistent the expansions will be.

You are a prompt engineer for AI coding assistants (Claude Code, Codex, ChatGPT, Claude Desktop).

The input is a minimal Vietnamese keyword sketch of a coding task. Expand it into a complete English prompt that:

- Restates the task explicitly in one sentence
- Lays out a numbered plan of what the assistant should do
- Asks for code + tests + a brief explanation of trade-offs
- Notes reasonable assumptions about the tech stack if the keywords omit it
- Requests clarifying questions only if essential context is genuinely ambiguous

Output ONLY the final English prompt. No translation commentary, no preamble, no markdown fences.

Example: in → out

You type (Vietnamese)

fix bug login keychain mac swift, prompt password hiện lại mỗi lần update

App pastes (English prompt)

Debug a macOS Swift app where the login keychain prompts for the user password on every Sparkle update. Investigate why the Keychain ACL is re-validating on each release, and propose a fix that preserves the grant across notarized Developer ID releases.

Plan:
1. Inspect how the app currently calls SecItemAdd / SecItemCopyMatching.
2. Identify whether kSecAttrAccessGroup or a custom SecAccess ACL is in use.
3. Propose the minimal change needed for Keychain entries to be readable across Developer-ID-signed releases without prompting.
4. Provide patched Swift code + a Swift Testing unit test that pins the new ACL behaviour.

Assume: macOS 14+, Swift 6, Developer ID Team ID known. Ask if you need the bundle identifier.

Limitations of the workaround

  • Translate framing leaks through. The backend system prompt still tells the model it is translating. Your custom instruction overrides that most of the time, but occasionally the model will literal-translate a Vietnamese keyword instead of expanding it.
  • Same temperature as translation. Outbound bindings use a low-temperature prompt for fidelity. Prompt expansion benefits from a slightly higher temperature for variety; this will be tunable once v0.11 ships the dedicated mode.
  • No per-AI variants. The same expanded prompt flows to Claude Code, Codex, ChatGPT, and Claude Desktop. Each tool has slightly different prompt conventions; v0.11 will add an assistant-specific switch.

Coming in v0.11

A first-class Prompt Engineer mode with its own dedicated system prompt, a higher temperature, and a tech-stack context profile that gets prepended to every expansion. Per-AI variants (Claude Code / Codex / ChatGPT) so the output matches each assistant's prompting conventions.

Download the appRead the changelog