What This Skill Does

One line: translate-cli is the most underrated productivity skill in the OpenClaw ecosystem. It lets you translate arbitrary text, stdin streams, single files, or entire directories of glob-matched files — with multi-provider support, preset configurations, custom prompt templates, and dry-run validation.

At SFD Lab, we use it to batch-translate articles to English and Traditional Chinese daily, enabling trilingual publishing on smallfiredragon.com.

Installation

clawhub install translate-cli

Verify installation:

translate --version
# translate-cli v1.x.x

Basic Usage

Translate Text

# Direct text translation
translate "深度学习入门指南"

Specify target language
translate --lang en "深度学习入门指南"
→ Introduction to Deep Learning

Stdin Input

# Pipe input
echo "Context window technology" | translate --lang zh
From file
cat article.txt | translate --lang zh-TW > article_tw.txt

Translate a Single File

translate --file ./docs/readme.md --lang en --out ./docs/readme_en.md

Batch Translation (Glob)

# Translate all markdown files in a directory
translate --file "./content/*.md" --lang en --out-dir ./content_en/

Output:
✓ article-1.md → article-1.md (en)
✓ article-2.md → article-2.md (en)
✓ 12 files translated in 18.3s

Provider Selection

translate-cli supports multiple AI providers. Configure in .openclaw/config.toml:

[translate]
default_provider = "openai"    # openai | anthropic | deepseek | gemini
default_model = "gpt-4o-mini" # budget-friendly default

Override per-command:

translate --provider anthropic --model claude-haiku "Hello world"
translate --provider deepseek "Quantum computing basics"

Pro tip from our experience: DeepSeek handles Chinese technical documentation best for zh→en. For en→zh, GPT-4o-mini produces more natural-sounding output. Pick your provider by direction, not just cost.

Presets — The Killer Feature

Most people skip this. Presets let you codify language pairs, models, and prompt styles:

# List existing presets
translate presets list

Create a preset
translate presets add tech-zh-to-en 
  --from zh 
  --to en 
  --provider openai 
  --model gpt-4o 
  --prompt "Translate Chinese technical content to English. Preserve all code blocks, technical terms, and markdown formatting exactly."
Use the preset
translate --preset tech-zh-to-en --file article.md

Our three core presets at SFD Lab:

• sfd-zh-to-en: Chinese tech articles → English, code blocks preserved
• sfd-zh-to-tw: Simplified → Traditional Chinese, Taiwan terminology
• sfd-product: Product copy, SEO keyword structure preserved

Dry-Run Validation

# Validate without actually translating
translate --preset sfd-zh-to-en --file "./content/*.md" --dry-run

Output:
[dry-run] Would translate 12 files
[dry-run] Provider: openai / gpt-4o
[dry-run] Estimated tokens: ~45,000
[dry-run] Estimated cost: ~$0.09

Always dry-run before batch jobs. Avoid surprise API bills.

TOML Configuration File

[settings]
default_lang = "en"
concurrency = 3     # parallel file translations
retry_on_fail = 2   # retries on failure
[presets.blog]
from = "zh"
to = "en"
provider = "anthropic"
model = "claude-haiku-3-5"
prompt_template = "./prompts/blog-translation.txt"
[output]
preserve_structure = true
append_lang_suffix = true  # article.md → article_en.md

Common Pitfalls

Issue 1: Code Blocks Get Mangled

Cause: Default prompt doesn't emphasize format preservation. Fix: Add to your preset prompt: "Preserve all code blocks exactly as-is. Do not translate code or technical identifiers."

Issue 2: Traditional Chinese Output Contains Simplified Characters

Fix: Explicitly add to prompt: "Use Traditional Chinese (Taiwan standard). Do not use Simplified Chinese characters or Mainland China terminology."

Issue 3: Batch Job Stops Mid-Way

Cause: API rate limiting. Fix: Lower concurrency from default 5 to 2, add retry_on_fail = 3 in config.toml.

Skill Combinations

• translate-cli + summarize: Summarize long articles first, then translate summaries — 80% cost reduction
• translate-cli + humanizer: Run translated content through humanizer to remove AI flavor before publishing
• translate-cli + smart-web-scraper: Scrape foreign-language pages → translate → store as Chinese knowledge base

SFD Lab Note

We use translate-cli for three things: ① Daily trilingual publishing on smallfiredragon.com (zh → en + zh-TW); ② Content processing by our 小春蚕 agent for BACAKU (Japanese/English → Chinese); ③ Security document analysis where 小猎鹰 needs English audit reports translated before review.

Over a month, this skill saves roughly 30 minutes of manual translation daily. For any content-driven project, this is a permanent install.