Context is one of the most powerful ways to improve translation accuracy. Without it, translators are left to guess the meaning of short strings, UI labels, and ambiguous phrases - leading to inaccurate translations, longer review cycles, and frustrated localization teams.
The new Crowdin CLI context management commands make it easy to manage string context at scale, directly from your terminal.
In this post, we'll walk through a practical workflow: download your strings, enrich them using an AI agent with the help of Crowdin Skills, and push the result back to Crowdin.
The Workflow
A string like Name or Post means something very different depending on where it appears in your app. Is it username or post title? Is it a label or a button? Without context, translators must guess - and guesses lead to mistakes.
The context enrichment workflow with Crowdin CLI is straightforward:
- Download strings context to a local file
- Enrich the context using an AI agent
- Upload the enriched context back to Crowdin
Let's walk through each step.
Step 0: Install Crowdin Skills
Crowdin Skills is a collection of agent skills that teach your AI agent how to work with Crowdin - including how to correctly fill string context.
Install all Crowdin skills with a single command:
npx skills add crowdin/skills
This installs two skills relevant to context enrichment:
crowdin-context-cli- documents thecontext downloadandcontext uploadCLI commands so the agent understands the full workflow and the JSONL format.context-extraction- teaches the agent how to write good context descriptions: which strings need context (ambiguous short words, plurals, UI labels with inline formatting), how to write concise 1–3 sentence descriptions that explain the UI element type and placement, and what not to touch in the file.
Step 1: Download Strings Context
The context download command exports your project's strings to a local .jsonl file, ready for AI enrichment.
crowdin context download
By default, this creates a crowdin-context.jsonl file in the current directory. You can also customize the output path with the --to option. You don't need to execute this command manually. There is no need to execute this command manually. The agent can do it for you when you ask it to.
JSONL is a line-delimited format that is easy to parse and write. It is a good format for AI agents to work with. There are many IDE extensions available, such as JSONL Gazelle for VS Code. It allows you to view the JSONL file content in a more readable format.
Filtering Strings
You rarely need to enrich all strings at once. The command supports a range of filters so you can focus on exactly what matters:
| Filter | Description | Example Command |
|---|---|---|
| By file | Process strings from specific source files | crowdin context download -f "src/ui/**" |
| By label | Target strings with specific labels | crowdin context download --label needs-context |
| By context status | Only download strings that are missing context | crowdin context download --status empty |
| By date | Only strings created after a specific date | crowdin context download --since 2026-01-01 |
| By branch | Only strings from a specific branch | crowdin context download -b main |
These filters are especially useful when running context enrichment as part of a CI/CD pipeline or when incrementally updating context for new strings.
Step 2: Enrich Context with AI Agents
Your AI agent already knows your codebase, so it can write the best description possible for your strings. Once you have your crowdin-context.jsonl file, it's time to let the agent do the heavy lifting.
Just ask your agent to enrich the context. It will use the Crowdin CLI to download and upload context, and the context-extraction skill to write good descriptions.
Step 3: Upload Enriched Context
Once your AI agent has enriched the crowdin-context.jsonl file, upload it back to Crowdin with the context upload command:
crowdin context upload
Use the --from option if you used a custom path with --to during download.
Additional Commands
Check Context Coverage
The context status command gives you a quick overview of how much of your project has context filled in:
crowdin context status
✔️ Fetching project info
Context Status for Project "Demo" (ID: 00000)
Total strings: 3484
With AI context: 50 (1.44%)
Without AI context: 3434 (98.56%)
With manual context: 3484 (100.00%)
Use --by-file to see a breakdown per source file - useful for spotting which files need the most attention.
Reset AI-Generated Context
The context reset command removes AI-generated context from strings while preserving any manually written context:
crowdin context reset --all
You can scope the reset to specific files, labels, or branches using the same filtering options available on the other commands. The --all flag is required when no filter is specified, as a safety measure against accidental bulk resets.
Summary
This post showed how to enrich your project's string context using Crowdin CLI and an AI agent. The workflow: install Crowdin Skills, download strings to a local JSONL file, ask your agent to fill in context, then upload the result back to Crowdin. Use context status to check coverage and context reset to remove AI-generated context when needed.