The context commands feedback

[andrii-bodnar]

The 4.14.0 release adds commands to manage string context for your Crowdin project from the CLI. The context command and its subcommands let you:

• context download — export strings to a local .jsonl file (with filters by file, label, context status, date, branch)
• context upload — push enriched context from a local file back to Crowdin
• context status — see how much of the project has context (overall or per file)
• context reset — remove AI-generated context while keeping manual context

This fits well with Context Enrichment with AI Agents: download → enrich with an AI agent (e.g. using Crowdin Skills) → upload.

We’d love your feedback on the commands, the JSONL format, filtering options, and how this fits into your workflows (CLI, CI/CD, or agents). Share your experience, questions, or suggestions in the replies ✨

[KubaGrzeszczak]

Any suggestion on how this can be automated and included into our scheduled github actions? 🙏

[andrii-bodnar]

Hey @KubaGrzeszczak 👋

Good question. The thing that makes this different from a "normal" Crowdin CI job: the actual enrichment is done by an AI agent, not by the CLI itself. So your scheduled workflow needs a headless/non-interactive agent CLI available in the runner (Claude Code, or any coding-agent CLI that can load the Crowdin Skills). context download / upload are just the wrappers around that.

End to end:

1. Checkout your repo - the agent needs the source to see how each key is used.
2. Install CLI + Skills - npx skills add crowdin/skills (adds crowdin-context-cli + context-extraction). I think this step can be done separately, with the skills committed to the repository.
3. Download only what's missing - crowdin context download --status empty keeps each run incremental.
4. Let the agent enrich - run your agent headlessly against the downloaded JSONL; the context-extraction skill guides how it writes descriptions.
5. Upload - crowdin context upload.
6. (optional) crowdin context status for a coverage log.