Context Extraction

The loop

Commands, filters and the file format: the crowdin-cli skill's context reference. Read it before the download; its --status filter does not mean "needs context". With a JSONL file already in hand, start at step 3.

1. Coverage (optional): crowdin context status. Done when you can tell the user how many strings have manual, AI and no context.
2. Download: crowdin context download, with no --status. Done when crowdin-context.jsonl exists and holds one line per string in scope.
3. Fill ai_context following the rest of this skill. Done when every line that needs context (see "Strings that need context") has a value, and every changed line passes the validity checklist.
4. Preview: crowdin context upload --dryrun. Done when it lists exactly the strings you filled.
5. Upload: crowdin context upload, then crowdin context status. Done when the upload lists the same strings the dry run did; report how many, and the new coverage, to the user.

Information sources (priority order)

For each string, use these JSONL fields in order - stop when you have enough to write a confident ai_context: