Bitbucket Pipelines

Copy these pipeline files and customize as needed.

AI Translation Pipeline for Bitbucket

This pipeline automatically extracts untranslated strings and translates them using Claude AI, then applies the translations back to your repository.

Workflow:

1. Triggered when English locale files change on main branch
2. Extracts new/changed strings using LocaleOps
3. Sends strings to Claude API for translation
4. Applies translated strings and creates a PR

Requirements:

• ANTHROPIC_API_KEY: Must be set in repository variables (secured)
• BITBUCKET_TOKEN: Repository Access Token with write permissions (secured)

Setup Instructions:

1. Create a Repository Access Token:
   - Go to Repository Settings > Security > Access tokens
   - Click "Create Repository Access Token"
   - Give it permissions: Repositories (Write), Pull requests (Write)
   - Copy the generated token
2. Add repository variables:
   - Go to Repository Settings > Pipelines > Repository variables
   - Add ANTHROPIC_API_KEY as a secured variable
   - Add BITBUCKET_TOKEN as a secured variable (paste the access token)
3. Ensure Pipelines is enabled for your repository

bitbucket-pipelines.yml

image: node:24

pipelines:
  branches:
    main:
      - step:
          name: AI Translation
          # Prevent concurrent runs, queue and fifo order
          concurrency-group: localeops-translation
          # Only run if English locale files were changed
          condition:
            changesets:
              includePaths:
                - "src/i18n/locales/en/**"
          script:
            # Install jq for JSON processing
            - apt-get update && apt-get install -y jq

            # Configure git for commits with access token
            - git config user.name "Pipelines Bot"
            - git config user.email "pipelines@bitbucket.org"

            # Step 1: Extract untranslated strings using LocaleOps
            - out=$(npx @localeops/localeops extract)
            - echo "[DEBUG] Extraction output:" $out

            # Step 2: Transform extracted data into flat array for translation
            # Collect all items needing translation (added or changed strings)
            # Output format: [{locale, filePath, resourcePath, text}, ...]
            - |
              to_translate=$(echo "$out" | jq '[to_entries[] | .key as $loc | .value[] | select(.type == "added" or .type == "changed") | {locale: $loc, filePath, resourcePath, text: (if .type == "added" then .value else .newValue end)}]')

            # Exit early if nothing needs translation
            - |
              if [ "$(echo "$to_translate" | jq 'length')" -eq 0 ]; then
                echo "Nothing to translate"
                exit 0
              fi

            # Step 3: Send strings to Claude API for translation
            - |
              response=$(curl -s https://api.anthropic.com/v1/messages \
                -H "Content-Type: application/json" \
                -H "x-api-key: $ANTHROPIC_API_KEY" \
                -H "anthropic-version: 2023-06-01" \
                -d "$(jq -n --arg items "$to_translate" '{
                  model: "claude-sonnet-4-20250514",
                  max_tokens: 8192,
                  messages: [{role: "user", content: "Translate each item to its target locale. Keep placeholders like {name}, {{count}}, %s intact.\n\nReturn JSON array: [{\"locale\": \"...\", \"filePath\": \"...\", \"resourcePath\": \"...\", \"value\": \"<translation>\", \"from\": \"<original>\"}]\n\nItems:\n\($items)"}]
                }')")

            - echo "[DEBUG] API Response:" $response

            # Step 4: Extract translated text from Claude's response
            # Remove markdown code blocks if present
            - |
              translated=$(echo "$response" | jq -r '.content[0].text' | sed 's/```json//g; s/```//g')

            - echo "[DEBUG] Extracted translations:" $translated

            # Step 5: Reorganize translations by locale for LocaleOps apply command
            # Transform from array to object: {locale: [{filePath, resourcePath, value, from}]}
            - |
              translations_by_locale=$(echo "$translated" | jq 'group_by(.locale) | map({ (.[0].locale): map({filePath, resourcePath, value, from}) }) | add')

            - echo "[DEBUG] Grouped by locale:" $translations_by_locale

            # Step 6: Apply translations and create PR
            # LocaleOps will use the configured git remote with access token
            - npx @localeops/localeops apply "$translations_by_locale"

          # Artifacts to preserve (optional)
          artifacts:
            - "src/i18n/locales/**"

definitions:
  # Optional: Add caching to speed up subsequent runs
  caches:
    npm: ~/.npm