Troubleshooting¶
Use this page when a translation run succeeds unexpectedly, fails during configuration, or produces output that needs review.
Start Here¶
- Run a focused command first, such as
translate -l "ko" -md. - Add
-dfor console debug logs. - Add
-sto save debug logs under<root-dir>/logs/. - Run
co-op-reviewafter translation to check freshness, structure, and local links.
Configuration Errors¶
No Language Model Provider¶
Error:
Fix:
- Configure Azure OpenAI or OpenAI.
- Verify the variables are in the environment where the command runs.
- For local usage, put them in
.envat the project root.
See Configuration.
Image Translation Without Azure AI Vision¶
Error:
Fix:
- Add
AZURE_AI_SERVICE_API_KEY. - Add
AZURE_AI_SERVICE_ENDPOINT. - Or run a text-only command such as
translate -l "ko" -md.
Invalid Key or Endpoint¶
Symptoms can include 401, redacted permission errors, or endpoint access errors.
Fix:
- Confirm the key belongs to the same Azure resource as the endpoint.
- Confirm the resource supports Vision when using
-img. - Confirm Azure OpenAI deployment name and API version match your deployment.
- Run with debug logs:
translate -l "ko" -md -d -s.
No Files Were Translated¶
Common causes:
- The selected flags do not match your files.
- Existing translated files are already present.
- Source files are under excluded directories.
- The command is running from the wrong project root.
Checks:
Use --root-dir when the command is run outside the project root.
Unexpected Link Behavior¶
Link rewriting depends on selected content types:
-nbincluded: notebook links can point to translated notebooks.-nbexcluded: notebook links can remain pointed at source notebooks.-imgincluded: image links can point to translated images.-imgexcluded: image links can remain pointed at source images.
Run a full content translation when all internal links should prefer translated outputs:
Run link review after translation:
Markdown Rendering Issues¶
If translated Markdown renders incorrectly:
- Check that frontmatter starts and ends with
---. - Check that code fence counts match between source and translated files.
- Run
co-op-reviewto catch common structure issues. - Re-translate the specific file if the output was corrupted.
GitHub Action Ran but No Pull Request Was Created¶
If peter-evans/create-pull-request reports that the branch is not ahead of base, the workflow found no files to commit.
Likely causes:
- The translation run produced no changes.
.gitignoreexcludestranslations/,translated_images/, or translated notebooks.add-pathsdoes not match the generated output directories.- The translation step exited early.
Fixes:
- Confirm generated files exist in
translations/ortranslated_images/. - Confirm
.gitignoredoes not ignore generated outputs. - Use matching
add-paths:
- Temporarily add debug flags to the translate command:
- Confirm workflow permissions include:
Translation Quality¶
Machine translations may need human review. Use evaluate only when you want experimental quality scoring and low-confidence repair workflows.
Experimental
evaluate can use rule-based and LLM-based checks, and its scoring model and metadata behavior may change. Keep it out of required CI gates unless your workflow is prepared for changes.
For deterministic CI checks, use co-op-review instead.