[Jeremy Rose (jeremyrose-viam)]

tl;dr for human readers

The image_uri comment in ml_training.proto is parsed into a broken link and two broken images by TypeDoc. The backticks avoid this.

Summary

• The RegisterCustomTrainingContainerRequest.image_uri comment contains docker.io/library/<image>:<tag> as plain prose. The TypeScript SDK's TypeDoc build autolinks docker.io/library/ as http://docker.io/library/ (non-HTTPS target) and parses <image> as an HTML <img> tag with no src or alt. This was flagged by the TS SDK's scheduled Link Check workflow (run, auto-filed as DOCS-5035): 3 errors on classes/mlTrainingApi.RegisterCustomTrainingContainerRequest.html.
• Wrapping the reference in backticks makes it a code span, so TypeDoc renders it literally.
• Same class of fix as viamrobotics/api PR 897 (PackageConfig.Name).

Test plan

• Rebuilt the TS SDK's generated proto bindings from the pinned API version (v0.1.596) with this one-line change, then ran typedoc + htmltest: 0 errors across 1213 documents (3 errors without the change). The field now renders as <code>docker.io/library/&lt;image&gt;:&lt;tag&gt;</code>.
• Python (Sphinx autoapi+napoleon, nitpicky -n): built the Python SDK docs before and after, applying the backticked text to the generated ml_training_pb2.pyi docstring. A separate buf generate run with the SDK's mypy plugin confirmed the generated docstring is the proto comment verbatim, so the hand edit matches real output (the only diff between generating from the unfixed and fixed proto is that line). No new warnings, and 4 existing py:class reference target not found warnings for this field go away. The field now renders in full, docker.io/library/<image>:<tag>; before, napoleon cut it off after <image>.

Go and C++ are expected to be unaffected, as in PR 897: Go renders field comments as raw source text, and C++ bindings don't carry comments.

🤖 Generated with Claude Code