[leehack]

Summary

• Docs-only clarification for #1836: on native .litertlm, getMetadata() and engine.chatTemplate(...) show the file-name family template that drives Dart-side format detection and parsing. The runtime renders that template only for text-only Qwen3 requests. Qwen3.5, Gemma and every request with media render the bundle's embedded template, so the native prompt can differ.
• Updates website/docs/getting-started/model-families.md (the .litertlm bullet and the "To see what a model will receive" note) and skills/llamadart-tool-calling/SKILL.md.
• Refs LiteRT-LM .litertlm: separate the reported/rendered bundle chat template from the Dart-side format and capability template #1836. A code fix would change format detection and capability results. The issue is now the post-release design issue for that fix; see the findings comment there.

Production-readiness scope

• User-facing scope: the docs say accurately what chatTemplate()/metadata show for native LiteRT-LM bundles.
• Supported platforms/paths: docs-only (website and the published agent skill).
• Unsupported or intentionally unavailable paths: N/A. Runtime behavior is unchanged.
• Out of scope / follow-ups: report the rendering template without changing parsing: #1836.

Completeness checklist

• Declared scope is fully implemented, or explicitly reduced above.
• Unsupported platform/option combinations fail loudly with actionable diagnostics or are clearly documented as unavailable. (This PR documents the gap.)
• Public API docs, README/website docs, examples, support matrices, and changelog entries are updated where relevant. (No changelog bullet: docs clarification only.)
• Regression coverage covers the original issue plus key negative/version-skew paths where relevant. N/A: no behavior change.
• Public capability changes update the validation coverage registry with a pack or a reasoned, issue-linked exemption; inventory checks pass (or N/A with scope). N/A: no capability change.
• Security/privacy review completed: no secrets, bearer tokens, signed URLs, or raw secret-bearing paths leak through logs, cache keys, metadata, errors, or snapshots.
• Follow-up work that is useful but not required for this PR is tracked in GitHub Issues, or explicitly marked "None" above.

PR type guidance

• Feature PR: N/A
• Bugfix PR: N/A
• Docs-only PR: no runtime behavior change. I checked the wording against litert_lm_service.dart (useTextTemplate/promptTemplate) and the ModelParams.chatTemplate Dartdoc, and verified it on real bundles (Review Notes). The docs build and link check pass.

Test Plan

• dart run tool/prepare_workspace.dart
• dart format --output=none --set-exit-if-changed . (0 changed)
• dart analyze: N/A, Markdown-only diff (CI runs it)
• dart test -p vm -j 1 --exclude-tags local-only: N/A, ran the Markdown-selected suites instead (below)
• dart test -p chrome --exclude-tags local-only: N/A, docs-only
• coverage-lib matrix row when lib/ behavior changes: N/A, no lib/ change
• Other targeted/local-only validation: dart test -p vm -j 1 $(cat tool/ci/markdown_test_suites.txt) (728 passed, including package_skills_test.dart); ./tool/docs/build_site.sh --port 18736 (exit 0, link check passed)

Matrix Evidence

Matrix row	Scope covered	Platform / model / backend	Result	Evidence / notes
docs-site	website build, docs metadata, generated docs routes	macOS arm64, Flutter 3.47.1	PASS	./tool/docs/build_site.sh --port 18736 exit 0 (the default port 8080 was taken locally)

Review Notes

• Independent review status: N/A, standard-risk docs-only change.
• Real-bundle facts behind the wording (macOS arm64, native LiteRT-LM CPU):
  • Qwen3.5-0.8B_int8.litertlm: metadata reports the Qwen3 template (hermes). chatTemplate([user 'hi'], enableThinking: true) ends in <|im_start|>assistant\n. The embedded template, which the runtime renders, always ends in <think>\n\n</think>\n\n.
  • gemma-4-E2B-it.litertlm: the embedded template (11995 chars) differs from llamadart's gemma4 template (16430 chars).
• CI status / head SHA: green at 278b8dba7 (32 pass, 9 skipped, 0 failed).

High-risk regression review

• Classification: standard
• Implementation task: N/A
• Independent blocking QA task: N/A
• Exact head / current base: N/A
• Production-branch deletion, bypass, or miswire proof: N/A
• Affected-family real-model/artifact evidence: N/A
• Explicit unavailable-family or other N/A evidence: N/A
• Known PR-caused P1 regressions: 0
• Unresolved review threads: 0

[leehack]

Landed via batch integration PR #1843 (rebase merge), one commit per constituent.

[github-actions]

Chat app preview removed for leehack/llamadart-chat-pr-1842.