[daierenao]

Closes #1250

Purpose of change

The Tool Use guide mixes user-facing configuration with internal scheduling details and leaves parameter-injection behavior spread across pages. Clarify tool declaration and registration, injection sources and override/failure rules, and parallel-call configuration, independence requirements, and timeout boundaries. Correct incomplete declaration snippets, keep async configuration-reference entries focused on user-facing constraints, and link the YAML, chat, and configuration references to the relevant sections.

Use the implemented async.batch.parallelism option rather than the tool-call.parallelism name mentioned in the issue. Existing section anchors and the old Tool Use URL alias are preserved. No runtime behavior or configuration defaults change.

Tests

• Hugo extended 0.110.0 documentation build; checked 359 internal links/anchors across the four changed generated pages, the old Tool Use alias, and the example source links.
• All eight Python fenced examples parse. Minimal Java/Python API wrappers validate configuration, tool/resource declaration, and ReActAgent construction. The two Python and two Java Ollama descriptor snippets also construct the actual setup and expose qwen3:8b as the model; Python negative controls confirm a missing model is rejected.
• Clean Java reactor rebuild and focused ToolCallActionTest / FunctionToolTest suites: 30 passed. Python tool-action, function-tool, and decorator suites: 91 passed, importing the current checkout.
• 13 focused Python timeout and execution-mode tests passed for the review follow-up; these overlap the suites above.
• git diff --check passed.

Java validation uses JDK 26 with the Java 11 compilation target. No real Flink job, external model call, or JDK 11 runtime validation was performed. Documentation build and focused suites were rechecked after rebasing onto c8886cfa.

API

No public API changes.

Documentation

• doc-needed
• doc-not-needed
• doc-included

Was this patch authored or co-authored using generative AI tooling?

• Yes
• No

Generated-by: Codex CLI 0.162.0-alpha.17.2 (GPT-6)

[sangkyoonnam]

I checked the changed statements against the Java and Python tool-call code at e3f3bd0 and ran the Python injection, override and parallelism=1 tests; the Java side is from reading. Three inline notes correct facts (the snippets, the timeout paragraph, the reconciler advice) and one is a nit. None needs a design change.

[da-daken]

Thanks for your contribution! I left two nits below; overall, LGTM. Waiting for @wenjin272 to make the final decision.

[wenjin272]

Thanks for the update! The Tool Use guide now covers the main requirements of #1250.

The Configuration table still retains some internal details that #1250 asks to remove: JDK-specific thread pools, cross-key contention, and started/unstarted durable state. Could we simplify the async.batch.parallelism, async.batch.timeout.ms, and num-async-threads entries to defaults, user-facing constraints, and links to Tool Use?

[daierenao]

Thanks for catching that! Simplified those three entries and linked them to Tool Use, keeping the defaults and user-facing constraints.

[wenjin272]

Thanks for addressing my comments. LGTM