Add TypeScript tabs to the graph workflow pages

[kalenkevich]

What this does

Adds TypeScript coverage to the five graph-workflow pages under /graphs/.
Every section that had a Python tab now has a TypeScript tab, placed
between Python and Go to match the ordering used elsewhere in the repo — 30
tabs in total.

Page	TypeScript tabs
graphs/index.md	2
graphs/routes.md	7
graphs/data-handling.md	8
graphs/human-input.md	4
graphs/dynamic.md	9

The code lives in 26 new snippet files under
examples/typescript/snippets/graphs/, pulled in with --8<-- region
includes, following the convention already used by the Go snippets and by the
existing TypeScript snippets in callbacks/, runtime/ and agents/.

Why the Python tab wasn't enough

The Python examples don't translate line-for-line, so a TypeScript reader had
to guess at several things that differ:

• there is no @node decorator — node(fn, options) is the factory form, and
  new FunctionNode(name, fn, config) the explicit one
• schemas are Zod objects (or a genai Schema), not pydantic models
• state is written through ctx.state, not returned on an event
• a user-facing message is the event's content; output is what reaches the
  next node
• ctx.runNode() resolves to a node result, so you read .output

The tabs also call out three behaviours that are easy to get wrong and have no
Python counterpart:

• ctx.runNode() does not throw when a child interrupts — it resolves with
  interruptIds populated and output still undefined, so an orchestrator
  that doesn't check will decide on an answer the human never gave
• a second event carrying output silently overwrites the first
• LlmAgent.inputSchema is not the node's input contract inside a graph; the
  validating schema belongs on the node via node(agent, {inputSchema})

Provenance of the snippets

These are ports of the runnable samples in
google/adk-js under samples/workflows/,
which already map 1:1 to the section anchors on these pages and are executed in
that repo's CI (every sample is constructed, and the offline ones run
end-to-end with the model stubbed). So the code here is derived from examples
that are known to run, not written fresh against the docs.

Verification

• mkdocs build --strict passes.
• All 30 TypeScript tabs render, in the correct Python → TypeScript → Go order,
  with no tab set left without a TypeScript entry.
• All 53 --8<-- references on these pages (Go's 23 + the new 30) resolve to a
  real file and region; no unresolved directives in the output.
• All 26 snippets type-check clean under tsc --noEmit with strict,
  noUnusedLocals and noUnusedParameters, against the adk-js workflow API.
• License headers sit outside the include regions, so they don't leak into the
  rendered code blocks (verified against the built HTML).
• The markdown diff is purely additive apart from the five
  language-support-tag lines.

Open questions for reviewers

1. Version label. Resolved. @google/adk@2.0.0 is now published, so
   the ^2.0.0 pin installs and the TypeScript v2.0.0 tags are correct. All 26
   snippets type-check clean under strict against the published package.

2. No CI for TypeScript snippets. Go and Kotlin have
   go-snippets-pr-check.yaml / kotlin-snippets-pr-check.yaml, but nothing
   compiles the TypeScript ones, so they will rot silently. I'd suggest a
   follow-up workflow running tsc across examples/typescript/snippets/* —
   that would cover the whole tree, not just these pages. Can do it in this PR
   or separately, whichever you prefer. (It would be red until point 1 is
   resolved, since the package can't install today.)

Notes

• Draft while the version question above is open.
• No existing content was rewritten; Python and Go tabs are untouched.

[netlify]

✅ Deploy Preview for adk-docs-preview ready!

Name	Link
🔨 Latest commit	83af7c9
🔍 Latest deploy log	https://app.netlify.com/projects/adk-docs-preview/deploys/6a88ba1e304b64000865cc1a
😎 Deploy Preview	https://deploy-preview-2167--adk-docs-preview.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.
🤖 Make changes	Run an agent on this branch

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[joefernandez]

Thanks for the update. Please address comments.

[kalenkevich]

Thanks, changes applied.

Wording. Applied every suggestion. Beyond those, I did a pass over all 30 TypeScript tabs for the same problems rather than only the flagged lines:

• No sentence starts with code syntax.
• Removed the informal phrasing: "earn their keep", "dropped straight into", "reach for", "hands you", "two things to know going in", "kick the children off", "fails loudly".
• Spelled out / as "and".
• Described the ctx.runNode() interrupt behaviour concretely instead of "does not throw": it returns normally, with interruptIds populated and output still undefined, and an orchestrator that skips the check continues with a value the user never supplied.
• Replaced "the barrier" with what the JoinNode actually waits for.
• Tied rerunOnResume back to the sample it follows, and introduced the two orchestrator details by saying when they matter.

Same pass over the snippet header comments, including the nested_workflow.ts one you flagged.

Notes moved out of the tabs. All three are now single, language-neutral notes after the code examples:

• Response schema limitations — was duplicated in the Python tab as well, so both are replaced by one note.
• Stuck JoinNode — was in all three tabs, now one caution.
• Unbounded graph cycles — moved out of the TypeScript tab.

Two things worth your call

1. The three JoinNode cautions did not say the same thing. Consolidating them surfaced a behavioural disagreement rather than a wording one.

Python: "the JoinNode is stuck and workflow execution stops."
Go: "the JoinNode is stuck and workflow execution stops."
TypeScript (mine): the join still releases, with undefined for that predecessor.

The TypeScript one is right for adk-js. In workflow.ts:578 the barrier releases when every predecessor reaches NodeStatus.COMPLETED — not when it produces output — and the successor's record is built with loop.nodeOutputs.get(p), which is undefined for a predecessor that completed without one. So it does not stall; it hands the next node a missing value, and the failure appears further downstream.

I could not verify the Python or Go runtimes, so I did not want to assert either behaviour for all three languages. The consolidated caution states the actionable rule that holds regardless — every node feeding a join must produce an output — and describes the consequence without claiming a specific mechanism. If the Python and Go statements are accurate for those runtimes, this is a genuine cross-language difference and probably deserves to be documented explicitly; happy to add that if you can confirm the Python side.

2. Definition-list bullets left as they are. Your "fix the other sentences that start with code syntax" note — I read that as prose sentences, not the - **`message`:** Text shown to… bullets, since the Python and Go tabs on the same pages use that form. Say the word if you want those changed too, but it would mean changing the other languages' tabs to match.

[joefernandez]

thanks for updates! Approved for publish