feat(tools-registry): update tako-search entry

[robertabbott]

Updates the existing tako-search entry in content/tools-registry/registry.ts for @takoviz/ai-sdk 3.0.0. One file, +8/-6. No new entry and no new page — the listing already renders at /resources/tools/tako-search.

The code example demonstrated the routing the tool argues against

This was the substantive problem, and it is why the example changed rather than just the metadata.

The entry prompted 'What is the stock price of Nvidia?' against takoSearch. The package's own takoSearch description — shipped in 3.0.0, read out of the published tarball — says:

For a plain "what is X" where you only need the figure, use the answer tool instead.

So the card showed a single-figure lookup routed to the one tool the package tells the model not to use for it. It compiled and ran, which is why it survived; it just taught the wrong thing.

The new prompt asks the kind of question somebody evaluating the package would actually ask, and it is a breadth question rather than a single figure — which is what takoSearch is for. The example also passes the full toolset, the way the package's own README does, consistent with superagent, you-search, bedrock-agentcore and nitrosend, which already show more than one tool.

Note the prompt string uses double quotes, because it contains an apostrophe. Every other prompt in this file is single-quoted and none contain one, so this is the first exception. node --check on the extracted example confirms it parses exactly as a reader would copy it; oxfmt does not reach inside a template literal, so it has no opinion here.

Field changes

Field	Before	After
description	Describes takoSearch only	Leads with the two-knowledge-sources pitch and maps all three tools
codeExample	Single-figure prompt, one tool	Chart request, full toolset
docsUrl	github.com/TakoData/ai-sdk#readme	docs.tako.com/documentation/integrations/vercel-ai-sdk
apiKeyUrl	https://tako.com	https://tako.com/console/api-keys
tags	4 tags	adds extraction

description — the previous text described takoSearch alone, so the card understated a package that exports three tools. Rewritten against Tako's own positioning, which leads with the part that was missing entirely: Tako returns curated structured data and live web search from one API, so an agent need not choose between a data API that misses most questions and a search API that makes the agent read the pages itself. The new text also maps each tool to what it returns.

440 characters, against a median of 232 and a maximum of 505 in this file. Leads with the product name, matching exa, valyu and firecrawl. Deliberately contains no apostrophe, so single quotes remain correct under singleQuote: true — the previous value needed double quotes for one.

docsUrl is the card's "Learn More" link and landed on a GitHub README. Repointed at the integration guide, matching how neighbouring entries link to vendor-hosted pages (exa → docs.exa.ai/reference/vercel, tavily → docs.tavily.com/documentation/integrations/vercel, firecrawl → docs.firecrawl.dev/integrations/ai-sdk, valyu → docs.valyu.ai/integrations/vercel-ai-sdk).

apiKeyUrl is the "Get API Key" button and pointed at the site root. It now points at the key page.

extraction is already used by five entries in this file, and takoContents extracts web page text, so the tag applies.

Verification

The contributing guide asks for an example that is tested, so it was executed rather than only typechecked:

tools called: ["takoSearch","takoContents"]
tool errors:  0
answer chars: 515

takoSearch returns a "Top Performing Stocks" card plus five web results; the model then calls takoContents to read the rows behind the card, and the answer names real tickers with their percent change.

• Example executed end to end against ai@7 and @takoviz/ai-sdk@3.0.0 — the model reaches for takoSearch, then takoContents to read the rows behind a card. Zero tool errors.
• The extracted example passes node --check
• registry.ts typechecks under --strict
• oxfmt 0.62.0 with this repo's .oxfmtrc.jsonc reports no changes
• Both new URLs return 200
• Content-only, so no changeset (per CONTRIBUTING.md: changesets are not needed for docs)
• Branched from current main

Two deviations were needed to run it locally, neither affecting the published example: openai('gpt-5.2') instead of the plain 'openai/gpt-5.2' string, because that string routes through AI Gateway and only a direct OpenAI key was available; and an explicit Tako baseUrl, because the key on hand was not a production one. The example in the entry keeps the plain model string used by other entries in this file.

Also worth stating: the prompt was chosen from measurement, not taste. A first attempt — 'Compare Nvidia and AMD revenue growth over the last decade' — produced a good answer but routed to takoAnswer twice and never called takoSearch, which is odd for an entry named "Tako Search". Several candidates were run before settling on one that exercises the named tool.

One caveat a maintainer should have rather than discover: today's makes this prompt time-sensitive. It was run on a weekday; over a weekend or a market holiday the card reports the last trading session, which is still a correct answer but not literally today. That is a property of the question, not of the tool.

Deliberately unchanged

• slug stays tako-search, so the existing URL keeps working.
• name stays "Tako Search". "Tako" would fit the package better now that it is three tools, but renaming the card is gratuitous churn and the slug would no longer match.
• model: 'openai/gpt-5.2' is current and used by other entries here.
• installCommand, npmUrl, websiteUrl, apiKeyEnvName are all correct.

Supersedes

This replaces #18380 and #18355, both of which will be closed. #18380 changed three metadata fields and explicitly left codeExample alone; #18355 proposed a community-provider page, which is the wrong home — @takoviz/ai-sdk exports tools rather than implementing a provider interface, so the tools registry is the right listing.