Recipe: Look up a concept (the right way)

The task

Agents reach for a dictionary tool for everything — including “self-inductance,” “Lenz’s law,” “electromagnetic induction.” A dictionary is for English vocabulary, not technical concepts. This recipe is the decision tree, and the trap in the most common fallback (Wikipedia search on an ambiguous term).

Tools used: define_word, search_wikipedia, get_article_summary, get_concept (OpenAlex), crossref_search_works / openalex_search_works.

Calls: 1-3.

Copy-paste prompt

Define or explain "<term>" using Pipeworx. If it's a common English word, use the dictionary. If
it's a technical or scientific concept, use Wikipedia — but first tell me if the term is
ambiguous (a programming language vs. a place, an acronym with multiple expansions) and, if so,
disambiguate before answering rather than picking the first search hit.

Decision tree

Call tool names bare, as shown above — not as pack.tool_name dot-notation. That syntax doesn’t exist on this gateway and errors immediately: live-verified, dictionary.define_word({word: "sanguine"}) returns RPC error: Unknown tool: dictionary.define_word, not a lookup.

What a good answer looks like

define_word({ word: "sanguine" })

returns (live call, 2026-08-06):

{
  "word": "sanguine", "found": true, "phonetic": "/ˈsæŋ.ɡwɪn/",
  "meanings": [
    { "part_of_speech": "noun", "definitions": [ { "definition": "Blood colour; red." } ] },
    { "part_of_speech": "adjective", "definitions": [ { "definition": "..." } ] }
  ]
}

A trustworthy dictionary answer has found: true plus phonetic and per-part-of-speech definitions. A trustworthy Wikipedia answer (get_article_summary) has a description line distinct from the extract — when a term is genuinely ambiguous, Wikipedia is honest about it:

get_article_summary({ title: "Mercury" })
// → description: "Topics referred to by the same term"
//   extract: "Mercury most commonly refers to:\nMercury (planet)...\nMercury (element)...\nMercury (mythology)..."

That’s the good case — the ambiguity is visible in the response. Read on for the case where it isn’t.

A plausible-sounding failure looks like a complete, correctly-cited, entirely legitimate encyclopedia answer — about the wrong sense of the word. This is live, not hypothetical:

search_wikipedia({ query: "Java", limit: 1 })
// → { total_hits: 41599, results: [ { title: "Java",
//     snippet: "Java is one of the Greater Sunda Islands in the South East Asian country of Indonesia...",
//     pageid: 69336 } ] }

get_article_summary({ title: "Java" })
// → { title: "Java", description: "Island and region in Indonesia",
//     extract: "Java is one of the Greater Sunda Islands in the South East Asian country of
//     Indonesia... Java is the world's most populous island, home to approximately 56% of the
//     Indonesian population...", content_urls: {...} }

If the actual question was “what is Java” meaning the programming language, this response is a complete miss dressed as a hit — real title, real facts, real citation, description field present and populated exactly as a correct answer’s would be. Nothing here flags ambiguity, unlike the “Mercury” case above, because “Java” (island) is not a disambiguation stub — it’s a full primary article, and Wikipedia’s own disambiguation page lives at a different title (“Java (disambiguation)”) that search_wikipedia’s top-relevance ranking didn’t surface. Do not chain search_wikipedia({limit: 1}) straight into get_article_summary for a term you know is overloaded (language names, common nouns, acronyms) — check the top 3-5 hits’ snippet fields for topical fit first, or search the disambiguation title directly ("Java (disambiguation)"/"Java (programming language)") when you know which sense you want.

The shapes of failure

define_word({ word: "Lenz" }) → {found: false, hint: "..."}. Not a crash, but a wasted call — the dictionary doesn’t know about physics. Default to Wikipedia for anything that isn’t pure vocabulary.

get_concept (OpenAlex) can also fail loudly rather than quietly — live-verified, our platform key is currently over its daily OpenAlex budget:

get_concept({ query: "reinforcement learning" })
// → { error: "tool_error", message: "OpenAlex concepts search error: 429 — Insufficient
//     budget. This request costs $0.001 but you only have $0 remaining. Resets at midnight UTC." }

Same for openalex_search_works. When this happens, fall back to crossref_search_works, which is free and unaffected:

crossref_search_works({ query: "reinforcement learning" })
// → { total_results: 2698088, results: [ { doi: "10.5772/5275",
//     title: "Superposition-Inspired Reinforcement Learning and Quantum Reinforcement Learning",
//     authors: ["Chun-Lin Chen", "Dao-Yi Dong"], citations: 6, ... } ] }

search_works collides between the Crossref and OpenAlex packs. The bare name currently auto-routes to OpenAlex via the collision router — call crossref_search_works or openalex_search_works explicitly rather than relying on that.

Worked examples

”What does ‘self-inductance’ mean?”

// Wrong tool for the job:
define_word({ word: "self-inductance" })
// → { found: false, hint: "Word not in dictionary..." }

// Right:
get_article_summary({ title: "Inductance" })
// → { title: "Inductance", description: "Property of electrical conductors",
//     extract: "Inductance is the tendency of an electrical conductor to oppose a change...",
//     content_urls: {...} }

It wants the exact article title — when you don’t have it, run search_wikipedia first and take the top hit’s title, checking the snippet actually matches the sense you meant (see the “Java” failure above).

”What’s the field of ‘reinforcement learning’?”

get_concept({ query: "reinforcement learning" })
// → { display_name: "Reinforcement learning", level: 2,
//     ancestors: [{name: "Machine learning", level: 1}, {name: "Computer science", level: 0}],
//     related_concepts: [{name: "Deep reinforcement learning"}, {name: "Q-learning"}],
//     works_count: <count>, cited_by_count: <count> }

(Subject to the OpenAlex budget caveat above.)

Pattern: try-then-fallback

const dict = await define_word({ word: term })
if (dict.found) return dict

const hits = await search_wikipedia({ query: term, limit: 5 })
// check hits.results[].snippet for topical fit before picking one — don't blindly take [0]
if (hits.results.length) {
  return get_article_summary({ title: hits.results[0].title })
}

return get_concept({ query: term })

But honestly: most agent queries that hit define_word and miss are technical concepts. Just default to Wikipedia first for anything that isn’t a clearly common English word — and check more than the top hit when the term itself is a known overload (language names, acronyms, common nouns used as proper nouns).

Citation pattern

Self-inductance is the property of a circuit that opposes changes in current (Wikipedia). It’s quantitatively related to Lenz’s law and Faraday’s law of induction.

Caveats

• Call tools by their bare name, not pack.tool dot-notation — the latter errors immediately with “Unknown tool.” See the top of this page.
• A single-hit Wikipedia search can return a fully-formed answer about the wrong sense of an ambiguous term, with no disambiguation flag, when the top-ranked article happens to be a real primary topic rather than a stub. Check the snippet, or the top few hits, before trusting limit: 1.
• search_works collides between Crossref and OpenAlex — name the pack explicitly (crossref_search_works / openalex_search_works).
• OpenAlex tools can run out of daily budget (get_concept, openalex_search_works) — the error is explicit (429, "Insufficient budget"), not silent; fall back to crossref_search_works for literature search when it happens.
• Dictionary is English-only. For other languages, use Wiktionary via Wikipedia’s API or note this limitation.