expand_result

search_code deliberately withholds full source to keep responses small. expand_result is how the agent fetches it — but only for the results it actually needs to read in detail, usually just rank #1.

Inputs

How it works

After every search, the server caches the full result set in memory, keyed by rank. When you call expand_result(rank=3), it looks up result #3 in that cache and reads its exact line range from disk — no re-embedding, no re-searching, no re-ranking. Expansions are cheap, precise disk reads.

Example

search_code returns result #3 with a 234 more lines truncation note. To read it in full:

{
  "name": "expand_result",
  "arguments": { "rank": 3 }
}

The response is the complete source of that entity, with a header showing its name and file:start-end range.

Notes

• Only works after a search_code call in the same session. If the cache is empty you'll be told to run search_code first.
• rank must be between 1 and the number of results returned by the last search.
• To read arbitrary files or a named function outside of search results, use get_source instead.