github iwe-org/iwe iwe-v0.34.0
v0.34.0

latest releases: iwec-v0.34.0, iwes-v0.34.0
one hour ago

Lexical search now works across languages: it matches different forms of a word in Russian, Greek, Arabic, Tamil and accented languages, and finds words inside Chinese, Japanese and Korean text. retrieve now shows each document's frontmatter, memory session start can be capped and filtered with max_tokens and exclude, and a write that fails part way through now rolls back the files it already changed.

iwe

Added

  • session_start.max_tokens memory policy setting caps the whole session-start block, policy text included, by dropping listed entries from the end
  • session_start.exclude memory policy setting takes a filter for documents session start never lists, such as retired ones

Changed

  • retrieve output includes each document's frontmatter — a frontmatter object in JSON and YAML, nested under frontmatter: in the markdown header — and --frontmatter false leaves it out (previously frontmatter was never shown)
  • internal claude session inbox without a session id prints each candidate's evidence under its summary (previously only the first line of the body was shown)
  • retrieve --exclude also removes the excluded keys from the includes, includedBy, references and referencedBy lists (previously they were still listed there with their titles)
  • internal claude session stage refuses a proposal whose evidence has no double-quoted text, or quotes text that does not appear in the session transcript (previously any non-empty evidence was accepted)
  • internal claude session complete counts a staged candidate as offered when it settles as written or rejected, and --offered is only for offers that were never staged (previously distill passed the staged count, so running complete again counted the same offers twice)
  • Memory reflect and init texts pass --strict on every create, update and delete they suggest (previously only distill did)
  • Memory distill, init and query cookbook texts describe user selection as a rule distill follows rather than a guarantee, since other writers can still change the store (previously they said nothing reaches memory unselected)
  • internal claude session complete --drop-pending marks leftover proposals as dropped and keeps them out of the rejection list that distill and reflect learn from (previously they were recorded as rejected by the user)
  • internal claude session adopt records the adopted line range in the session record, and session list shows a session as adopted when its unread tail was adopted after a partial read (previously it showed as distilled)

Fixed

  • find --lexical matches different forms of a word in Russian, Greek, Arabic and Tamil, and in more cases for accented languages such as French, Portuguese and Turkish (previously stemming silently did nothing for non-Latin scripts)
  • find --lexical finds words inside Chinese, Japanese and Korean text, and no longer matches unrelated Chinese characters that sound the same
  • Memory session start still shows the policy's At session start section and the distill backlog when no document matches the listing (previously the whole block was dropped)
  • Memory session start keeps each listing within its own max_tokens (previously the first entry was always kept, even over the limit)
  • internal claude session complete checks an area hub against its schema before linking a written document into it, and warns instead of writing when the hub would fail (previously the hub was written without a check)
  • internal claude session complete saves the session record before linking area hubs and records a repeated capture or rejection only once, so a failed run can be run again (previously a failure could leave hubs linked with no record, and a repeated run counted twice)
  • A write that fails part way through restores the files it already changed (previously the earlier changes stayed on disk)
  • update -k <key> --content checks --expect against the document (previously the flag was silently ignored in body-overwrite mode)

iwes

Fixed

  • Workspace symbol search matches different forms of a word in Russian, Greek, Arabic and Tamil, and in more cases for accented languages such as French, Portuguese and Turkish (previously stemming silently did nothing for non-Latin scripts)
  • Workspace symbol search finds words inside Chinese, Japanese and Korean text, and no longer matches unrelated Chinese characters that sound the same

iwec

Changed

  • iwe_retrieve results include each document's frontmatter as a frontmatter object, and frontmatter: false leaves it out (previously frontmatter was never shown)
  • iwe_retrieve exclude also removes the excluded keys from every edge list (previously they were still listed there with their titles)

Fixed

  • iwe_query lexical search matches different forms of a word in Russian, Greek, Arabic and Tamil, and in more cases for accented languages such as French, Portuguese and Turkish (previously stemming silently did nothing for non-Latin scripts)
  • iwe_query lexical search finds words inside Chinese, Japanese and Korean text, and no longer matches unrelated Chinese characters that sound the same
  • A write that fails part way through restores the files it already changed (previously the earlier changes stayed on disk)
Library crates (liwe, diwe)

diwe

Added

  • RetrieveOptions::frontmatter and DocumentOutput::frontmatter, the document's frontmatter in retrieve output, counted against max_tokens

Changed

  • RetrieveOptions::exclude also removes the excluded keys from every edge list in DocumentOutput

Fixed

  • apply_changes puts back every file it already removed, created or rewrote when a later write fails, and removes the directories it created for them (previously a failure part way through left the earlier changes on disk)

Don't miss a new iwe release

NewReleases is sending notifications on new releases.