Added

  • Search shows what each memory concluded (ADR-105). Every entry now carries a one-line abstract (the outcome sentence of its result) and a short summary (the key details, what was asked, the files touched), extracted by code — no model. memory_search and hnl cloud recall show the abstract for every hit and the summary for the top three instead of a title and a text fragment; memory_get still returns the full entry. The session brief's Recent Work and the per-prompt memory lines show abstracts too. On the retrieval benchmark this is 226 characters per hit against 241 before.
  • Near-duplicate memories collapse to the newest, and you can forget one (ADR-106). Entries of one project about the same topic show as one result led by the newest, with an also: line naming the others, so a reversed decision no longer hides behind the old one (benchmark: the newer decision ranks first 50 % of the time, was 36 %). hnl cloud forget <rel_path> and the MCP tool memory_forget remove one entry from this machine (file, search index, MAIN.md and day summary) and from the cloud (entry, search index, vector); your other machines delete their copy at their next sync.

Update HarnessLink and every installed harness to its current release:

hnl update

New machine? Install with:

curl -fsSL https://harnesslink.sh/install.sh | bash

Newer: 17.21.1 · Older: 17.20.2 · All releases · Atom feed

Flag notifications