Changed

  • Sessions keep their full context by default. context.handoffAt: auto no longer means half the window or 100k tokens: HarnessLink starts no compaction and asks for no handoff, and omp, Claude Code and Codex compact on their own when their context needs it. Every compaction still saves a handoff first, and omp's still uses HarnessLink's summary. The early compaction disrupted work and saved few tokens in practice. A token count or a share ("40%") brings back the earlier behaviour. The session-start context no longer tells the model to keep sessions lean and clear freely.

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: 17.5.85 · Older: 17.5.83 · All releases · Atom feed

Flag notifications