Skip to content

0.5.11.md

Latest commit

 

History

History
24 lines (16 loc) · 1.27 KB

0.5.11.md

File metadata and controls

24 lines (16 loc) · 1.27 KB
Raw

Helm 0.5.11

Helm 0.5.11 adds explicit handoff state for long-running tasks and tighter preflight routing around planning work and skill selection.

Highlights

  • finalized profiled tasks now write markdown [STATE_SNAPSHOT] artifacts under .helm/state-snapshots/
  • helm context state-snapshot shows the latest snapshot or a specific task snapshot
  • profiled runs expose the previous snapshot through HELM_PREVIOUS_STATE_SNAPSHOT and OPENCLAW_PREVIOUS_STATE_SNAPSHOT
  • adaptive harness preflight now records interaction_workflow for divergence/convergence routing
  • explicit skill choices are scored with skill_relevance, and poor matches fail preflight
  • file-intake evidence probing can now be inferred from existing command paths during backfill

Why this release matters

Long-lived agent work fails less often when the next run can recover from a compact file artifact instead of relying on chat history. This release makes that handoff explicit and inspectable.

The harness changes also reduce two practical failure modes:

  • complex planning requests being treated as immediate execution without a divergence step
  • unrelated skills being forced onto a task just because a skill name was supplied

This is a patch release focused on operational continuity and safer routing.