v0.5.45: Reliable Idle Review and Clearer Skill Management
- DSH 0.2.2-rc.2: hide empty retired-memory context cards while preserving real memory snapshots, audit history and append-only context behavior (#618).
- Readable memory workers: identify supported Host subagents by operation and a unique suffix, retaining existing task labels and isolation (#620; fixes #616).
- Collapsible skill groups: collapse source directories, see paths and counts, and reveal matching or selected skills automatically (#621; closes #617).
- Optional linked-skill updates: an off-by-default Skills setting permits editing, restoring and adopting changes in the linked original while preserving symlinks and junctions (#622; closes #611).
- Quiet review cancellation: disposed Agent contexts no longer start idle reviews or install observers. Cancellation without committed writes stays quiet; real errors and partial-write receipts remain visible (#624; fixes #623).
- Review output budget: increase the default per-response cap from 4,096 to 16,384 tokens. Explicit
idleReview.maxTokenssettings continue to take precedence (#615).
Upgrade and compatibility
Update the complete Starter composition through Memory System → Status → Check versions → Update, then fully restart DSH. This release updates dsh-mnemon to 0.5.45 and dsh-mnemon-source-skills to 0.5.12; the other 17 plugins keep their published versions. Memory storage formats are unchanged, and linked-skill updates remain off unless enabled.
Official DSH 0.2.2-rc.1 and rc.2 are supported alongside the documented older Hosts. Alpha and arbitrary source builds are not compatibility promises. See compatibility.
The review budget is an upper bound for each model response, not a fixed token charge. Hosts do not universally clamp it to a model's output capability. If the review model or gateway accepts less than 16,384, explicitly set Model output token limit per request (idleReview.maxTokens) to its supported value, such as 4,096. Existing saved overrides are retained; no automatic retry or limit adaptation is added in this release.