v1.19.12

Make catalog pruning explicit: normal model refreshes preserve stale registry entries and only report drift; use --prune for intentional removal. See docs/CHANGELOG.md.

v1.19.11

Cross-adapter catalog freshness and safe model-registry updates: probe-only snapshots, pending diffs, cached preflight membership checks, corrected source mapping, and opt-in adapter isolation capabilities. See docs/CHANGELOG.md.

v1.19.10

Price-aware Cursor model routing, current GPT-5.6 defaults, catalog drift warnings, and multi-worker Cursor reliability fixes. See docs/CHANGELOG.md.

v1.19.8

Hashline content-hash edits and mtime read cache for agentic workers (OMP-inspired). See docs/CHANGELOG.md.

v1.19.7

Hermes-style segmented tool-batch concurrency for agentic workers. See docs/CHANGELOG.md.

v1.19.6

Same-adapter model fallback for unavailable/rate-limited models when only one platform is funded. See docs/CHANGELOG.md.

v1.19.5

Per-role swarm routing_policy defaults. See docs/CHANGELOG.md.

v1.19.4

Plan-then-cheap prewalk (OMP-style) with durable plan-artifact injection. See docs/CHANGELOG.md.

v1.19.3

Bedrock ConverseStream for live chat deltas. See docs/CHANGELOG.md.

v1.19.2

Keep default swarms read-only through adapter routing. See docs/CHANGELOG.md.