1Helm 0.0.5
Native model autonomy restored
- Replaced large behavioral manifestos with compact factual identity, machine, workspace, tool, memory, skill-inventory, and authority context.
- Added lazy skill discovery/loading so full procedures enter context only when selected.
- Removed prose-policing evidence/outcome gates that rewrote capable-model answers.
- Kept human display names out of model transcripts and added one concise retry for empty conversational web/news searches.
- Verified the exact WeHo sinkhole prompt and resident Linux/skill awareness against the configured real model.
Apple Silicon DMG and updater ZIP are Developer ID signed, Apple-notarized, stapled, and Gatekeeper accepted. The Linux host artifact is for the signed host-owned systemd updater.