Release v0.57 · agent0ai/space-agent

Downloads

Architecture	MacOS	Linux	Windows
x86	Mac Intel	Linux x86	Win x86
ARM	Mac ARM	Linux ARM	Win ARM

This release focuses on documentation alignment, runtime prompt refinements, and new tests covering widget imports and watchdog behavior.

Browser-control requirement: Off-runtime website visits now require a stand-alone browser-control window; location-based navigation is explicitly forbidden.
Skill-loading semantics: Successful skill loads are treated as read-stage operations and must reference exact catalog/loaded-skill IDs.
Widget-authoring guidance: Added compact documentation for the widget renderer signature (async(parent, currentSpace, context)) and context.import("scripts/...") pattern, along with a new widget-import.js helper.
Watchdog/CLI freshness model clarified: CLI writes are recognized as out-of-process; clustered workers must publish logical-path commits; the watchdog full-tree reconcile is documented as a rare backstop that jobs and code should not depend on.
New tests: Added spaces_widget_import_test.mjs, watchdog_external_changes_test.mjs, and several agent_llm_performance test cases with updated histories and results.

Docs: browser-control, widgets, watchdog & tests — Aligned runtime prompts and docs around browser-control requirements, skill-loading semantics, widget-authoring guidance, and watchdog/CLI freshness model. Updated AGENTS.md and onscreen prompt files, tweaked server user-crypto handling and watchdog implementation, and added widget-import.js helper with accompanying test suites and agent LLM performance test updates.