Orca v1.1.0

1.1.0 - 2026-10-01

Added

• Automatic bounded loading of root res://AGENTS.md as subordinate, private request-scoped project guidance.
• Bounded metadata discovery for res://skills/<slug>/SKILL.md and explicit read_project_skill loading for one exact skill body.
• Read-only inspect_godot_api reflection for ClassDB signatures, hierarchy, properties, signals, constants, enums, and safe editor Open Docs topics without prose scraping or object construction.
• Read-only read_gdscript_function extraction from saved scripts or exact unsaved editor source, with provenance and no patch hash for editor source.
• Read-only discover_dependencies traversal for bounded forward serialized dependencies and reverse dependents.
• Tool-loop thrashing detection for repeated calls/results, alternating cycles, repeated rounds, and no-progress rounds, followed by one no-tools finalization request beneath existing hard caps.
• Permanent automated suites for project instructions, project skills, Godot API reflection, focused function reads, dependency discovery, and tool-loop guarding.

Changed

• Improved post-1.0 responsive editor scaling, compact composer sizing, narrow action containment, transparent branding, expanded diff sizing, and active-turn auto-follow for delayed content and sequential approvals.
• Updated release metadata and About-page coverage to 1.1.0.

Security And Privacy

• Project guidance remains subordinate to system, user, mode, approval, path, and runtime safety checks and is removed from stored continuation history after each turn.
• Skill bodies are not automatically loaded, symlink traversal is rejected, and instruction/skill text and metadata use strict byte, line, and count limits.
• Unsaved editor source is never represented as a disk patch base, API reflection does not construct reflected objects, and dependency discovery does not load or instantiate resources.