Release Changelog & Roadmap

The following capabilities are merged or being tested on the development branch, and are not included in the 0.1.0-preview.2 download. Run via local source build to evaluate:

• Agent Lab Paired Worktrees: Runs baseline and candidate tasks in independent Git worktrees branched from the same commit, evaluated by a separate verifier script.
• Empirical Outcomes & Inconclusive: Explicitly classifies results into Improved, Worse, and Inconclusive (insufficient evidence or stochastic variance).
• SessionStart Hook Generation & SafeApply/Undo: Generates read-only hook definitions from promoted rules, safely applies them to local projects, and retains Undo hash snapshots.

Key Improvements

• Workspace Redesign: Major overhaul of the native session workspace and inbox views, improving log browsing efficiency and context extraction responsiveness.
• Notification & Sound Categorization: Introduced granular notification toggles and sound feedback settings, allowing mute controls for routine completions according to engineer preference.
• Native Interface Integration: Standardized empirical UI representations, connecting authentic macOS native screenshots and approval details.
• Core Verification Suite: Consolidated automated test pipelines in the repository (including system test scripts, native RPC communication tests, and full-repo contract checks).

Code Signing & Gatekeeper Instructions

Initial Core Capabilities

• Native AppKit Host: WebKit process isolation and menu bar integration, with ⌘1..6 module navigation and global ⌘K search.
• Agent Log Ingestion: Bounded discovery of Claude Code and Codex logs (Cursor format supported via read-only import), timeline rendering, and inferred session states.
• Workflows & Dry Run: Markdown workflows (*.md) orchestrating Git checks and build commands with read-only real execution and mutating stub rehearsals.
• Inbox Gating & Parameter Freezing: Commands queue with frozen parameter hashes requiring explicit single-action approval.
• Memory Lifecycle & Recall: Four-state lifecycle (Candidate ➔ Active ➔ Superseded ➔ Archived); lexical recall strictly limited to active entries; Private Library completely blocked from agent retrieval.
• Deterministic Improvement Suggestions: Generated suggestions requiring 3+ occurrences across 2+ sessions; sequential worktree runs comparing exit codes and durations.
• stdio MCP Interface: Built-in vela mcp providing read-only project context to external agents.

For full commit history and changelog details, see CHANGELOG.md and Feature Status (docs/status.md).