1.7.0

Changelog

1.7.0

Navigation

• Replaced the top nav tab bar with a collapsible sidebar — icon-only rail on desktop (state persists across reloads), slide-out drawer on mobile.

Themes

• Added new themes — Dracula, Rosé, and more.
• Every theme now has a subtle animated accent sweep across the top bar and sidebar edge, respecting the OS "reduce motion" setting.

Intelligence

• The Decision Engine now weighs the current session's run history when scoring tools — a tool that just failed is deprioritized (not excluded) on the next recommendation.

Run page & live output

• Run page tool execution now streams live output and progress instead of blocking silently until the command finishes.
• Fixed the process dashboard/stream showing a fake "Running for Xs" placeholder — last_output now carries the tool's real stdout/stderr, visible live on the Tasks page.
• Selecting a tool on the Run page now updates the URL to #/run/<tool> — bookmarkable/shareable, reopens straight into that tool with empty inputs (no params prefilled).

Accessibility

• Added focus-trap keyboard navigation to the Command Palette and Chat widget.

Error recovery

• Fixed /api/error-handling/execute-with-recovery raising a server error on every real call.
• Wired automatic error recovery (retry/backoff, alternative-tool swap) into recon, scanning, fuzzing, and forensics tool executions.

Process pool

• Fixed /api/process/execute-async returning the cached result payload instead of a task ID on a cache hit, which broke polling it via /api/process/get-task-result/<task_id>.
• Fixed the process pool's worker auto-scaling miscounting active workers after scaling down.
• All tool command execution now runs through the process pool's auto-scaling (previously bypassed it); pooled task results no longer leak memory on long-running instances.

Internals

• Migrated the entire tool registry to a unified ToolSpec-driven registration system, replacing legacy per-category registration code across all tool categories.