Skills overhaul, policy unification, and UI polish

The skills ecosystem received a major overhaul with the introduction of skill_builder ability, enabling full CRUD operations on user-defined YAML playbooks. A dynamic Brain dashboard tab now displays My Skills and curated playbooks, with click-to-expand functionality for details. Proactive skill suggestions were added to automatically analyze complex ACT loops and recommend reusable workflows via a background LLM call. The entire skill suggestion pipeline was rewritten as a MessageProcessor subclass, eliminating custom parsing logic and cutting 285 lines of code while adhering strictly to the ACT loop.

Policy management was decoupled from hardcoded constants and now dynamically derives labels and categories directly from each Ability class via AbilityRegistry. Meta-tools are automatically filtered from the policy UI, and the blocked-log schema was corrected. Path resolution across the codebase was consolidated into a single FileMapperService singleton, replacing procedural path modules and removing 63 lines of duplicated logic. Security and file handling were strengthened with a new WebDownloadAbility featuring streaming support and a shared SSRF guard, alongside a FileWriteAbility with act-trail read guards and strict tmp/system write policies.

The web interface saw targeted UX improvements: chat Enter keys now insert newlines instead of submitting messages, provider API key links were made clickable, and timestamps are formatted server-side to avoid client timezone drift. HTML table support was added to the output sanitiser and TTS pipeline so LLM-generated tables render and speak correctly. Significant architectural cleanup removed the PWA stack, Deno interface SDK, dead related_abilities fields, and stale Codex CLI provider code, while migrating pre-trained assets to the backend directory.

The curated skill library expanded from 5 to 62 validated playbooks across multiple domains, with full FTS5+vec indexing rebuilt in skills.sqlite. Architecture documentation was updated to reflect the new FileMapperService paths, policy enforcement metadata, and Skills dashboard layout. Several internal scripts and test paths were corrected following the pre-trained asset relocation, ensuring pipeline stability.

• Skill library expanded from 5 to 62 curated playbooks with FTS5+vec indexing in skills.sqlite

• Policy UI now dynamically derives categories and labels from Ability class attributes instead of hardcoded constants

• Proactive skill suggestions triggered after ACT loops with four or more iterations, using a background LLM analysis gate

• Path resolution consolidated into FileMapperService, eliminating paths.py and standardizing document routing

• Chat interface updated: Enter inserts newlines, timestamps format server-side via locale_service, and HTML tables render correctly