refactor: remove redundant attachment type inference chain

Simplify attachment handling by removing fallback type-guess logic and keeping MIME detection content-header driven. Align failing permission tests with DSL-only configuration and harden async hook test stability.

Made-with: Cursor

Co-authored-by: Cursor <cursoragent@cursor.com>

feat: finalize permissions DSL runtime flow and upgrade memory search…

… with bleve

Move permission DSL validation to runtime registration flow, wire additionalDirectories into built-in tool sandbox allowlists, and improve deny/ask observability at the executor decision point. Replace custom memory search ranking with Bleve-based retrieval and align memory_search output policy to default top-3 with a max_results cap of 5 while updating README docs and tests to the new behavior.

Made-with: Cursor

Co-authored-by: Cursor <cursoragent@cursor.com>

feat: add voice pipeline and plain-text TTS output rule

Introduce provider-based ASR/TTS support in the SDK runtime, unify attachment metadata detection across integrations, and enforce plain readable text when TTS is enabled.

Made-with: Cursor

Co-authored-by: Cursor <cursoragent@cursor.com>

release v0.1.7

fix: use sentinel error for prompt guard detection

Made-with: Cursor

Co-authored-by: Cursor <cursoragent@cursor.com>

fix: use sentinel error for prompt guard detection

Made-with: Cursor

Co-authored-by: Cursor <cursoragent@cursor.com>

fix: keep realtime params unescaped in progress updates

Made-with: Cursor

Co-authored-by: Cursor <cursoragent@cursor.com>

feat: unify memory write operations and refine flush signaling

Made-with: Cursor

Co-authored-by: Cursor <cursoragent@cursor.com>

feat: improve memory search and agent skill injection

Co-authored-by: Cursor <cursoragent@cursor.com>

feat: inject available skills into every prompt; add ScanSkillsList a…

…nd dynamic SkillTool FS execution

Co-authored-by: Cursor <cursoragent@cursor.com>