Release v0.1.10¶

✨ Highlights¶

Voice/media outbound stabilization: explicit voice requests now route to real media delivery on chat channels even when model output is stale.
Runtime guard hardening: stale text-only/approval denial responses are rewritten or recovered into contextual voice payloads.
Command/channel reliability: Telegram /pack command is now handled through the standard inbound processing pipeline.

Auto media-delivery path in AgentLoop for explicit voice/image/sticker requests on Telegram/WhatsApp when message tool was not called.
Voice-content recovery pass to regenerate contextual payloads when draft output contains stale capability denial text.
Expanded regression coverage for runtime checkpoints:
stale voice denial rewrite (EN/ID)
approval-required/meta reply rewrite recovery
non-delivery voice-topic and negated-voice protection
persistent memory truth enforcement for Indonesian phrasing.
Telegram channel test coverage for /pack command forwarding.

Prompt/tool guidance now clearly states message supports outbound media (voice, image, audio, sticker, document) and should not be treated as text-only.
Delivery-mode detection in runtime now includes broader Indonesian phrasing (including dengan suara) with negation-aware matching.
Workflow/approval test scenarios tightened for silent media paths.

Prevented stale “text-only / approval required” replies from leaking as final voice output in explicit voice requests.
Ensured auto voice-delivery tests are deterministic in CI by stubbing local TTS generation path.
Improved correctness of memory truth guard against “no cross-conversation memory” Indonesian variants.