v0.13

What's Changed

• fix: per-action write state so one write failure doesn't block the rest by @mattmezza in #20
• Thinking-level support across all inference kinds (LLM tab) by @mattmezza in #21
• feat(voice): clean non-pronounceable chars from TTS replies by @mattmezza in #22
• fix(jobs): block create only on live duplicate id (#11) by @mattmezza in #24
• feat: deepseek-v4-flash default + terminal REPL by @mattmezza in #23
• feat(agent): batch write-action approvals into one prompt (#12) by @mattmezza in #25
• feat: persona profiles — swappable agent identities (#13) by @mattmezza in #27
• feat: vision fallback — caption images via a secondary model on non-vision models by @mattmezza in #26
• feat: optional headless browser automation (#16) by @mattmezza in #28
• Secrets vault: encrypted two-vault backend with per-persona access (#19) by @mattmezza in #31
• Per-chat persona binding + isolated contexts (channel-agnostic) (#14) by @mattmezza in #32
• Show CoT thinking trace in admin UI log tab (#33) by @mattmezza in #38
• feat(secrets): migrate static credentials onto the infra vault + collapse redundant UI (#35) by @mattmezza in #40
• feat(artifacts): agent-crafted web pages served at /artifacts/ by @mattmezza in #37
• feat: bot-per-persona — each persona as its own Telegram bot (#29) by @mattmezza in #39
• feat(memory): two-tier scoped memory — shared pool + per-persona private (#42) by @mattmezza in #45
• fix(agent): inject memory + reflections per-turn, not in frozen snapshot (#41) by @mattmezza in #43
• feat(memory): recall_memory tool — deliberate full-store semantic lookup (#47) by @mattmezza in #49
• fix(agent): inject skills index per-turn, not in the frozen snapshot (#46) by @mattmezza in #48
• feat: subagents — on-demand + scheduled spawning with lifecycle management (#15) by @mattmezza in #44
• feat(skills): serve the index on-demand via search_skills/list_skills (#50) by @mattmezza in #52
• Fix Telegram formatting: avoid tables, use code blocks for commands by @mattmezza in #54
• feat: reply-decision gate to stop infinite loops in group chats (#36) by @mattmezza in #53
• fix(telegram): typing/thinking indicator on Telegram Web (K) (#57) by @mattmezza in #58
• feat(vault): enable secret removal from vaults (#56) by @mattmezza in #60
• feat: image generation tool (OpenRouter, fal.ai, OpenAI) with admin UI + wizard (#55) by @mattmezza in #61
• feat(telegram): group multi-agent rooms — turn-taking + loop guard + speaker tags (#30) by @mattmezza in #59
• Auto-dismiss permission approval messages after user interaction by @mattmezza in #63
• feat(voice): advertise voice reply as a base capability by @mattmezza in #66
• Scoped /yolo mode + close the permission-generalisation bypass by @mattmezza in #67

Full Changelog: v0.12...v0.13