v1.1.14 — catch-up release (1.0.1 → 1.1.14)

v1.1.14 — catch-up release

This release ships everything that landed in the source repo between v1.0.1 (Apr 30) and v1.1.14 (May 19). 13 versions, ~2,400 lines of changes across the bridges, templates, and SKILL.md.

Install / upgrade

/plugin marketplace add pattern-ai-labs/agentcall
/plugin install join-meeting@agentcall

Already installed? /plugin update join-meeting@agentcall.

Highlights

Multi-ecosystem support (1.1.2 – 1.1.5)

• Install matrix in README for Claude Code, Cursor, Codex, Gemini CLI, Windsurf, Copilot, Junie, OpenClaw
• New gemini-extension.json for Gemini CLI
• New pattern avatar template (radial sunburst, larger Inter typography) — now the default --template

TTS pipeline rewrite (1.1.6 – 1.1.13)

• Word-weighted sentence split + played/not_played categorization in agentcall-audio.js
• Backend TTS-ordering fix: drain-on-interrupt cancels pending TTS so the "speaking flash + ~1s of audio after barge-in" cycle is gone
• 2-second interruption window with AudioContext.suspend() for clean false-alarm recovery
• Conversational-style rule in SKILL.md: write tts.speak text the way you'd say it aloud; no markdown / emojis / symbols (Kokoro reads unknown chars by Unicode name — ^ → "circumflex", € → "euro")
• Bridge-side sanitizer normalizes em-dash + en-dash → , across all 6 bridge files
• Multi-sentence tts.speak now auto-splits per sentence in the bridge for sub-1s first-audio latency, then aggregates N backend tts.done events into ONE tts.done back to the agent — agent's 1:1 mental model preserved

State-machine refactors (1.1.0 – 1.1.7)

• VAD as a 3-state machine driven by transcript.final + 1.25 s cooldown
• Barge-in gate as a 3-state machine driven by transcript.final + 1.5 s cooldown
• Auto-thinking: bridge-visual broadcasts voice.state=thinking on every user.message in direct mode; clears on agent activity or 10 s fallback
• Avatar interrupted state: red flash on confirmed interruption, last-write-wins handoff to auto-thinking
• Gate raise-hand: bridges auto-raise the bot's hand if a gated tts.speak waits >10 s for the human to stop talking
• Non-blocking barge-in: tts.speak spawns as a background task with TTS-only lock so send_chat, set_state, and screenshare don't wait for the silence gate

Agent UX

• tasks.set agent command shows a work-in-progress task list below the avatar (HTTP-polled, no backend change)
• Screenshare debugging block in SKILL.md (curl + screenshot recipes)

SKILL.md cleanup (1.1.14)

• Trimmed to contract-level guidance; removed internal mechanism details
• Clarified tts.done semantics: it's the TTS generation-complete signal, not playback-complete

Source

Full history at https://github.com/jomon003/AGC-AgentCall-Skills (private). Public source code in this repo is the canonical install artifact.

Validation

$ npx -y @anthropic-ai/claude-code plugin validate .
✔ Validation passed

Archive

Attached agentcall-skill-1.1.14.zip is the same artifact you get by cloning this repo at this tag.