Roadmap / known limitations

[leonidb]

This is the single tracking issue for planned work and known limitations. Kept in one place rather than scattered across many issues. Items here are candidates and current thinking - not commitments or release targets.

Known limitations

• Narration bleeds into mic when dictating - If Claude Code's /voice mode or a system-wide dictation tool (Wispr Flow, Superwhisper, macOS Dictation) is active while Audio Recap is narrating, TTS output can bleed across the microphone. The intent is to suppress narration while the mic is in use, need research - might be impossible for a thin plugin.

• No speaker identity across concurrent sessions - Running Audio Recap in two or more Claude Code sessions at once produces narrations from the same default voice with no label. There is no way to tell which session is speaking. Options under consideration: prepend the project directory name, or assign distinct macOS voices per session from a small pool. Advanced feature, not sure it is really relevant in practice, need to see if the use case makes sense first.

Planned improvements

• Silent during interactive prompts - Plan-mode approval gates and multi-option choice prompts are between-turn UI that don't fire the Stop hook. Narration is also needed in those cases. Requires a different hook event probably.

• Instant replay for /audio-recap:repeat - Currently /audio-recap:repeat re-synthesizes speech from cached text. Caching the rendered audio output would give zero-latency replay at no synthesis cost. First measure the difference, might not be worth it.

• Linux support - possibly with additional installation, not sure what's available for narration by default.