Catalog

Feedback and utilities

saysay

Speaks every response aloud on macOS, summarizing screen content into spoken language with a per-user phonetic dictionary and serialized, ducked playback across sessions.

Claude + Codex 2.0.9 1 skill

Speech mode via the macOS say command. Once enabled, the agent speaks every response aloud, translating screen content into spoken language instead of reading it out verbatim.

Use it for hands-busy or eyes-away sessions where the agent should keep the operator informed audibly while still writing the normal text answer. It is a local macOS feature; no speech output leaves the machine.

Commands

/saysay

Enter speech mode. Subsequent responses are spoken aloud after rendering.

/saysay off

Exit speech mode.

Requirements

macOS, for the say binary. Node.js on $PATH for the post-update broadcast check (bin/check-broadcast). The saysay and say-phonetic commands ship in the plugin's bin/ directory, which Claude Code and Codex add to $PATH automatically, so there is no manual install step.

Optional: a "Saysay Duck" / "Saysay Unduck" pair of macOS Shortcuts lets saysay lower other audio while it speaks. When the shortcuts are absent, saysay simply skips ducking.

Phonetic mappings

say-phonetic keeps a per-user pronunciation dictionary so that names, acronyms, and code identifiers come out the way you want. Mappings live in ${LAICLUSE_HOME:-$HOME/.laicluse}/saysay/phonetics.json. A dictionary left over from an earlier ~/.local/share/saysay/phonetics.json install is migrated into the new root automatically on first use.

say-phonetic add "laicluse" "lay-clue-zee"
say-phonetic remove "laicluse"
say-phonetic list