[adapter] OpenAI Codex: lifecycle hook or wrapper for /say

[omarelkhal]

Goal

Research OpenAI Codex (CLI or IDE) for a “response finished” event and pipe assistant text through the same Aftertone pipeline as Cursor / Claude (speak_summary_prepare.py → tts_daemon /say).

Context (May 2026)

Adapter	Status
Cursor	✅ Global afterAgentResponse
Claude Code	✅ Global Stop hook — #10
Codex	❌ This issue
OpenCode	Not started

Shared install: scripts/install.sh → ~/aftertone, config in .cursor/hooks/speak_summary.toml, daemon on localhost.

Deliverable

• Design note in docs/adapters/codex.md (can start docs-only PR)
• Optional proof-of-concept hook script (Windows + Linux)
• README / site row moves from “soon” to “available” when smoke-tested

Checklist (CONTRIBUTING.md § OpenAI Codex)

• Research lifecycle / hook API
• Speech adapter → prepare or /say (no bash-only trap on Windows without documenting it)
• Install root + shared TOML
• Spoken-summary guidance for models (<spoken_summary>, lang)
• Smoke test (reuse py/test_speak_summary_pipeline.sh patterns)
• (Optional) MCP + CLI parity — separate from speech hook

Help wanted

Comment with links to official Codex hook / extension docs. First PR can be research-only.