feat(voice): Add VOICE_STRICT_MODE for conditional voice notifications

[sti0]

Summary

Adds VOICE_STRICT_MODE environment variable to control voice notification behavior in the stop-hook-voice system. When enabled, voice notifications only trigger when an explicit COMPLETED message is detected in the assistant's response—eliminating unwanted "Completed task" fallback notifications.

Problem

The current implementation falls back to speaking "Completed task" whenever no COMPLETED pattern is found in the response. This creates unwanted voice notifications for:

• Intermediate responses during multi-step workflows
• Conversational exchanges that don't represent task completions
• Error messages or clarification requests

Solution

Introduce VOICE_STRICT_MODE environment variable with the following behavior:

VOICE_STRICT_MODE	COMPLETED Found	Result
unset / false	Yes	✅ Speak completion message
unset / false	No	✅ Speak "Completed task" (fallback)
true / 1	Yes	✅ Speak completion message
true / 1	No	🔇 Skip silently

Changes

Code (Packs/kai-voice-system/src/hooks/stop-hook-voice.ts)

• Add isStrictMode constant reading VOICE_STRICT_MODE env var (supports true and 1)
• Change extractCompletion() return type from string to string | null
• Return null instead of fallback when strict mode enabled and no COMPLETED found
• Add if (completion) guard in main() to skip notification when null

Documentation (Packs/kai-hook-system/README.md)

• Add VOICE_STRICT_MODE to environment variables table
• Bump version to 1.1.0
• Add changelog entry

Configuration

Enable in your environment or $PAI_DIR/.env:

VOICE_STRICT_MODE=true

Test Plan

• Default behavior preserved: Without VOICE_STRICT_MODE, fallback to "Completed task" still works
• Strict mode + COMPLETED present: Voice notification fires with extracted message
• Strict mode + no COMPLETED: Silent skip (no notification sent)
• Build verification: TypeScript compiles without errors

Backward Compatibility

✅ Fully backward compatible - Existing behavior unchanged when VOICE_STRICT_MODE is unset or false.

---

🤖 Generated with https://claude.com/claude-code

[danielmiessler]

Thank you @sti0 for the VOICE_STRICT_MODE concept! 🙏

Your voice system contributions have been fantastic. PAI v2.1 restructured everything (kai-* → pai-*), so this would need to be re-applied to pai-voice-system.

See the release: https://github.com/danielmiessler/PAI/releases/tag/v2.1.0