Aiden Live voice actions and 0.41.3

[sambitcreate]

Summary

• move Aiden Live to Gemini 3.8 Live Extended Thinking with explicit non-blocking Computer Use
• make per-action Computer Use approval voice-only with strict finalized-user-caption fencing
• remove visible orb button chrome and expose Google model visibility controls
• bump the release version to 0.41.3

Validation

• npm run build
• npm test
• npm run type-check
• npm run lint
• npm run test:gemini-live
• npm run test:voice
• npm run test:provider-model-catalog
• npm run test:branding
• npm run release:check-consumers

Remaining operator gate

• Real Google acceptance remains credentialed and requires the explicit operator flow documented in docs/gemini-live-google-acceptance.md.

[pullfrog]

Caution

Three correctness issues need to be resolved before merging:

• The new default model has no verified direct-Google Live + Computer Use contract and is marked tool_call: false in the only checked-in capability record.
• The voice approval baseline is derived from a render-updated ref, so notification ordering can admit a pre-prompt command.
• Caption IDs are reused across unsealed same-direction finalized input events, so a valid approval phrase can be rejected.

npm run test:gemini-live, npm run test:voice, npm run test:provider-model-catalog, npm run type-check, and npm run lint all pass. The credentialed real-Google acceptance was not run in this review environment.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using GPT Luna ｜ 𝕏}

[pullfrog]

Caution

The prior three findings are addressed, but two correctness issues remain in the new implementation.

Validation: npm run test:gemini-live passed (146 Live tests and 46 Computer Use tests); the worktree is clean. The credentialed real-Google acceptance remains unrun because it requires the documented operator/API-key flow.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using GPT Luna ｜ 𝕏}

[pullfrog]

ℹ️ No critical issues. Two minor edge cases are noted inline.

Reviewed changes

This incremental pass covered changes since the prior Pullfrog review at 5a3dc425: the new FIFO receipt state and the disabled-capability dock boundary.

• Queued finalized voice approval receipts separately from coalesced display captions and added batched-delivery coverage.
• Gated the dock presentation on the Live capability and added a default-off render regression.

Focused validation passed: npm run test:gemini-live (148 Live tests and 46 Computer Use tests), npm run test:voice, npm run type-check, and npm run lint.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using GPT Luna ｜ 𝕏}

[pullfrog]

Important

The latest Live command-gating change leaves two registered renderer contract tests failing; update them before merging.

Reviewed changes

This incremental pass reviewed the changes from faa031f2 to 32638fda, focusing on bounded voice-receipt retention and disabled Live command registration.

• Retained the earliest exact post-baseline voice decision across bounded receipt eviction and wired the retention floor through the Live approvals hook.
• Gated the assistant.open command with live.visible and moved the exact approval phrase parser into a shared module.

Focused validation passed: npm run test:gemini-live (149 Live tests and 46 Computer Use tests), npm run type-check, and npm run lint. The focused renderer contract run failed both stale command-shape assertions noted below.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using GPT Luna ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes

This incremental pass reviewed the renderer hotkey contract update since the prior Pullfrog review at 32638fda.

• Aligned the assistant.open source-contract assertion with the runtime live.visible enablement argument.

Focused validation passed: npx tsx --test renderer/lib/assistant-motion-contract.test.ts renderer/components/environment-subagents-contract.test.ts (32 tests), npm run test:gemini-live (149 Live tests and 46 Computer Use tests), npm run test:voice (72 passed, 1 skipped), npm run type-check, and npm run lint.

  ^{｜ View workflow run ｜ Using GPT Luna ｜ 𝕏}