Robonix Webots CI Report

Provider: deepseek · Model: deepseek-chat · Generated On: 2026-07-13T15:04:18+00:00 · Stage: build · Confidence: high

The CI build failed because the speech service's post-codegen import verification could not find the 'speech_mcp' module. The build script runs `rbnx codegen` without the `--mcp` flag, so the MCP types directory is not generated, but the verification script expects it. The build log shows 'ModuleNotFoundError: No module named speech_mcp' at line 329. This is a build-time regression introduced by changes to the speech service's build script and package manifest.

PR Changes

• Added new service contracts: GetAudioBridgeInfo, GetHandsfreeStatus, SetHandsfree, WatchHandsfreeEvents, DetectWakeWord
• Added capability TOML files for bridge_info, wake_word, handsfree events, set_enabled, and status
• Modified liaison/submit.v1.toml and liaison/voice.v1.toml: changed kind from 'system' to 'service'
• Updated audio_client_bridge README, main.py, reverse.py, package_manifest.yaml, and tests
• Updated audio_driver/main.py: added exclusive mic lock, improved device selection, and deployment-configured device handling
• Updated sim/stop.sh: added voiceprint_service kill pattern
• Updated services/speech: README, package_manifest.yaml, pyproject.toml, build.sh, service.py, and added wake_word.py
• Added system/liaison/src/handsfree.rs: new module for hands-free voice mode orchestration
• Modified system/liaison/src/main.rs: integrated handsfree controller, registered new gRPC services
• Modified system/liaison/src/voice.rs: added barge-in support, no-speech timeout, VoiceSessionStream, and improved endpoint resolution
• Modified system/pilot/src/service.rs: added executor cancel_all_plans on abort_turn

Test Result

• No Webots simulation was started because the build step failed. The build summary shows 10 packages built successfully, 3 fetched, 0 skipped, and 1 failed (speech service). The caps.final.json shows Atlas was unreachable because the boot step never ran.

Likely Root Cause

• The speech service build script (services/speech/scripts/build.sh) runs `rbnx codegen -p "$PKG"` without the `--mcp` flag, but the post-codegen verification script (line 329) tries to import `speech_mcp`, which is only generated when `--mcp` is passed. The previous build script used `rbnx codegen --mcp` but the new one removed that flag.

Suggested Fix

• Add the `--mcp` flag back to the `rbnx codegen` command in services/speech/scripts/build.sh, or update the verification script to skip the MCP import when the MCP types directory is absent.

Risks Or Watchouts

• The speech service build script was refactored to support multiple backends (local/tencent/mock) and the codegen invocation was changed; the MCP flag was accidentally dropped.
• Other services (memsearch, voiceprint, scene) still use `rbnx codegen --mcp` and build successfully.
• The build failure prevents any simulation or scenario testing from running.

Evidence