Build, upload, and debug Arduino CLI projects for Waveshare ESP32-S3 Touch AMOLED boards, especially ESP32-S3-Touch-AMOLED-1.75C on macOS. Use when setting up arduino-cli, installing the Espressif esp32 core, compiling Waveshare Arduino-v3.3.5 examples, choosing the correct ESP32-S3 FQBN/options, detecting /dev/cu.usbmodem serial ports, flashing sketches, or validating hardware bring-up through serial monitor output.
Build, upload, and debug Arduino CLI projects for Waveshare ESP32-S3 Touch AMOLED boards, especially ESP32-S3-Touch-AMOLED-1.75C on macOS. Use when setting up arduino-cli, installing the Espressif esp32 core, compiling Waveshare Arduino-v3.3.5 examples, choosing the correct ESP32-S3 FQBN/options, detecting /dev/cu.usbmodem serial ports, flashing sketches, or validating hardware bring-up through serial monitor output.
Waveshare ESP32-S3 AMOLED
Overview
Use this skill to bring up Waveshare ESP32-S3 Touch AMOLED Arduino projects through arduino-cli, with reproducible CLI setup, vendor library handling, compile/upload commands, and serial validation.
Default Workflow
Inspect the hardware and local state:
Run arduino-cli version, arduino-cli core list, and arduino-cli board list.
Prefer /dev/cu.usbmodem* on macOS; the known local port has been /dev/cu.usbmodem83101.
Use esp32:esp32@3.3.5 for Waveshare examples/arduino (upstream renamed this from examples/Arduino-v3.3.5).
If esp32:esp32@3.3.10 is installed, expect the bundled GFX v1.6.4 library to fail around spiFrequencyToClockDiv; install 3.3.5 instead.
Even with esp32:esp32@3.3.5, a freshly re-cloned vendor repo can still fail on spiFrequencyToClockDiv (nullptr to uint32_t) if the vendor bumped GFX_Library_for_Arduino past what that core's single-arg spiFrequencyToClockDiv(uint32_t) supports — this is an upstream vendor/core mismatch, not a local misconfiguration; re-check both versions if it recurs.
Add the Espressif package URL if missing: https://espressif.github.io/arduino-esp32/package_esp32_index.json.
Arduino examples/libraries: examples/arduino (was examples/Arduino-v3.3.5 in older vendor checkouts).
The current Espressif core exposes Waveshare AMOLED 1.43/1.64/1.8/1.91/etc., but not a dedicated 1.75C FQBN. For 1.75C, use generic esp32:esp32:esp32s3 plus explicit options and the 1.75C pin_config.h.
Compile serially:
Do not run concurrent arduino-cli compile jobs into the same cache/build path.
Use --jobs 1, --clean, and a dedicated --build-path for reliable validation.
Upload and validate:
Upload to the detected /dev/cu.usbmodem* port.
On macOS, prefer raw serial capture with stty -f "$PORT" 115200 cs8 -cstopb -parenb -ixon -ixoff -echo followed by cat "$PORT".
Use arduino-cli monitor --config baudrate=115200,dtr=on,rts=off only as a fallback; it may open successfully but capture no bytes on this USB Serial/JTAG path.
Treat repeated sketch serial lines plus visible AMOLED output as the baseline pass condition.
For visual display validation:
Upload sketches/display_ocr_check when the project has it.
If OCR framing or orientation is uncertain, run make camera-aligner first. Copy the generated CAMERA_CROP='...' OCR_ROTATE=... environment values.
If the board appears upside down in the camera, prefer DISPLAY_ROTATION=2 make visual-smoke so the sketch renders upright text for OCR.
Run CAMERA_CROP='...' OCR_ROTATE=... DISPLAY_ROTATION=... make visual-smoke to capture one camera frame with ffmpeg and OCR it with macOS Vision.
For bright AMOLED overexposure or color mismatch, start with the calibration sketch: DISPLAY_ROTATION=2 DISPLAY_BRIGHTNESS=96 OCR_PREPROCESS_MODE=color CAMERA_EXPOSURE_POINT=0.5,0.65 CAMERA_FOCUS_POINT=0.5,0.65 COLOR_SWATCH_CHECK=1 make visual-smoke. It renders large OK text plus red/green/blue/yellow swatches, then reports the largest connected color blocks, average RGB, centroids, and row/order geometry.
Use make camera-color-check IMAGE=/path/to/raw.jpg to check red/green/blue/yellow swatches on an existing camera image without OCR. Tune with COLOR_SWATCH_MIN_X_GAP, COLOR_SWATCH_MAX_Y_SPREAD, or COLOR_SWATCH_GEOMETRY=0 only when the camera angle demands it.
Use CAMERA_RAW_IMAGE=/path/to/raw.jpg ./scripts/camera-ocr.sh to retune crop, rotation, or OCR_PREPROCESS_MODE without recapturing.
Camera capture is bounded by CAMERA_CAPTURE_TIMEOUT; if it times out before saving a frame, debug macOS camera availability or another app owning the camera before changing board firmware.
Run make camera-ready before expensive visual smokes when camera availability is uncertain; it uses Swift capture only and fails unless a frame is saved. Run make camera-diagnose when capture fails; it records camera inventory, Swift AVFoundation status, related processes, and bounded video-only capture probes under .logs/. Use CAMERA_DIAGNOSE_FFMPEG=0 when you want to avoid ffmpeg device enumeration. If Swift diagnostics report running=true frames=0 drops=0, the capture session started but macOS/USB delivered no video buffers, so fix the host camera pipeline before rerunning board smokes.
Run make camera-reset to restart macOS camera services and rerun the strict camera-ready check. It does not quit chat apps unless is set.
For official demo bring-up:
Run make official-demos to list the Waveshare Arduino demo manifest.
Run make official-build-all before debugging higher-level AI features; all official Arduino examples should compile first.
Run make official-audio-preflight when audio should stay quiet; it compiles only the official ES7210/ES8311 demos and checks source/serial markers without uploading or using audio devices.
Run make official-audio-physical-plan to inspect the gated ES7210/ES8311 physical-audio plan without uploading or using audio devices.
Run ALLOW_AUDIO=1 make official-audio-physical-smoke only during an allowed audio window; ES7210 uses a host stimulus while serial capture is open, and ES8311 requires supervised OFFICIAL_AUDIO_OUTPUT_CONFIRM=heard.
Run make official-coverage for a read-only audit of build artifacts, source presence, quiet audio marker readiness, and existing physical smoke logs across every official demo.
Run SMOKE_SECONDS=8 make official-smoke DEMO=01-helloworld to upload the official display baseline and verify runtime serial output.
Run OFFICIAL_VISUAL_SMOKE=1 OFFICIAL_OCR_EXPECTED="Hello World" SMOKE_SECONDS=8 make official-smoke DEMO=01-helloworld when camera OCR should run after the serial gate; treat saved frames with failed OCR as partial visual evidence.
Run OFFICIAL_VISUAL_STABLE_MARKER=1 OFFICIAL_VISUAL_SMOKE=1 SMOKE_SECONDS=8 make official-smoke DEMO=01-helloworld for a staged-only stable OK OCR marker on the official HelloWorld baseline; it leaves vendor source directories untouched and defaults OCR_ROTATE=180 for the current camera mount.
Official smoke uses scripts/serial-capture.py to open capture before RTS reset; keep this path for one-shot setup logs.
03-power-axp2101 uses a staged-only OFFICIAL_POWER_WIFI_TIMEOUT_MS patch in automation so invalid vendor Station credentials do not block AP, PMU, and LVGL initialization; treat the resulting Setup done as PMU/LVGL evidence, not Station Wi-Fi proof.
The project runner stages vendor examples under .arduino-build/official-sketches/<id> because several official .ino filenames do not match their parent folder names, which arduino-cli requires.
For feature coverage auditing:
Run make feature-matrix-check to verify all tracked feature directions have matching Makefile targets, scripts/sketches, docs, and Skill helper wiring.
Run make feature-matrix-doc after changing config/feature-matrix.tsv to regenerate docs/hardware-verification-matrix.md.
Run make hardware-evidence-audit to identify lanes with missing Verified Locally sections or missing smoke-suite evidence.
Run make hardware-evidence-doc to regenerate docs/hardware-evidence-audit.md.
Run make visual-evidence-audit to identify lanes that lack recorded camera/OCR artifacts in their local evidence sections; it also reports the latest camera-ready preflight and camera-diagnose recommendation when visual suite runs are blocked by the host camera.
Run make visual-evidence-doc to regenerate docs/visual-evidence-audit.md.
Run make goal-completion-audit for the stricter requirement-level completion gate; use python3 scripts/goal-completion-audit.py --strict only when a non-zero result should fail CI or handoff.
Run make goal-completion-doc to regenerate docs/goal-completion-audit.md.
Run make evidence-index for a consolidated handoff map across strict completion, feature docs, hardware suite, visual OCR, and remaining-gate status. Run make evidence-index-doc to regenerate docs/evidence-index.md.
Run make remaining-gates-preflight to refresh the safe side of incomplete lanes without flashing XiaoZhi or using audio hardware, then make remaining-gates-doc to regenerate docs/remaining-gates-preflight.md from the latest summary. It must still leave strict completion at 9/13 until approved physical/external/manual evidence exists. Web AI Button physical tap is recorded as skipped/manual-required by default unless REMAINING_GATES_ARGS=--include-manual is passed.
Run make remaining-gates-runbook to regenerate docs/remaining-gates-supervised-runbook.md before handing off the four remaining gates. It keeps the required approval, supervised command, expected evidence, and completion boundary visible for audio, XiaoZhi flash, and physical tap work.
For XiaoZhi AI bring-up:
Run make xiaozhi-latest to locate the latest official waveshare-esp32-s3-touch-amoled-1.75c release asset.
Run make xiaozhi-inspect before flashing; it should confirm the zip contains merged-binary.bin.
Run make xiaozhi-preflight for the non-destructive readiness gate; it hashes merged-binary.bin, reports esptool/serial/source/ESP-IDF readiness, and confirms destructive=0 audio=0.
Run make xiaozhi-readiness for the default no-audio readiness bundle. It runs preflight, source check, ESP-IDF env/build, and verifies the latest rollback image. Set XIAOZHI_READINESS_BACKUP=1 immediately before an approved flash when the current board flash must be freshly captured.
If make xiaozhi-preflight reports release_source=cache, it used a local firmware zip because live release metadata failed; set XIAOZHI_RELEASE_CACHE_FALLBACK=0 when live upstream freshness is required.
Run make xiaozhi-backup before any XiaoZhi flash; it reads the current board flash to .vendor/xiaozhi/backups/ and prints xiaozhi_backup_summary ... destructive=0 audio=0. The backup path defaults to --no-stub at XIAOZHI_BACKUP_BAUD=115200 because long stub reads can stop on this USB Serial/JTAG path.
Run BACKUP=<path> CONFIRM=--yes make xiaozhi-restore only when you intentionally want to write a saved flash image back to the board.
Run make xiaozhi-flash only when the user is ready to replace the current Arduino demo with XiaoZhi firmware.
After an approved XiaoZhi flash, run make xiaozhi-runtime-check before any audio interaction. It captures serial logs after an RTS reset, looks for XiaoZhi onboarding or activation markers, and reports xiaozhi_runtime_summary ... destructive=0 audio=0.
Run make xiaozhi-visual-check for a camera-only OCR gate, or make xiaozhi-runtime-visual-check for serial plus AMOLED OCR after an approved flash. These gates do not use microphone or speaker hardware.
For source builds, run make xiaozhi-source-clone then make xiaozhi-source-check; the local defaults select CONFIG_BOARD_TYPE_WAVESHARE_ESP32_S3_TOUCH_AMOLED_1_75C=y.
For the self-developed cloud AI terminal:
Run make cloud-ai-build to compile the board-side display/serial terminal.
Run make cloud-ai-smoke to upload it and verify the host serial relay reaches AI_DISPLAYED.
Run make cloud-ai-pipeline-smoke to upload it and verify the silent ASR -> LLM -> TTS serial pipeline reaches PIPELINE_DONE.
Run make cloud-ai-cache-smoke to upload it and verify board-local NVS cache, session id, cloud request id, recoverable cloud error metrics, and STATE? runtime status commands.
Run CLOUD_AI_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make cloud-ai-smoke when the camera is positioned for OCR; serial verifies AI OK, while OCR should at least see stable OK.
Treat this as the verified non-audio cloud terminal control-plane slice; pipeline smoke does not use microphone or speaker hardware. ES7210 microphone input and ES8311 speaker output still need dedicated audio stream validation in an allowed audio window.
For the local web AI button:
Run make local-ai-server to start the Mac-side HTTP AI trigger server.
Run make web-ai-button-build to compile the Wi-Fi + touch-button sketch.
Run make web-ai-button-smoke to start the local server, upload the sketch, pass ignored .env Wi-Fi credentials over serial, configure the board endpoint to the Mac LAN IP, and verify WEB_AI_RESPONSE status=ok.
Run make web-ai-button-tap-smoke for the supervised physical tap gate; it follows the same setup path but waits for WEB_AI_TOUCH_EVENT, WEB_AI_TRIGGER source=touch, and WEB_AI_RESPONSE status=ok instead of sending serial TRIGGER.
Use WEB_AI_BUTTON_VISUAL_SMOKE=1 make web-ai-button-smoke when camera OCR should also verify the board reaches the AI response screen.
Do not hard-code or commit SSIDs, passwords, local IPs, or AI command credentials. The default local server mode is mock; use WEB_AI_SERVER_MODE=command and AI_TRIGGER_COMMAND=... only for supervised local AI experiments.
This path is safe for late-night validation because it does not play audio or use the host microphone.
For microphone/audio-front-end validation:
Run make audio-vad-build to compile the ES7210 microphone probe.
Run make audio-afe-readiness when audio should stay quiet and you only need machine-readable AFE readiness; it rebuilds and reports implemented source/build/checker readiness versus source-integration and physical-audio requirements.
Run make audio-vad-preflight when audio should stay quiet; it rebuilds and checks artifacts, serial port, ES7210 source markers, checker options, and config/audio-afe-profile.tsv without uploading, playing stimulus, or opening audio devices.
Treat the AFE profile as coverage metadata: ES7210 capture and ESP-SR VAD are implemented, while AEC, noise suppression, and real WakeNet audio frames still require source integration and physical audio evidence.
Run make audio-vad-smoke to upload it, play a host-side say stimulus, and validate RMS/peak serial metrics.
Use AUDIO_VAD_REQUIRE_SPEECH=1 make audio-vad-smoke only when the host speaker is close enough for reliable VAD.
Use AUDIO_VAD_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make audio-vad-smoke when camera OCR should verify the screen reaches OK.
For speaker/audio-output validation:
Run make speaker-output-build to compile the ES8311 speaker tone probe.
Run make speaker-output-smoke to upload it, trigger a 1 kHz / 1.5 kHz tone over serial, and validate the output through host microphone capture.
Use SPEAKER_AUDIO_DEVICE=<avfoundation audio index> when the default camera microphone is not close to the board speaker.
Use SPEAKER_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make speaker-output-smoke when camera OCR should verify the screen reaches OK.
Do not run audible speaker or microphone smoke tests late at night unless the user explicitly asks for them.
For PMU/IMU sensor validation:
Run make sensor-status-build to compile the AXP2101 + QMI8658 status probe.
Run make sensor-status-smoke to upload it and validate silent serial metrics from the PMU and IMU.
Use SENSOR_STATUS_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make sensor-status-smoke when camera OCR should verify the screen reaches OK.
This path is safe for late-night validation because it does not play audio or use the host microphone.
For battery and low-power lifecycle validation:
Run make power-lifecycle-build to compile the AXP2101 power lifecycle probe.
Run make power-lifecycle-smoke to upload it and validate DIM, STANDBY, ACTIVE, brightness, capacity, load-profile, wake, and runtime-estimate serial behavior.
Use POWER_REQUIRE_BATTERY=1 make power-lifecycle-smoke only when a battery is physically connected; the default smoke should not fail USB-only benches.
Use POWER_LIFECYCLE_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make power-lifecycle-smoke when camera OCR should verify the screen reaches OK; the script defaults to color preprocessing plus centered focus/exposure for this probe because generic gray OCR can miss the white marker.
This path is safe for late-night validation because it does not play audio or use the host microphone. Its standby mode keeps serial alive and should not be treated as proof of true ESP32 deep sleep or measured current draw.
For Wi-Fi connectivity validation:
Run make wifi-connectivity-build to compile the ESP32-S3 Wi-Fi scan/join probe.
Run make wifi-connectivity-smoke to upload it and validate Wi-Fi radio initialization plus scan completion without storing credentials.
Use WIFI_TEST_SSID=... WIFI_TEST_PASSWORD=... make wifi-connectivity-smoke only for a supervised join check; never commit SSIDs or passwords.
Use WIFI_CONNECTIVITY_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make wifi-connectivity-smoke when camera OCR should verify the screen reaches OK.
This path is safe for late-night validation because it does not play audio or use the host microphone. The default serial log avoids printing SSID names.
For touch controller validation:
Run make touch-status-build to compile the CST92xx touch controller probe.
Run make touch-status-smoke to upload it and validate the touch controller is online through serial status and AMOLED OCR.
Use TOUCH_REQUIRE_EVENT=1 make touch-status-smoke only when a human can touch the screen during the smoke window.
Use TOUCH_STATUS_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make touch-status-smoke when camera OCR should verify the screen reaches OK.
For combined non-audio app validation:
Run make interaction-dashboard-build to compile the combined display, touch-controller, PMU, and IMU dashboard.
Run make interaction-dashboard-smoke to upload it and drive page changes, serial-simulated IMU gesture handling, brightness, standby, and wake transitions without requiring a human tap.
Use INTERACTION_DASHBOARD_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make interaction-dashboard-smoke when camera OCR should verify the final dashboard page reaches OK.
This path is safe for late-night validation because it does not play audio or use the host microphone.
For dedicated IMU interaction validation:
Run make imu-interaction-build to compile the QMI8658 wrist wake, shake switch, posture menu, and step counter probe.
Run make imu-interaction-smoke to upload it and validate deterministic serial-injected IMU samples for WRIST_WAKE, SHAKE_SWITCH, POSE_MENU, STEP, and MENU_NEXT.
Use IMU_INTERACTION_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make imu-interaction-smoke when camera OCR should verify the screen reaches OK; serial verifies the IMU event details.
This path is safe for late-night validation because it does not play audio, use the host microphone, or require physically shaking the board.
For LVGL visual-agent validation:
Run make lvgl-visual-agent-build to compile the repo-owned LVGL tabview app.
Run make lvgl-visual-agent-smoke to upload it and validate LVGL initialization, display flush, CST92xx touch input registration, chat bubbles, cards, settings, agent thoughts, and tab/page changes over serial.
Use LVGL_VISUAL_AGENT_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 DISPLAY_BRIGHTNESS=96 make lvgl-visual-agent-smoke when camera OCR should verify the screen reaches the large OK marker; serial verifies LVGL tabview, touch registration, chat, cards, settings, and agent thoughts.
Treat this as the agent-specific LVGL UI slice; the official 05-lvgl-widgets demo remains the vendor baseline.
This path is safe for late-night validation because it does not play audio or use the host microphone.
For desktop AI widget validation:
Run make desk-widget-build to compile the serial-driven desktop widget.
Run make desk-widget-smoke to upload it and validate CI/GitHub/alert/calendar/timer/AI-summary pages without network credentials.
Run make desk-widget-relay-smoke to validate the host event adapter that maps mock, JSON, or HTTP events into the widget serial protocol.
Use DESK_WIDGET_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make desk-widget-smoke when camera OCR should verify the screen reaches OK.
This path is safe for late-night validation because it does not play audio or use the host microphone.
For IoT control panel validation:
Run make iot-panel-build to compile the serial-driven Home Assistant / MQTT / HTTP control panel.
Run make iot-panel-smoke to upload it and validate device state changes, explicit IOT_HA Home Assistant service calls, MQTT-style inbound updates, HTTP-style outbound actions, and scenes without Wi-Fi credentials.
Run make iot-panel-relay-smoke to validate the host event adapter that maps mock, JSON, or HTTP smart-home events into the panel serial protocol, including board-side IOT_HA output and the ha= state counter.
Use IOT_PANEL_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make iot-panel-smoke when camera OCR should verify the screen reaches the stable IOT marker; serial verifies device, Home Assistant, MQTT, HTTP, and scene details.
This path is safe for late-night validation because it does not play audio or use the host microphone.
For offline voice-control state-machine validation:
Run make offline-voice-build to compile the WakeNet/MultiNet-facing serial harness.
Run make offline-voice-smoke to upload it and validate pre-wake command rejection, wake events, command recognition, runtime command add/modify/delete, continuous mode, sleep/wake state, and local actions without using the microphone.
Use OFFLINE_VOICE_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 OCR_ROTATE=180 make offline-voice-smoke when camera OCR should verify the screen reaches the stable VOICE marker.
Treat this as the deterministic control-plane gate before wiring real ESP-SR audio frames and models.
This path is safe for late-night validation because it does not play audio or use the host microphone.
For TinyML / IMU classifier validation:
Run make tinyml-imu-model-check to validate the checked-in nearest-centroid model metadata and validation set.
Run make tinyml-imu-build to compile the QMI8658 TinyML classifier.
Run make tinyml-imu-smoke to upload it, verify the board-reported model hash, disable live IMU mode, inject deterministic serial feature vectors, and verify REST, TILT_LEFT, TILT_RIGHT, FACE_UP, and SHAKE labels.
Use TINYML_IMU_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 OCR_ROTATE=180 make tinyml-imu-smoke when camera OCR should verify the screen reaches the stable TINY marker.
Treat this as a small TinyML automation model. The current embedded nearest-centroid classifier can be replaced by ESP-DL or a larger trained model later without removing the deterministic serial sample gate.
This path is safe for late-night validation because it does not play audio or use the host microphone.
For ESP-Claw / OpenClaw agent harness validation:
Run make esp-claw-agent-build to compile the Arduino compatibility harness for the ESP-Claw/OpenClaw direction.
Run make esp-claw-agent-smoke to upload it and validate local rule add, Lua-style rule loading, event sensing, rule decision, MCP-style tool registration/invocation, IM chat input, tagged memory put/get, and LLM fallback routing over serial.
Use ESP_CLAW_AGENT_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 DISPLAY_BRIGHTNESS=96 make esp-claw-agent-smoke when camera OCR should verify the screen reaches OK; serial verifies rules, MCP calls, IM chat, memory, and fallback actions.
Treat this as a deterministic compatibility harness, not the official ESP-Claw firmware image. It exists so automation can prove the agent loop before IM credentials, Wi-Fi, and full ESP-Claw source builds are introduced.
This path is safe for late-night validation because it does not play audio or use the host microphone.
For NES Contra emulator validation:
Run make nes-contra-preflight to inspect the local Contra disassembly checkout, cc65 tools, and user-supplied ROM state without modifying tracked files.
Run make nes-contra-build to compile the diagnostic emulator lane scaffold.
Run make nes-contra-smoke to upload the scaffold and validate NES_CONTRA_READY, CAPS?, ROM?, serial input injection, and frame progress.
Use NES_CONTRA_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 make nes-contra-smoke when camera OCR should verify the large OK marker before real game pixels are used as evidence.
Treat this lane as partial until a reviewed emulator core boots a Mapper 2 ROM; do not commit Contra ROM bytes or generated ROM headers.
For Skill automation wiring:
Run scripts/waveshare-arduino-cli.sh verify <project-dir> from this skill to prove the agent-facing entrypoint can inspect the toolchain, see the USB board, list official demos, and clean-compile cloud_ai_terminal, web_ai_button, audio_vad_probe, speaker_output_probe, sensor_status_probe, power_lifecycle_probe, wifi_connectivity_probe, touch_status_probe, interaction_dashboard, imu_interaction_probe, lvgl_visual_agent, desk_widget, iot_control_panel, offline_voice_control, tinyml_imu_classifier, esp_claw_agent, and nes_contra_emulator.
Run scripts/waveshare-arduino-cli.sh feature-matrix <project-dir> check to validate coverage metadata before claiming all tracked directions are wired.
Run scripts/waveshare-arduino-cli.sh visual-evidence <project-dir> audit to check camera/OCR evidence coverage through the Skill helper.
Run scripts/waveshare-arduino-cli.sh evidence-index <project-dir> to print the consolidated evidence map, or pass doc/markdown to regenerate or render it.
Run scripts/waveshare-arduino-cli.sh ok-qoder-evidence <project-dir> to build, upload, serial-smoke, and camera-OCR the default OK Qoder hello sketch into a committed evidence pack under docs/evidence/.
Run make claude-skill-smoke from this repo to ask the local Claude CLI to read the repo Skill and invoke this helper for the non-audio visual smoke path. The wrapper records Claude's transcript plus the command log under .logs/, and CLAUDE_SKILL_SMOKE_MODE=audit switches to a no-upload feature-matrix check.
verify/doctor is intentionally compile-only; it does not upload firmware or run camera OCR.
Run explicit hardware smokes when the user wants board validation:
CLOUD_AI_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 scripts/waveshare-arduino-cli.sh cloud-ai <project-dir> smokeCLOUD_AI_VISUAL_SMOKE=1 DISPLAY_ROTATION=2 scripts/waveshare-arduino-cli.sh cloud-ai <project-dir> pipelinescripts/waveshare-arduino-cli.sh cloud-ai <project-dir> cache
and
.
Known 1.75C FQBN
Use this FQBN unless a future Espressif core adds a real 1.75C board profile:
If upload fails, retry with UploadSpeed=460800 before changing board families.
Helper Script
Use scripts/waveshare-arduino-cli.sh from this skill for a portable setup/build/upload/monitor wrapper when the target project does not already provide scripts.
If the project contains scripts/setup.sh, scripts/build.sh, scripts/upload.sh, or scripts/monitor.sh, prefer those project scripts because they encode project-local sketch paths.
For visual validation in this repo, prefer:
SMOKE_SECONDS=8 ./scripts/smoke.sh
make camera-aligner
make camera-diagnose
make camera-ready
make camera-reset
make ok-qoder-evidence
make feature-matrix-check
make feature-matrix-doc
make hardware-evidence-audit
make hardware-evidence-doc
make visual-evidence-audit
make visual-evidence-doc
make goal-completion-audit
make goal-completion-doc
make evidence-index
make evidence-index-doc
make remaining-gates-list
make remaining-gates-preflight
make remaining-gates-doc
make remaining-gates-runbook
make hardware-smoke-list
make hardware-smoke-suite
make visual-smoke
make official-demos
make official-build-all
make official-audio-preflight
make official-coverage
SMOKE_SECONDS=8 make official-smoke DEMO=01-helloworld
OFFICIAL_VISUAL_SMOKE=1 OFFICIAL_OCR_EXPECTED="Hello World" SMOKE_SECONDS=8 make official-smoke DEMO=01-helloworld
OFFICIAL_VISUAL_STABLE_MARKER=1 OFFICIAL_VISUAL_SMOKE=1 SMOKE_SECONDS=8 make official-smoke DEMO=01-helloworld
make xiaozhi-latest
make xiaozhi-inspect
make xiaozhi-preflight
make xiaozhi-readiness
make xiaozhi-backup
make xiaozhi-runtime-check
make xiaozhi-visual-check
make xiaozhi-runtime-visual-check
make xiaozhi-source-check
make xiaozhi-idf-env
make xiaozhi-idf-build
make cloud-ai-build
make cloud-ai-smoke
make cloud-ai-pipeline-smoke
make cloud-ai-cache-smoke
make local-ai-server
make web-ai-button-build
make web-ai-button-smoke
make web-ai-button-tap-smoke
make audio-vad-build
make audio-afe-readiness
make audio-vad-preflight
make audio-vad-smoke
make speaker-output-build
make speaker-output-smoke
make sensor-status-build
make sensor-status-smoke
make power-lifecycle-build
make power-lifecycle-smoke
make wifi-connectivity-build
make wifi-connectivity-smoke
make touch-status-build
make touch-status-smoke
make interaction-dashboard-build
make interaction-dashboard-smoke
make imu-interaction-build
make imu-interaction-smoke
make lvgl-visual-agent-build
make lvgl-visual-agent-smoke
make desk-widget-build
make desk-widget-smoke
make desk-widget-relay-smoke
make iot-panel-build
make iot-panel-smoke
make iot-panel-relay-smoke
make offline-voice-build
make offline-voice-smoke
make tinyml-imu-model-check
make tinyml-imu-build
make tinyml-imu-smoke
make esp-claw-agent-build
make esp-claw-agent-smoke
make nes-contra-preflight
make nes-contra-build
make nes-contra-smoke
/Users/phodal/.codex/skills/waveshare-esp32s3-amoled/scripts/waveshare-arduino-cli.sh verify /path/to/project
References
Read references/board.md when you need board facts, source links, library names, or troubleshooting notes.
CAMERA_RESET_QUIT_CHAT_APPS=1
Pass only if OCR sees OK; use the saved raw/processed images to debug focus, glare, rotation, or garbled output.
Run make hardware-smoke-list to inspect the default serialized non-audio smoke selection.
Run make hardware-smoke-suite HARDWARE_SMOKE_ARGS="--target <id>" for a narrow serialized hardware lane; the default suite skips audio, conditional, and external lanes. If --with-visual is used, the suite runs make camera-ready before any target upload and aborts when the host camera cannot save a frame.
Run make hardware-smoke-suite HARDWARE_SMOKE_ARGS="--target xiaozhi-ai --allow-external" for the non-destructive XiaoZhi firmware archive check; it does not flash firmware or use audio hardware.
Treat matrix status values as coverage metadata, not proof that a partial or external feature is complete.
Run make xiaozhi-idf-env to prove the local ESP-IDF shell can be activated without flashing firmware or using audio hardware.
Run make xiaozhi-idf-build to compile XiaoZhi from source without flashing; the script prints xiaozhi_idf_build_summary ... destructive=0 audio=0 on success.
The current source requires ESP-IDF >=5.5.2; the local default is .vendor/esp-idf-v5.5.4 with ~/.espressif/python_env/idf5.5_py3.14_env.
