Esmail-sarhadi/esp32-agent-skills

1. **Brownout when Wi-Fi starts** — see: `Brownout detector was triggered` right after `WiFi.begin()`. cause: the radio's transmit peaks pull ~350 mA and a thin USB cable or a small LDO sags below 2.5 V. fix: 470 µF bulk capacitor at the mo

2. **ESP32-CAM browns out at boot** — see: reboot loop as soon as the camera initialises. cause: camera and Wi-Fi start together and exceed the board's regulator. fix: separate 5 V supply with a large capacitor; the on-board regulator is no

3. **Board will not enter download mode** — see: `Failed to connect ... Timed out waiting for packet header`. cause: the auto-reset circuit did not assert GPIO0 low, or the cable is charge-only. fix: hold BOOT, tap EN, release; try another 

4. **Invalid head of packet** — see: `Failed to connect to ESP32: Invalid head of packet (0x00)`. cause: something else is talking on the UART at boot, or the baud is wrong. fix: disconnect other serial devices, lower the upload speed to 11

5. **The port does not exist** — see: no COM port at all. cause: missing CP210x/CH340 driver, or the board is in a state where the USB bridge is unpowered. fix: install the bridge driver, check Device Manager, try a powered hub. · [73923341

6. **ESP32-CAM sends nothing on serial** — see: upload succeeds, monitor is silent. cause: the camera boards power the USB-serial chip only when GPIO0 is low, and the sketch reconfigures the UART pins. fix: use a USB-TTL adapter on U0TXD/U0

7. **Stack canary watchpoint triggered** — see: `Debug exception reason: Stack canary watchpoint triggered`. cause: a stack buffer overflow, usually a large local array or `sprintf` into a small buffer. fix: move large buffers to static or 

8. **TG1WDT_SYS_RST on minimal code** — see: an ESP32-C3 resets immediately even with an empty sketch. cause: a bootloader or partition mismatch, often after flashing a binary built for another target. fix: erase the flash (`esptool erase_f

9. **Board stops booting after a partition change** — see: boot loop, or `invalid header` in the ROM log. cause: `boot_app0.bin` missing, or the app partition offset no longer matches the table. fix: reflash bootloader, partitions and boot_

10. **Driving GPIO6-11 kills the board** — see: it compiles, flashes, then never boots again. cause: those pins are the SPI flash bus. fix: never use them; `pincheck.py` rejects them. · [65125302](https://stackoverflow.com/q/65125302)

11. **`INPUT_PULLUP` does nothing on GPIO34-39** — see: a button reads floating noise. cause: those pins are input-only and have no internal pull-up or pull-down at all. fix: external 10 kΩ resistor, or move the button to a pin that has one

12. **It boots on the bench, fails in the enclosure** — see: works until power is cycled. cause: a strapping pin (0, 2, 12, 15) is held at the wrong level by attached hardware at reset. fix: keep GPIO12 low at boot, do not load GPIO0/2/15 w

13. **The on-board LED will not blink** — see: `digitalWrite(2, HIGH)` changes nothing. cause: on many boards the LED is on GPIO2 but wired active-low, or it is a WS2812 on a different pin. fix: invert the logic, and check the board's schem

14. **`digitalWrite` is too slow** — see: a bit-banged protocol misses timing. cause: the Arduino API costs microseconds per call. fix: direct register access (`GPIO.out_w1ts`), or better, use the RMT/LEDC peripheral. · [52348930](https://s

15. **Touch pins stop working with Wi-Fi** — see: `touchRead` returns garbage. cause: the touch peripheral shares hardware with the radio. fix: do not mix touch and Wi-Fi on the same channel plan; read touch only while the radio is idle. · 

16. **A pin works as input but not output** — see: the output never changes. cause: it is one of GPIO34-39, which have no output driver. fix: pick another pin; there is nothing to enable. · [53324715](https://stackoverflow.com/q/53324715)

17. **Board reboots when a relay switches** — see: reset exactly when the load turns on. cause: inductive kickback on the GPIO rail, or a ground loop. fix: flyback diode across the coil, opto-isolation, separate supply for the relay board. 

18. **`analogRead` returns 0 once Wi-Fi connects** — see: readings worked before `WiFi.begin()`. cause: ADC2 is unavailable while the radio is on; there is no workaround. fix: move the sensor to ADC1 (GPIO32-39). · [65125302](https://stacko

19. **The top of the range is compressed** — see: values flatten near 3.2 V. cause: at 11 dB attenuation the ADC is not linear above ~2.5 V. fix: use a divider so the signal lands mid-range, or calibrate with `analogReadMilliVolts`. · [5683

20. **Measuring battery voltage** — see: a divider reads low and drifts. cause: the sample-and-hold capacitor loads a high-impedance divider. fix: keep the divider under 10 kΩ, add 100 nF at the pin. · [56266857](https://stackoverflow.com/q

21. **Reading VCC without an external pin** — see: no way to know the supply. cause: there is no internal VCC channel. fix: measure an internal reference through a known divider, or use `esp_adc_cal` with a factory-calibrated reference. · [

22. **Readings are noisy** — see: ±20 counts of jitter. cause: the ADC is noisy by design. fix: average N samples (noise drops by √N), add a capacitor, keep the sensor wire short. · [65725609](https://stackoverflow.com/q/65725609)

23. **Sampling audio or a fast signal** — see: `analogRead` cannot keep up. cause: it is a slow, blocking call. fix: use I2S in ADC mode with DMA; it samples continuously without the CPU. · [65725609](https://stackoverflow.com/q/65725609)

24. **The sensor is never detected** — see: `i2c.scan()` finds nothing. cause: no pull-ups, wrong pins, or the device is on a different address than the datasheet's default. fix: 4.7 kΩ to 3.3 V on SDA and SCL, confirm the address with a sc

25. **SCL stuck low** — see: the bus is dead until power cycle. cause: a slave is holding the line after a partial transfer. fix: clock SCL manually nine times to let it release, then issue a STOP. · [79237471](https://stackoverflow.com/q/7

26. **`ENODEV` but the scan works** — see: MicroPython finds the address then throws `[Errno 19]`. cause: the driver was constructed with the wrong address or a stale bus object. fix: recreate the bus, pass the scanned address explicitly. ·

27. **SPI transfer ends early** — see: the transaction stops mid-buffer. cause: the length was given in bytes where the driver expects words, or CS was released too soon. fix: check the length units and keep CS asserted for the whole transa

28. **MPU-6050 FIFO returns nonsense** — see: the first samples are correct then garbage. cause: the FIFO is read without checking the count register, or without clearing it. fix: read `FIFO_COUNT`, read exactly that many bytes, reset the F

29. **I2C works alone, fails with Wi-Fi** — see: timeouts only when the radio is active. cause: the radio's interrupts delay the bus, and the default timeout is short. fix: raise the I2C timeout, lower the clock to 100 kHz, retry the transa

30. **Two devices with the same address** — see: both respond, both return the other's data. cause: fixed-address parts cannot share a bus. fix: use a TCA9548A multiplexer, or move one device to a second I2C bus. · [79330446](https://stacko

31. **Garbled serial output** — see: readable at first, then symbols. cause: baud mismatch, or the monitor is attached to the wrong UART (USB CDC vs UART0). fix: match the baud exactly, and print to the same port the monitor is on. · [79670

32. **`Serial.print` and `Serial.printf` disagree** — see: different values for the same variable. cause: the format specifier does not match the type (`%d` for a `long`, `%f` for a double that was promoted). fix: use `PRIu32`/`%lu` and cas

33. **Long lines get truncated** — see: only the first ~120 characters arrive. cause: the RX buffer is 256 bytes by default. fix: raise `SERIAL_RX_BUFFER_SIZE` at build time, or drain faster. · [60879021](https://stackoverflow.com/q/6087902

34. **`SoftwareSerial` does not work** — see: nothing received. cause: the ESP32 has three hardware UARTs; bit-banging is unreliable and unsupported. fix: use `HardwareSerial(1)` or `(2)` with explicit pins. · [60094545](https://stackoverfl

35. **Waking from light sleep on serial data** — see: the UART does not wake the chip. cause: the wake source must be enabled before sleeping and the pin must be a valid wake pin. fix: `gpio_wakeup_enable` on the RX pin plus `esp_sleep_enab

36. **The board drops off and reconnects in a loop** — see: connects, disconnects, repeats every few seconds. cause: the application reconnects on top of the core's own reconnect logic, and both fight. fix: disable one of them; add exponent

37. **Wi-Fi and Bluetooth together are unstable** — see: enabling both makes one of them fail. cause: they share one radio and one antenna; coexistence needs time-slicing. fix: enable coexistence in `menuconfig`, expect lower throughput on 

38. **The antenna choice is not what you think** — see: terrible range on a module with an external antenna connector. cause: the RF switch must be told which antenna to use, and some modules need a solder jumper moved. fix: check the modul

39. **The access point always uses 192.168.4.1** — see: clients cannot reach the device on another subnet. cause: the default AP IP is fixed. fix: `WiFi.softAPConfig()` before `softAP()`. · [58976055](https://stackoverflow.com/q/58976055) 4

40. **mDNS works on one device and not another** — see: `device.local` resolves from a laptop but not a phone. cause: Android's resolver does not do mDNS by default. fix: do not rely on mDNS for discovery; publish the IP over the protocol y

41. **`connect()` returns errno 104** — see: `Connection reset by peer` after a few requests. cause: the server closed a keep-alive connection and the client reused it. fix: handle the reset by reconnecting, do not assume a socket stays ope

42. **HTTP GET takes seconds** — see: each request is slow even on a good network. cause: DNS resolution and a new TLS handshake per request. fix: reuse the connection, cache the resolved IP, or move to a persistent protocol. · [72242061](h

43. **Checking whether the internet is up** — see: `WiFi.status()` says connected but nothing loads. cause: it only reports the link, not reachability. fix: ping the gateway or fetch a small known URL with a short timeout, and cache the res

44. **Advertising stops once a client connects** — see: no longer discoverable after the first connection. cause: advertising is not restarted on disconnect, and some stacks stop it on connect. fix: restart advertising in the disconnect han

45. **BLE keeps the chip awake** — see: deep sleep never reaches microamps. cause: the BLE stack holds the radio and a timer. fix: deinit BLE before sleeping, and use `esp_bluedroid_disable`. · [61811536](https://stackoverflow.com/q/6181153

46. **A device will not pair** — see: the phone sees it, pairing fails. cause: missing bonding flags, or a security request the peripheral never answers. fix: set bonding and MITM explicitly and store the keys in NVS. · [67864472](https://s

47. **BLE and Wi-Fi share the antenna** — see: throughput collapses when both are active. cause: single radio, time-sliced. fix: expect it, or schedule BLE advertising only in Wi-Fi idle windows. · [65636417](https://stackoverflow.com/q/656

48. **"Sketch too big"** — see: the build fails at link time. cause: the default partition table gives the app only ~1.2 MB. fix: choose a partition scheme with a bigger app slot; do not shrink the code. · [75396355](https://stackoverflow.c

49. **NVS runs out of space** — see: `ESP_ERR_NVS_NO_FREE_PAGES`. cause: the NVS partition is 20 KB by default and entries have overhead. fix: enlarge the NVS partition in the table, or store large blobs in SPIFFS/LittleFS. · [55222899](htt

50. **Writing flash every second** — see: the device dies after months. cause: flash endurance is ~100k erase cycles per sector. fix: buffer in RAM and flush on a timer or on change; never log to NVS in a loop. · [73783849](https://stackove

51. **EEPROM, NVS or SPIFFS?** — see: no clear choice. cause: they solve different problems. fix: NVS for small key-values, SPIFFS/LittleFS for files, EEPROM emulation only for compatibility. · [70826170](https://stackoverflow.com/q/7082617

52. **EEPROM wear** — see: `EEPROM.commit()` in a loop. cause: each commit erases a sector. fix: commit on change with a debounce, or keep the value in NVS which does wear levelling. · [59817278](https://stackoverflow.com/q/59817278)

53. **Storing a string permanently** — see: `EEPROM.write` truncates. cause: it writes bytes, not strings. fix: write length + bytes, or use NVS with `nvs_set_str`. · [56139657](https://stackoverflow.com/q/56139657)

54. **The best way to store frequent data** — see: samples every second. cause: flash is the wrong medium for that rate. fix: ring buffer in RAM, flush batches to flash, or use RTC memory across deep sleep. · [63780850](https://stackoverflo

55. **Writing NVS during production flashing** — see: each unit needs a unique ID. cause: NVS is initialised at first boot, not at flash time. fix: generate an NVS partition image and flash it with the rest, or provision over serial on firs

56. **`LoadProhibited`** — see: `Guru Meditation Error: Core 1 panic'ed (LoadProhibited)`. cause: dereferencing NULL or a freed pointer. fix: check every pointer that came from `malloc`, `strchr`, or a failed lookup; `EXCVADDR` in the dump 

57. **Reading a backtrace** — see: a list of addresses. cause: the addresses are only meaningful with the matching ELF. fix: keep the ELF from that exact build and run `xtensa-esp32-elf-addr2line -pfiaC -e firmware.elf <addresses>`. · [6563

58. **`Interrupt wdt timeout on CPU1`** — see: panic with no obvious code path. cause: an ISR ran longer than the interrupt watchdog allows. fix: move work out of the ISR into a task via a queue; never call `delay`, `printf` or allocate ins

59. **`Task watchdog got triggered`** — see: reset after ~5 s of silence. cause: a task blocked or looped without yielding. fix: find the blocking call; add `vTaskDelay` only if the loop is genuinely idle. · [66278271](https://stackoverflow

60. **The watchdog fires during a long operation** — see: a legitimate long flash write or computation trips it. cause: the task never yields to the idle task. fix: subscribe the task to the TWDT and call `esp_task_wdt_reset()` inside the l

61. **Zeroing PSRAM eats all memory** — see: allocation fails after initialising a large buffer. cause: `calloc`-style zeroing touches every page and commits it. fix: allocate, then zero only what you use; check `ESP.getFreePsram()` before 

62. **MQTT publish crashes with a stack overflow** — see: reboot inside the publish call. cause: TLS plus MQTT needs more stack than the default task size. fix: raise the task stack to 8-16 KB, or publish from a dedicated task. · [67824787]

63. **`multiple definition of ...`** — see: link error after adding a second source file. cause: a variable defined in a header included twice. fix: `extern` in the header, define it in exactly one `.c`. · [57759461](https://stackoverflow.c

64. **A task must never return** — see: crash or silent stop when the function ends. cause: FreeRTOS tasks are not allowed to return; there is nowhere to return to. fix: end with `vTaskDelete(NULL)` or an infinite loop. · [63634917](https:/

65. **Stopping a task cleanly** — see: `vTaskDelete` from outside leaves resources held. cause: the task never gets to free its own memory. fix: signal it with a notification or a flag, let it clean up, then delete itself. · [62113104](http

66. **`std::mutex` versus `xSemaphore`** — see: a C++ mutex deadlocks or does not protect. cause: the C++ wrapper is not integrated with the RTOS scheduler the way the native primitives are. fix: use `xSemaphoreCreateMutex` / `std::mutex` c

67. **Rotary encoder misses steps** — see: counts are lost when the motor runs. cause: the ISR is too slow, or the pins have no hardware filter. fix: keep the ISR to a single counter increment, use the PCNT peripheral for real encoders. · [

68. **`esp_wifi.h: No such file or directory`** — see: the build cannot find an IDF header. cause: the component is not in `REQUIRES`, or the include path is not set. fix: add the component to the CMake `REQUIRES`/`PRIV_REQUIRES` list. · [7

69. **`idf.py` is not recognised** — see: the command is missing after install. cause: the export script was not sourced in this shell. fix: run `export.sh` (or `. ./export.sh`) in the same terminal, every time. · [63615242](https://stackov

70. **`collect2.exe: error: ld returned 1 exit status`** — see: the build fails only at link. cause: two libraries define the same symbol, or a library is missing from the build. fix: read the lines above the error — the linker always names

71. **Deep sleep still draws milliamps** — see: 8 mA instead of 10 µA. cause: the USB-serial chip is still powered, a GPIO is left driving a load, or the regulator has high quiescent current. fix: measure on the battery rail only, `gpio_hol

72. **A GPIO change does not wake the chip** — see: the interrupt works while awake, nothing after deep sleep. cause: only RTC-capable pins (0-15 on the classic ESP32) can wake from deep sleep. fix: use `esp_sleep_enable_ext1_wakeup()` with

73. **The ULP cannot touch a normal peripheral** — see: ULP code reads zero or hangs. cause: the ULP has its own tiny instruction set and can only reach RTC memory and a few registers. fix: keep the ULP to sampling and simple arithmetic, an

74. **ULP wake period drifts by minutes** — see: a "60 s" wake drifts. cause: the RTC slow clock is an RC oscillator with several percent tolerance. fix: calibrate it against the XTAL, or accept the drift and resync time when the radio is o

75. **Variables lose their value after sleep** — see: a counter resets every wake. cause: deep sleep powers down normal RAM; only RTC memory survives. fix: `RTC_DATA_ATTR` on the few values that must persist, and keep them small. · [5654870

76. **`ext0` wake never fires** — see: no wake on a valid pin. cause: `ext0` works on a single RTC pin and a specific level, and the pin must be configured as an RTC IO first. fix: `rtc_gpio_init` + `rtc_gpio_pullup`, then `esp_sleep_enable

77. **Touch does not wake the board** — see: touch works while awake, not as a wake source. cause: touch wake has to be armed separately and only works on touch-capable pins. fix: `esp_sleep_enable_touchpad_wakeup()` plus `touch_pad_set_fsm

78. **Light sleep and deep sleep behave differently** — see: a sketch that survives light sleep dies after deep sleep. cause: light sleep keeps RAM and the peripherals; deep sleep keeps almost nothing. fix: treat them as different features,

79. **Brownout right after waking** — see: a reset exactly when the radio starts. cause: the bulk capacitor is discharged after a long sleep and the RF inrush pulls the rail down. fix: larger capacitor, and bring the radio up after a short 

80. **OTA uploads but nothing changes** — see: the write succeeds, the old firmware still runs. cause: the OTA API writes to the inactive partition; without `set_boot_partition` and a reboot the bootloader still starts the old one. fix: `es

81. **A bad OTA bricks the device** — see: the new image boots once and dies. cause: no rollback configured, so a broken image stays active. fix: enable `CONFIG_BOOTLOADER_APP_ROLLBACK_ENABLE` and call `esp_ota_mark_app_valid_cancel_rollbac

82. **Boot loop after editing the partition table** — see: `invalid header` or an immediate reset. cause: the table moved the app offset but the bootloader and `otadata` were not reflashed with it. fix: flash bootloader, partition table and

83. **SPIFFS shows far less than 4 MB** — see: `SPIFFS.totalBytes()` reports a few hundred KB. cause: the filesystem size is whatever the partition table gives it, not the flash size. fix: enlarge the `spiffs`/`littlefs` partition in the ta

84. **The filesystem corrupts after a power cut** — see: mount fails after an unexpected reset. cause: SPIFFS is not power-loss resilient. fix: move to LittleFS, which keeps a journal, or call `format` on mount failure and rebuild. · [63568

85. **The device crashes during OTA writes** — see: a panic inside a flash write. cause: writing flash disables the cache, so any code running from flash — including ISRs — faults. fix: put interrupt handlers in IRAM (`IRAM_ATTR`) and stop 

86. **HTTPS OTA fails the handshake** — see: `HTTPClient` returns -1 over `https`. cause: no certificate, or a clock so wrong that the cert looks not-yet-valid. fix: set the time first and pin a CA; `setInsecure()` works but removes the onl

87. **OTA over a slow link times out** — see: the download stalls near the end. cause: the default HTTP timeout is shorter than the transfer. fix: raise the timeout, stream into `esp_ota_write` in chunks, and never buffer the whole image. ·

88. **TLS handshake fails with a valid certificate** — see: `setCACert` still fails. cause: the system clock is at 1970, so the certificate is not valid yet. fix: sync time over SNTP before the first HTTPS call. · [68356924](https://stackov

89. **The certificate expires and the product dies** — see: every device stops talking on the same day. cause: a leaf certificate was pinned instead of a CA, or the root changed. fix: pin a CA bundle, ship an update path for the bundle, and

90. **`time(nullptr)` returns a 1970 timestamp** — see: time is never valid. cause: `configTime` is asynchronous and returns immediately. fix: wait until the value is above a sane floor (e.g. 1.6e9) before trusting it, with a timeout. · [68

91. **WebSocket closes with code 1006** — see: the socket dies with no close frame. cause: nothing was sent for longer than the proxy or server idle timeout. fix: send a ping every 20-30 s and handle a reconnect instead of assuming the link

92. **UDP packets vanish on the soft-AP** — see: heavy loss when the ESP32 is the access point. cause: modem sleep parks the radio between beacons. fix: `WiFi.setSleep(false)` while streaming, and expect lower throughput than the datasheet 

93. **TWAI receives nothing** — see: no frames at all. cause: no transceiver on the bus, TX/RX swapped, no 120 Ω termination, or a bitrate mismatch. fix: check the transceiver first, then the termination, then the timing. · [68527367](https

94. **The LED strip flickers when CAN is active** — see: WS2812 glitches under bus load. cause: the strip refresh is timing-critical and the CAN ISR delays it. fix: drive the strip from RMT, keep the CAN ISR short, and do not refresh from i

95. **`rmtWrite` blocks the whole loop** — see: the main loop stutters during a strip update. cause: the blocking variant waits for the transmission. fix: `rmt_write_items(..., false)` and wait on the done semaphore, or double-buffer two tr

96. **DAC output does nothing** — see: `dacWrite` has no effect. cause: the DAC exists only on GPIO25 and GPIO26, and both are also ADC2 channels. fix: use those two pins only, and do not expect them to work as ADC while Wi-Fi is on. · [792

97. **Touch readings drift over hours** — see: thresholds that worked stop working. cause: the touch baseline is not recalibrated and the environment changed. fix: run `touch_pad_set_meas_time` sensibly, recalibrate the baseline periodicall

98. **A PCNT pin cannot be used for wake** — see: `ext1` wake does nothing on a pin that works elsewhere. cause: the pulse counter has taken the pin. fix: `pcnt_unit_disable` and release the GPIO before arming the wake source. · [54316980](

99. **The servo jitters** — see: random twitches even with a clean PWM signal. cause: the servo's current spikes brown out the shared 5 V rail. fix: separate supply, common ground, and a capacitor at the servo. · [60171641](https://stackove

100. **A queue send from an ISR crashes** — see: `LoadProhibited` inside the interrupt. cause: `xQueueSend` was used instead of the ISR variant. fix: `xQueueSendFromISR` with the woken flag, and create the queue before attaching the interru

101. **The task stack overflows when parsing JSON** — see: a reset inside `deserializeJson`. cause: JSON parsing needs several KB of stack, and the default task stack is smaller than that. fix: parse on a dedicated task with 8-16 KB, or use

102. **PSRAM allocation fails** — see: `ps_malloc` returns NULL on a module with PSRAM. cause: PSRAM is not enabled in the build, or the allocator was asked for internal RAM. fix: enable SPIRAM in the board config and use `heap_caps_malloc(

103. **The watchdog still fires after disabling it** — see: WDT resets continue. cause: there are three watchdogs — task, interrupt and RTC — and they are configured separately. fix: identify which one from the reset reason, then handle tha

104. **`LoadProhibited` only in a task** — see: the same code works in `loop()`. cause: the task stack is too small, or it holds a pointer to another task's stack. fix: size the stack in bytes, pass data by value or by heap, never by a poin

105. **Heap shrinks over days** — see: free heap falls and never recovers. cause: fragmentation from `String` concatenation and many small allocations of different sizes. fix: fixed-size buffers, reserve() up front, and check `heap_caps_get

106. **The captive portal never opens on Android** — see: the phone connects and does nothing. cause: Android probes `connectivitycheck.gstatic.com/generate_204` and expects a 204 or a redirect. fix: answer that path explicitly and resolve 

107. **Clients on the soft-AP have no internet** — see: they connect but cannot reach the web. cause: the ESP32 is an access point, not a router; there is no NAT in the default stack. fix: add a NAT layer (`esp_netif_napt_enable` on newer I

108. **The AP IP is printed before it exists** — see: `WiFi.softAPIP()` returns 0.0.0.0. cause: the AP starts asynchronously. fix: wait for the `AP_START` / `got_ip` event before printing or advertising the address. · [73999244](https://sta

109. **Two BLE profiles fight over one callback** — see: notifications stop when a second service is added. cause: a single GATTS handler cannot own two profiles' events. fix: register one handler per profile and dispatch on `gatts_if`. · [

110. **A phone will not reconnect after reflashing** — see: the device is invisible to a phone that paired before. cause: the phone still holds the old bond while the ESP32 lost its keys. fix: delete the bond on the phone, or call `esp_ble_

#!/usr/bin/env python3

#!/usr/bin/env python3

#!/usr/bin/env python3