Session Handoff [hw-cli]: hw CLI separated; upstream dev defect PRs #27-#32, #35 #36

Next steps, in priority order

1. Close rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35's four open documentation points, which the PR body lists: the stale ZephyrRTCDiscover.h header contract; the "restored" log on the save-path re-probe; and two commit-message inaccuracies (the residual EEPROM risk is wider than "0Eh odd"; Date and Month reset to 01h, not 0). Done means: fixed on work/rak4631-rak12002-rtc through local-strict-review, pushed, the PR body's open list emptied, and Copilot's headline on 9e9452f answered with the fixing SHA.
2. Read CodeRabbit's results for Add the hw hardware-report CLI command (rebased, hw CLI only) #16, Add hw hardware-report CLI command (iteration branch) #1 and rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35. They were requested one per hour by a bounded background job (the OSS plan allows one review per hour per repository), which may not survive the restart: see External blockers. Done means: each PR shows No actionable comments for its head in the edited CodeRabbit summary comment, or each finding is closed per pr-review-conduct.
3. Upstream, one PR at a time, only on the maintainer's explicit go-ahead naming it. For each approved fork PR (companion: mark a CLI reply cut to the app frame (iteration branch) #27 to companion: never treat an over-the-air CLI command as local (iteration branch) #32, rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35), cut a clean branch from the synced mirror dev (upstream/dev was f9d01cf; fetch and compare first), squash the dirty branch's net change onto it with the "Proposed upstream PR" text from the fork PR body, and open it against liquidraver/ZephCore dev. The fork PRs stay open and are never merged (hub upstream-contribution-workflow).
4. hw CLI on hardware (Add hw hardware-report CLI command (iteration branch) #1, Add the hw hardware-report CLI command (rebased, hw CLI only) #16) is still not re-run since the separation. The RAK4631 is attached (see State). Done means each hw subcommand answers as docs/Repeater_CLI_commands.md says.

External blockers

• Maintainer go-ahead for every upstream PR (step 3), and for any write to meshcore-dev/MeshCore.
• The CodeRabbit queue ran as a background shell job in the session that wrote this. If it did not survive the Claude restart, request @coderabbitai review by hand with pr_review.py comment, no more than one an hour across the repository.

Internal dependencies

• Step 3 for rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35 waits on step 1, so the squashed upstream text carries no known-inaccurate statement.
• rtc: do not adopt a chip whose status read failed (iteration branch) #30 (fork issue RTC probe adopts a chip whose status-register read failed, then writes time into it #23) and rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35 both edit rtc_probe() in adapters/clock/ZephyrRTCDiscover.c. Whichever goes upstream second needs rebasing onto the first. Add the hw hardware-report CLI command (rebased, hw CLI only) #16's hw CLI touches the same file, so it also conflicts with both once they land.
• server: leave room for a local "xx|" prefix in the CLI reply buffer (iteration branch) #28 (local xx| prefix room) and Add the hw hardware-report CLI command (rebased, hw CLI only) #16's hw both reserve prefix room. If server: leave room for a local "xx|" prefix in the CLI reply buffer (iteration branch) #28 lands first, hw's own local prefix deduction becomes redundant but stays harmless.

State (re-derive at resume: git worktree list, gh pr list)

• Fork dev = upstream/dev = f9d01cf when this was written.
• hw CLI: Add hw hardware-report CLI command (iteration branch) #1 work/hw-cli (the primary west workspace checkout) at ad70d15, and Add the hw hardware-report CLI command (rebased, hw CLI only) #16 hw-cli-only at cc54670. The trees are byte-identical and hold the hw CLI alone.
• Defect iteration PRs, all open into the fork's dev, one worktree each named after the branch:

• Review state: companion: mark a CLI reply cut to the app frame (iteration branch) #27-companion: never treat an over-the-air CLI command as local (iteration branch) #32 have clean Copilot and CodeRabbit rounds on their heads. rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35 has a Copilot round on its head with no new findings plus the four open documentation points; CodeRabbit is pending. Add hw hardware-report CLI command (iteration branch) #1 and Add the hw hardware-report CLI command (rebased, hw CLI only) #16 have their Copilot findings fixed, are attested, and CodeRabbit is pending. Copilot coverage reads "unstated" on several PRs, which is acceptable on this fork.
• Hub tooling ran from a private detached worktree of ptr727/ProjectTemplate at origin/main 6b1b147.
• Hardware:
  • The RAK4631 runs rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35's debug repeater build: SoCal, 1 dBm, handle lpxz, repeat and adverts off. Its RAK12002's RV3028 EEPROM holds 37h BSM=01, TCE=1, TCR=00 and 35h CLKOE=0, committed by that build.
  • The ThinkNode M7 is back on Elecrow ThinkNode M7: board, CH390 ethernet driver, wired TCP companion (iteration branch) #17 a7e2698, after the Zephyr session's T15-T17 rounds (evidence under the M7 retest worktree's ladder-evidence/T15/).
  • The Heltec V4-R8 is untouched.
• A separate MeshCore session was briefed on RV3028: backup switchover and trickle charging written to RAM only, so they revert at the next midnight; 0xB4 also overwrites a calibration bit meshcore-dev-MeshCore#10 and owns that work.

The parked decision queue

0 issues carry a decision label (this fork has none). Open maintainer questions, ranked:

1. Upstream go-ahead per fork PR (companion: mark a CLI reply cut to the app frame (iteration branch) #27-companion: never treat an over-the-air CLI command as local (iteration branch) #32, rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35).
2. First CLI command after opening the USB serial port returns "Unknown command" #6 does not reproduce on dev. Close it?
3. TX power accepts -9..0 dBm everywhere, though 1 dBm is the floor #19, TX power -9..0 dBm: the maintainer chose "leave open, no PR".

What the last round did

• Separated the hw CLI from upstream dev defects. Add hw hardware-report CLI command (iteration branch) #1 and Add the hw hardware-report CLI command (rebased, hw CLI only) #16 now carry only hw. The fixes for Companion CLI replies over 175 bytes are truncated silently #3, RAK4631 never uses a fitted RAK12002 RTC #4 and RTC probe adopts a chip whose status-register read failed, then writes time into it #23 moved to their own branches through append-only commits (bf3e96a, 1dc07d6). Several local passes and a Copilot round fixed hw documentation that did not match the code.
• Opened the defect iteration PRs companion: mark a CLI reply cut to the app frame (iteration branch) #27-companion: never treat an over-the-air CLI command as local (iteration branch) #32 and rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35, each with "Proposed upstream PR" text in Problem / Solution / Testing shape.
• Found two defects during review, both without fork issues by the maintainer's choice:
  • cli: cap the replies that echo a remote command (iteration branch) #31: remote admin replies that echo the command overran the 161-byte buffer.
  • companion: never treat an over-the-air CLI command as local (iteration branch) #32: on a companion, an over-the-air CLI command with timestamp 0 was treated as local, bypassing the allow-list.
• Verified on the RAK4631:
  • companion: mark a CLI reply cut to the app frame (iteration branch) #27: a cut reply ends in ... on a UTF-8 boundary.
  • rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35: a 30 s power cut is survived once the RV3028 backup switchover is committed to EEPROM.
• Closed ESP32-S3 builds fail to link on dev 5b77947: l32r literal after use in the Xtensa double-exception vector #24: a worktree-layout artifact, not a defect.
• Closed tcp_work stack overflow under TCP load: Ethernet fixed in #17, WiFi companion unmeasured #25: the ESP32 WiFi driver already gives tcp_work 2048 bytes.
• Filed RV3028 loses the time on a power cut: ZephCore never enables its backup switchover #33 (the RV3028 switchover, fixed in rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35) and RTC probe assumes 24-hour mode: an RV3028 left in 12-hour mode is misread or never adopted #34 (the RTC probe assumes 24-hour mode).
• Filed RV3028: backup switchover and trickle charging written to RAM only, so they revert at the next midnight; 0xB4 also overwrites a calibration bit meshcore-dev-MeshCore#10: MeshCore writes RV3028 37h to RAM only, so it reverts at midnight, and 0xB4 overwrites the calibration bit.
• Served the Zephyr session's T15, T16 and T17 M7 rounds (CH390 driver), reporting to it directly.

What not to repeat

• Never trust an ESP32 link result from a worktree whose zephyr/ is a symlink. The archive gets a mangled name, so the linker script's IRAM rules miss it and the link fails. That cost ESP32-S3 builds fail to link on dev 5b77947: l32r literal after use in the Xtensa double-exception vector #24's whole investigation. Export ZEPHYR_BASE to the resolved path.
• Never change a board's role without baking the frequency and name defaults into the build. A role switch formats both roles' prefs. The RAK4631 came up as a repeater on 869.618 MHz (at 1 dBm) this round.
• Never take a review's datasheet claims on recall. Twice a reviewer's or my own recollection was wrong: the 1 kΩ trickle resistor (it is 3 kΩ), and "level switching is the vendor's advice for a supercap" (§7.3 says direct switching). Read the manual text; the maintainer's copies are in the home directory.
• Never run local_review.py status with build output sitting untracked in the worktree. Build .diag files shifted the digest and blocked an attest. Keep build summaries inside the build directory.

New learnings

• The symlinked-zephyr ESP32 layout, the role-switch prefs wipe, and the RAK4631's intermittent 1200-baud touch are recorded in this host's project memory.
• RV3028 configuration must be committed to EEPROM, by the manual's §4.6.3/§4.6.9 procedure. It is recorded in rak4631: use a fitted RAK12002 RTC, and keep its backup switchover enabled (iteration branch) #35, RV3028 loses the time on a power cut: ZephCore never enables its backup switchover #33 and MeshCore Tracking: hw hardware-report command — state and handoff #10.