The style pass rewrote an em dash as a colon inside an unquoted name:
scalar, which YAML reads as a nested mapping. Comma instead; file
re-validated with a YAML parse.
Replace every em dash with real punctuation (rewrites, not hyphen swaps)
in docs, code comments, scripts, configs, and the port records; convert
en dashes, curly quotes, ellipsis glyphs, arrows, and section signs to
ASCII; drop machine-writing tell phrases from living docs. ADR titles
now use a colon. The M2 bring-up DTS model string carried an em dash
into the patch series and its record echoes; fixed at both, and the full
series re-verified to apply cleanly onto pristine 6.18.46. One comment
in freshness.h deliberately names the em dash glyph the UI renders as
the unknown mark; that is data, kept as prose naming it.
Verified: cargo tests (sim, config-lint, rs485-bridge), shellcheck,
both driver MC/DC harnesses, patches-apply.
Drop the Why / What Works sections and the product-origin story; the
text before Quick Start is now one comparison table of the upgrades over
the vendor SDK. Coverage methodology stays in the internal docs as the
reliability standard rather than the project's identity.
Second conciseness pass per review feedback: bullets and tables over
prose (vendor-SDK failings, repo layout, sim models, kernel-port
rationale, QEMU scenarios), duplicated content cut (provenance pointer,
scenario prose the qemu README already carries). Also fixes two stale
claims in architecture.md: the flared seam unification and real-image
boot both landed (flare-edge #110, PR #8) but were still marked pending.
The last product-first artifact after ADR-0008 was the repo name itself.
Current and forward-looking references updated; period-bound history in
the ADRs keeps the old name, with the rename recorded in ADR-0008.
Crate names, WARDEN_* env vars, and binaries are unchanged.
The repo's documentation framed it as a support repo for one product
(WardenOS). Since going public the real audience is anyone with a Luckfox
Pico 86 Panel: a maintained 6.18 kernel, an off-device development loop,
and a device simulator that exist nowhere else for this board. Reframe the
README and top-level docs board-first, with WardenOS documented as the
downstream consumer it is (ADR-0008).
Also an editorial pass over the whole doc set:
- every H1/H2 is now a short title, not a sentence (ADRs, qemu/, patches/,
drivers/, architecture, NPU feasibility, config-lint, payload); workflow
flowchart titles fixed at the source in tools/flowgen.py and regenerated
with fresh bench numbers
- README Quick Start commands verified against the scripts; requirements
corrected (curl, bare python, gcc >= 14) and the MC/DC gate added as a
step (run green locally on gcc 14.2)
- dropped the 'needs python (not python3)' vendor dig: build-kernel.sh
inherited the same requirement (filed #10 to remove it)
- glossed MC/DC and HPMCU on first use; marked the tests/uboot-ab
reference as flare-edge; deduplicated the three-simulator list into the
root README table
mkimage gains --rootfs-image/--oem-image (raw ext4 matched pair placed
into slot A, size-checked fail-closed; slot B keeps the skeleton as a
known-good fallback), and stage-1 rc now hands the live devtmpfs to the
new root across switch_root — busybox switch_root moves nothing, our
skeleton init remounted defensively, but a REAL image's getty opens
/dev/console immediately and looped on ENOENT without it.
Verified with the 2026-08-24 flare-edge build (rootfs 192M + oem 26M):
the image's own rcS chain runs on the 6.18 virt kernel, real daemons
start (warden-modbus confirmed running via root shell; the shipped
0.2.4 flared reproduces flare-edge#106 exactly, and S99hciinit reports
the issue-#4 btlpm symptom — the VM doubles as a faithful reproducer of
known field bugs), and a getty answers on the console.
qemu/tests/real-image-boot.sh asserts switch_root + daemon starts +
login prompt, credential-free. REAL-IMAGE-BOOT-PASS.
Regression sweep after the shared-path changes: OTA-APPLY-PASS,
PORTAL-SCENARIO-PASS, UI-SHOT-PASS, boot smoke OK.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
Closes the loop every prior harness stopped short of, entirely
off-hardware: the real flared (WARDEN_HARD_RESET-gated build) pulls a
real signed tier-1 .wfw whose payload is a BOOTABLE rootfs stamped
0.0.2, verifies it, writes /dev/block/by-name/rootfs_b inside disk.img,
and flips the AvbABData (slot B: priority 15, 3 tries, unsuccessful —
the exact pre-first-boot arming state, round-tripped through a portal
check-in). The harness then boots slot _b and asserts the applied
version + marker are what runs. OTA-APPLY-PASS verified end to end.
- mkimage: the misc partition now carries REAL provisioned AvbABData
(bytes mirror flare-edge's provisioning defaults) — slotctl fail-closes
on bad AB magic before writing, which a zeroed misc tripped.
- run.sh --allow-apply / cmdline warden.fwapply: per-boot opt-in that
makes stage-2 init export WARDEN_FW_ALLOW_APPLY=1; never the default.
- stage-2 init also exports WARDEN_HARD_RESET=0 (the CRU poke is fatal on
virt, same class as the HPMCU probe); the harness performs the reboot.
- ADR-0006 boundary documented in the scenario and README: BCB slot
CHOICE and the physical reset remain emulated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
The hermetic build only ever made zImage+dtb — no 6.18 .ko set existed,
so the panel's only aic8800_btlpm.ko was the stale 5.10 build, which
fails the struct-module ABI check at load. New WARDEN_MODULES_COLLECT
env runs the full `make modules` (required: zImage alone emits no
Module.symvers, so per-directory M= builds cannot link) and collects the
listed dirs' modules fail-closed; CI's kernel-build collects
drivers/net/wireless/aic8800 and ships the .ko files in the artifact.
Verified: all three aic8800 modules (bsp 326K, fdrv 1.5M, btlpm 36K)
compile clean against 6.18, and bsp+btlpm INSMOD WITH RC=0 on the 6.18
kernel in the qemu/ device sim — no linkonce/this_module ABI error;
btlpm's rfkill init runs and only hardware power-up fails (no AIC silicon
on virt, expected). On-panel BT bring-up remains bench-gated (c8a3
currently dark).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
The qemu/ device sim answered the first question off-board: the same
kernel family under -M virt gives musl vDSO rate ratio 0.99963 — the
generic 6.18 armv7 vDSO is correct, so the board symptom is RV1106
register state (CNTFRQ/CNTVOFF, firmware-owned, secure-world boot chain).
- qemu/tests/clockprobe: interval-based musl probe separating RATE error
(CNTFRQ) from boot OFFSET (CNTVOFF) — the original single absolute
sample cannot distinguish them.
- qemu/tests/clock-sanity.sh: VM regression guard asserting the vDSO rate
within 1% (PASSES: 1.00026); cross-builds the probe and stages it as
payload itself.
- kernel/rv1106-enablement/timer/PLAN.md: the DT fix
(arm,cpu-registers-not-fw-configured + measured clock-frequency on the
board dts) gated on the two bench measurements; c8a3 is currently
physically dark, needs hands at the bench.
- clockprobe joins the CI test loop.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
ubuntu-latest ships shellcheck 0.9.0, which warns that ash scripts are
checked as dash; 0.10 (local) does not. shell=dash is the accurate dialect
for busybox ash. Verified clean under BOTH versions; boot smoke re-run.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
- ci-cd.md phrasing made time-invariant (true before and after the
visibility flip; the prior wording asserted a state not yet real).
- ui-shot: liveness check before the tap QMP call too — the commit-message
claim of "before every QMP call" is now actually true.
- busybox fetcher checks its pin before downloading, mirroring the
kernel-tarball fetcher's ordering.
- Fragment disable-arm: absence of a deps-gated symbol is a valid disable
outcome; only "still set" is fatal (the iteration-2 symmetrization
overcorrected). Negative-tested both directions under set -e.
Iteration 3 produced four findings, none blocking, all applied — the
recursive review is converged.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
- build-kernel.sh fragment assertion: survives a missing trailing newline
(the read footgun, reproduced) and treats an absent symbol on a disable
line as FATAL, symmetric with the enable arm.
- fetch-kernel-tarball.sh checks the pin before downloading; both fetchers
add --retry-connrefused.
- mkimage rejects '.'/'..' state keys.
- ui-shot: VM liveness checked before every QMP call, console.log preserved
as evidence on every failure path, repaint deadline widened to 90s with
the contended-runner rationale documented.
- rs485-bridge: overflow discards back off one gap and rate-limit their log
line, mirroring the accept-loop fix; clippy nit fixed.
- .gitignore ignores *.elf/*.map so the untracked artifacts cannot silently
return; CI shellcheck glob now covers build/ and the rootfs boot scripts
(directives added for the deliberate in-guest source paths).
- Docs: NPU parity row matches its sibling verification docs; line-pinned
audit cross-references unpinned; CROSS_COMPILE documented in the build
header; payload README lists warden-ui; ci-cd tense settled.
Verified: guards negative-tested (bad state keys, no-newline fragment);
boot smoke, portal scenario, ui-shot all PASS; 53 tests green; shellcheck
clean across the widened glob; clippy zero.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
CI/pipeline:
- KERNEL_TARBALL passed as a YAML env literal '~' was never tilde-expanded
and would have failed every hosted kernel-build dispatch; the path is now
exported from the shell. Verified reproducible before the fix.
- Every job gets timeout-minutes; boot smoke uses timeout -k so a wedged
qemu is SIGKILLed instead of holding the job.
- Tarball fetch + fail-closed sha256 verification deduplicated into
build/fetch-kernel-tarball.sh (with curl retries), used by build-kernel.sh
and both CI jobs. busybox fetch gains retries too.
- ccache layer for kernel-build (cache keyed on defconfig+patches) recovers
the incremental-compile speed the ephemeral-runner move cost.
- build-kernel.sh now asserts every fragment option survived olddefconfig —
merge_config -m pastes text and Kconfig silently drops unmet symbols.
rs485-bridge:
- pending-buffer cap (2x max RTU ADU) instead of unbounded growth;
explicit accept-loop error handling with backoff instead of .flatten();
per-arm inline bounds instead of the string-keyed lookup whose default
would have mis-bounded a future get-input; control-socket cleanup errors
surfaced; flag-shaped values rejected in arg parsing; doc example uses a
private mktemp dir. Test timing margins widened for contended runners
(gap 25->120ms, 60x margin on the split-frame test).
VM harness:
- stage-1/stage-2 boot scripts share one validated slot parser and one
by-name populator (qemu/rootfs/etc/warden-lib.sh) — the duplicated
parser had already diverged on validation; userdata/oem mount failures
now fail fast with a greppable sentinel; udhcpc fallback keys off the
interface actually having an address; switch_root applet guarded.
- boot-smoke delegates the qemu invocation to run.sh (machine shape lives
in ONE place); run.sh port 0 disables a hostfwd.
- mkimage: unknown partition names fail at build time; DISK_END is a max,
not last-entry; --state keys validated as filenames.
- portal-scenario: mock readiness is asserted (no silent fall-through),
hostfwd port collisions retried, mount-failure sentinel fails fast.
- ui-shot: fixed sleeps replaced with bounded screendump polling; the
repaint assertion is real and documented as such. qmp.py loses its
module-global and gains argv validation.
Docs/scrub: bench-host paths and the site AP name removed from six more
port docs and two evidence tables; path-bearing build artifacts (.elf,
.map) untracked (the 154-byte firmware .bin is path-free and stays);
ADR-0003 marked visibility-superseded by ADR-0007; stale section
cross-reference fixed; flare-edge noted as private for outside readers;
stale root-level review report removed per the new workspace rule.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
Replaces the dual MIT/Apache-2.0 licensing (and its GPL carve-out) with a
single GPL-2.0-only license for the whole repository: coherent with the
kernel patch series it exists to carry, the embedded-Linux norm, and the
stack's openness principle — anyone shipping products on this SDK publishes
their changes. GPL-2.0 (not 3) deliberately: the Installation-Information
clause of GPLv3 sits badly with WardenOS's signed A/B firmware chain.
LICENSE is now the canonical GPLv2 text; crate manifests updated;
inbound = outbound noted in the README.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
Decision/directive lines keep their dates and content, drop the name;
hands-on confirmations become "confirmed by hand at the panel"; a dangling
CLAUDE.md reference becomes "repo policy". Upstream vendor author
attributions in the GPL kernel sources (MODULE_AUTHOR, copyright headers,
LKML citations) are third-party legal attributions and stay.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
Preparing the repo to go public (ADR-0007):
- CI: kernel-build moves from the self-hosted runner to ubuntu-latest
(installs its own cross toolchain + qemu, caches the pristine tarball).
On a public repo a registered self-hosted runner is reachable from
approved fork-PR workflows — i.e. arbitrary code on private
infrastructure — and the build never actually needed the SDK host.
ADR-0004 marked superseded-in-part; docs/ci-cd.md rewritten (site
specifics now live only in the private deployment log).
- Licensing: LICENSE gains the GPL-2.0 carve-out for patches/ and the
kernel source excerpts (Linux derivatives; per-driver provenance was
already tracked in PROVENANCE.md); patches/README.md states it too.
- Scrubbed from the tip: bench-unit dev credentials and its gadget IP
(m2-boot notes), the site AP SSID+BSSID and a neighboring AP's BSSID
and the device WLAN MAC (wifi bring-up evidence — BSSIDs are
geolocatable), the runner mesh IP. NOTE: these remain in git history;
decision on a pre-publication history rewrite is separate.
- Emoji cleanup across 21 tracked files (kernel port docs, review report,
enforce-mcdc.sh) per repo text conventions: status marks became
[x]/[wip]/[ ]/OK plain text.
- "[maintainer]-gated" process phrasing normalized to "maintainer-gated"
(attributions in dated evidence docs kept).
Verified: zero emojis tracked; scrub grep clean; patches carry no internal
references; ci.yml parses; shellcheck unchanged.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
Installed 2026-08-30 on bfe-mpc-0640 (QEMU 10.0.11; cpio/curl already
present); recorded in flare-deployment deploy-log. End-to-end CI check of
the smoke step waits on the account-wide Actions billing stall.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
Phase 4+5 of the device sim:
- Display + touch verified end-to-end: virtio-gpu at 720x720 (fbdev
emulation) renders the real WardenOS dashboard from the static LVGL
fbdev+evdev UI build (flare-edge qemu-vm-support tools/build-ui-vm.sh);
QMP input-send-event taps the Metrics tab and qemu/tests/ui-shot.sh
asserts the repaint from screendumps. Two load-bearing QEMU flags found
and documented: -global virtio-mmio.force-legacy=false (gpu/input are
VERSION_1-only) and the 200ms press hold (an instantaneous press+release
lands inside one LVGL indev poll and never clicks).
- qemu/tests/qmp.py: minimal QMP client (screendump, tap, quit).
- stage-2 init starts warden-ui when present and fb0 exists.
- docs/decisions/0006-qemu-device-sim.md: virt-not-custom-board, the
enters-at-kernel boundary, fragment policy, naming, consequences.
- docs/architecture.md: new section 7 (device emulation), order-of-work
item 7; modbus cross-reference to the bridge.
- qemu/README.md: emulated-vs-not table, scenarios, gotchas, host/runner
requirements. docs/ci-cd.md: runner needs one-time qemu-system-arm
install (fail-closed smoke until then, [maintainer]-gated). Repo README updated.
Final sweep on this commit: shellcheck clean, bridge 7/7 tests, boot smoke
PASS, portal scenario PASS (check-in + fw pull + signed .wfw download),
ui-shot PASS (touch navigates to Metrics) — all under the final flags.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
Phase 3 of the device sim, all verified on QEMU 10.0.11:
- qemu/rs485-bridge/: std-only crate bridging a QEMU serial chardev (unix
socket) to warden_sim::ModbusSlave — gap-based RTU framing (CRC failures
degrade to real-slave silence), line-protocol control socket for register
seeding and fault injection (drop/exception/clear), bounds-checked so a
scenario typo answers err instead of panicking the bus. 7 unit tests,
bench in the sim_bench pattern. Verified end-to-end: guest master frame
on /dev/ttyS4 (pci-serial) answered from the sim slave, CRC-correct.
- virt machine gains highmem=off: the 32-bit non-LPAE kernel cannot reach
virt's default 40-bit PCIe ECAM (pci-host-generic EOVERFLOW); with it the
full PCI set probes (16550A ttyS0, i6300esb).
- Watchdog scenario verified: guest arms /dev/watchdog, no petting, i6300esb
resets the VM ~30s later (first environment where this arm is testable).
- qemu/tests/portal-scenario.sh: the real static-musl warden-flared inside
the VM against flare-edge's mock portal on the host — authenticated
check-in, firmware desired-state pull, and download of a real signed
tier-1 .wfw offer, asserted from the portal log. Found and filed
flare-edge#106 (fatal SIGBUS in the HPMCU boot-loaded probe on
non-RV1106 memory maps); runs against a flared built from the
qemu-vm-support fix branch.
- stage-2 init: WARDEN_FLARE_INSECURE=1 + WARDEN_HPMCU=0 (documented VM
deviations), firmware-version stamp, newline-terminated state seeds.
- CI: rs485-bridge joins the test loop and bench job; new qemu-tools job
(shellcheck + initramfs + disk image on hosted runners); kernel-build
gains a fail-closed qemu boot-smoke step. All qemu scripts shellcheck-clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
The third simulator (deliberately not named "sim"): a QEMU -M virt VM that
boots the real 6.18.46 kernel and enters at -kernel zImage — everything below
(BootROM/idblock/U-Boot/real BCB A/B selection) is closed blobs + mask ROM
and is explicitly out of scope.
- qemu/mkinitramfs.sh: pinned static busybox (sha256 fail-closed) + rootfs/
- qemu/mkimage.sh: unprivileged sparse disk image with the device's canonical
12-partition blkdevparts A/B layout (vda == mmcblk0 mapping)
- qemu/rootfs/: stage-1 init (by-name symlinks from PARTNAME uevents,
whole-token warden.slot= parse, switch_root) + stage-2 init (userdata/oem
mounts, slirp networking, payload daemon start)
- qemu/run.sh: runner with --slot/--rtc/--watchdog/--rs485/--qmp/--display
- qemu/configs/virt.fragment + WARDEN_KCONFIG_FRAGMENT hook in
build/build-kernel.sh (canonical RV1106 build untouched when unset):
adds PCI, pci-serial, i6300esb watchdog, WireGuard, virtio-gpu/input
- qemu/tests/boot-smoke.sh: sentinel-asserting boot test
Verified on QEMU 10.0.11: canonical zImage boots -M virt unmodified (the
feared DEBUG_UNCOMPRESS decompressor hang does not exist in 6.18); full
stack boots both slots; 12 by-name symlinks; userdata persists across
reboot; -rtc base=2021-01-01 reproduces the no-RTC wrong-clock class.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018HUayid7W5w7jBdb9Rrj1K
Mirror flare-edge's artifact GC so the shared Actions storage quota can't fill
from either repo. A new prune-artifacts job (ubuntu, gh + built-in token,
actions:write) runs before the kernel-build upload and deletes older
kernel-rv1106 artifacts beyond the newest 3; kernel-build needs it. retention-days
14 -> 5 as the backstop. Best-effort so it never blocks a build. Completes the
account-wide auto-clear (an over-quota once blocked ALL runs, not just uploads).
The .c/.dts/.frag files here are a point-in-time port-provenance snapshot (with
standalone test programs + instructional fragments the narrative docs reference),
not a maintained second copy of the kernel delta. Add a prominent canonical-source
note so the snapshots (which can lag the final series, e.g. the clk CLK_IGNORE_UNUSED
fix) are never mistaken for current code. patches/ + build/build-kernel.sh remain the
single source of truth.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017wB8KB3MMQztRDXCMCkPrf
The kernel build itself is the gate; uploading the zImage/dtb can fail on an
account-wide GitHub artifact-storage-quota hit unrelated to the build. Mark the
upload continue-on-error so a good build isn't red-X'd by it. Outputs also remain
on the self-hosted runner host.
Verified: kernel-build compiled the RV1106 6.18.46 kernel end-to-end on
bfe-mpc-0640 (zImage 8.25 MB + rv1106-warden.dtb 37 KB); the patch series applied
(issue #1 fix confirmed).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017wB8KB3MMQztRDXCMCkPrf
The kernel-build job set WORK inside the repo checkout, so build-kernel.sh's
`git apply` ran from a subdirectory of the warden-sdk git repo — where git apply
silently ignores out-of-subdir paths and exits 0 without applying anything. The
build then failed at the (never-created) rv1106-warden.dtb target, and the
unconditional `echo "applied"` had masked it.
- ci.yml: WORK -> ${{ runner.temp }}/kbuild-out (outside the checkout); artifact
paths follow.
- build-kernel.sh: the patch loop now fails loudly instead of echoing on silent
failure, and a post-apply sentinel assert (rv1106-warden.dts must exist) makes a
silent no-op impossible to ship — the regression guard for this class.
Refs #1.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017wB8KB3MMQztRDXCMCkPrf
Make the self-hosted kernel-build runnable with zero manual toolchain/python setup
on the runner host (docs/ci-cd.md steps 3-4 move into the workflow):
- build/build-kernel.sh: honor a caller-provided CROSS_COMPILE (default stays the
Luckfox uclibc prefix). The kernel is freestanding, so a generic arm cross
compiler links it.
- kernel-build job: set CROSS_COMPILE=arm-linux-gnueabihf- (Debian
gcc-arm-linux-gnueabihf, already on 0640) instead of depending on the ephemeral
SDK checkout path; symlink python->python3 into $RUNNER_TEMP/bin on $GITHUB_PATH
for the bare-`python` SDK quirk.
- docs/ci-cd.md: only steps 1-2 (systemd service + cgroup cap) still need 0640 sudo.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017wB8KB3MMQztRDXCMCkPrf
Point the badges job's branch guard at refs/heads/main and update the README /
ci-cd docs to name main as the canonical branch, ahead of establishing main from
bringup as the SDK's first main commit.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017wB8KB3MMQztRDXCMCkPrf
noah
BFE Engineering