- New site: https://agent0.localhost -> reverse_proxy 192.168.0.141:8090
(first non-localhost backend; remote LAN host running uvicorn)
- Config in dedicated /etc/caddy/conf.d/agent0.localhost.conf
- Regenerated mkcert cert: 19 -> 20 SANs, valid until 2028-10-21
- Key learning: *.localhost wildcard is NOT honored by curl/system
trust even for single-label subdomains - .localhost is a special-use
TLD per RFC 6761; every subdomain needs an explicit SAN. Corrects
prior assumption in stack.md and manual-overrides.md.
Updates: docs/tech/stack.md, docs/tech/manual-overrides.md, CHANGELOG.md
- Added platform.localhost (-> localhost:10080) and packages.localhost
(-> localhost:20080) to stack.md sites table (total now 17)
- Documented onlyoffice.docs.localhost (pre-existing in Caddy config, was undocumented)
- Regenerated mkcert cert with explicit SANs for the 3 new hostnames
(onlyoffice.docs.localhost is multi-label, not covered by *.localhost wildcard)
- Fixed wildcard note: multi-label hosts require explicit SANs
- Updated manual-overrides with cert regeneration row + site delta
- Updated CHANGELOG with cert regen + site additions
Record the Caddy reverse-proxy at /etc/caddy/ (14 *.localhost sites,
mkcert TLS, systemd caddy.service) as a manual override in
docs/tech/manual-overrides.md - no caddy Ansible role exists yet.
Add a 'Local Dev Services' section to docs/tech/stack.md listing all
*.localhost sites and useful Caddy commands.
New learning record docs/learnings/2026-07-06-caddy-onlyoffice-mcp-sse-
investigation.md: Caddy emits 'context canceled' SSE aborts (~1.5ms) on
onlyoffice.localhost/mcp -> localhost:3847. Upstream verified alive
(HTTP 406 to plain GET = expected). Handed off to GitLab issue
satware/mcp/onlyoffice#92 with 5 hypotheses, 7 tasks, and a proposed
flush_interval -1 + transport http { versions 1.1 } mitigation (not
applied yet to avoid masking root cause).
- Added learning doc for 2026-07-03 session: diagnosed SIGSEGV in
XKeysymToKeycode (missing NULL check on XOpenDisplay), forked Remmina
to gitlab.com/satware, built with FreeRDP3, verified fix, submitted MR.
- Added manual override entry for patched remmina-plugin-rdp.so in
~/.config/remmina/plugins/
- Add hyprland.lua.j2 Ansible template with full Lua DSL config
- Update tasks to deploy hyprland.lua instead of hyprland.conf
- Add docs/tech/hyprland-lua-migration.md with API reference,
dispatcher mapping, pitfalls, and validation commands
- Update keybindings doc to reflect .lua config format
Co-authored-by: Junie <junie@jetbrains.com>
- Deleted alesi-routine-morning.service and .timer from ~/.config/systemd/user/
- Timer failed daily at 07:45 (exit 128, git outside repo)
- daily-routine.sh script retained for manual use
- Updated CHANGELOG.md and workstation-health.md
Replace obsolete WebGPU flags with state-of-the-art NVIDIA VA-API
hardware decode config for Chrome 148 on Wayland/Hyprland.
- Disable seccomp sandbox to fix NVIDIA GBM "Permission denied" error
- Disable VulkanFromANGLE/DefaultANGLEVulkan (incompatible with Wayland)
- Force ANGLE->OpenGL backend to avoid Vulkan/Wayland conflicts
- Enable modern VA-API flags: VaapiOnNvidiaGPUs, AcceleratedVideoDecodeLinux*
- Add gnome-libsecret password store for gnome-keyring integration
- Install libva-utils and libva-nvidia-driver packages
- Update driver version to 580.159.03 in hardware inventory
WebGPU remains hardware accelerated via ANGLE+OpenGL. Chrome verified
working with full video decode acceleration (H.264, VP9, HEVC) via NVDEC.
- Post-update fio benchmark: random read +12.4%, write +15.9% IOPS
- Latency reduced 10-14% for random workloads
- SMART health unchanged: 0 media errors, 0 error log entries
- Firmware research status updated to complete
- Pre-benchmark checklist marked done
Comprehensive benchmark of Samsung PM9A1 NVMe SSD before firmware
update from GXA7401Q to GXA7802Q. Includes SMART health, hdparm,
and fio results (seq/rand read/write, 4K blocks, iodepth=32).
- Researched 7 component categories for available firmware updates
- NVMe SSD (PM9A1): update available GXA7401Q -> GXA7802Q
- Motherboard BIOS F53 is current (no update needed)
- SATA SSD 840 EVO, GPU GTX 1050 Ti: EOL, no updates
- NIC/CPU microcode: managed via kernel/firmware packages
- USB peripherals: no LVFS/fwupd support
- Recommend installing fwupd for future firmware management
- Created crash analysis for 2026-03-17 OOM event.
- Documented stability test results and memory tuning effects.
- Added PARA-structured workflows for EOD protocol.
- Updated workstation health report for current 16GB swap status.
- Delete deprecated CLAUDE.md
- Create root-level AGENTS.md for AI assistant guidance
- Create SDD constitution in .specify/memory/constitution.md
- Align documentation with PARA structure (docs/product/project-brief.md, docs/tech/stack.md, docs/plans/)
- Add standardized YAML frontmatter to all documentation files
- Update README.md with saTway branding and structure
- Synchronized Ansible roles with removed redundant packages.
- Performed system audit (performance, boot, disk usage).
- Organized home directory by removing 22 clean remote repositories.
- Reclaimed ~21GB of disk space in ~/Projects.
- Retained dirty repositories and those without remote origins for safety.
- Updated documentation (performance tuning, software stack, setup).
- Created NEXT_STEPS.md and documented learnings.
mw
ja
