- F# 97.2%
- Shell 2.8%
1b42c3dae7
Intrinsic-path sibling of the merged sidecar empty-read fix. An empty CHRONICON read (CREATE inotify event at 0 bytes, body not yet landed) is now treated as "not ready" -- skipped, not quarantined as MissingField "provenance" -- and re-read on the writer's subsequent Changed event. |
||
|---|---|---|
| docs | ||
| src/Watcher | ||
| systemd | ||
| tests | ||
| .build-image | ||
| .gitignore | ||
| AGENTS.md | ||
| AGENTS.source.md | ||
| build | ||
| CLAUDE.md | ||
| Directory.Build.props | ||
| global.json | ||
| LOGOS-WATCHER.slnx | ||
| README.md | ||
NEXUS LOGOS Watcher
Long-running daemon that watches /NEXUS/LOGOS/, batches filesystem
writes from other NEXUS projects, and commits + pushes them to the
central LOGOS forgejo remote — one commit per origin_project per
debounce window.
Why this exists
/NEXUS/LOGOS/ is the central knowledge store. Project sessions
(Claude Code, Codex CLI, etc.) write findings to
/NEXUS/LOGOS/CHRONICON/<uuid>.toml, notes to
/NEXUS/LOGOS/HYPOMNEMATA/<project>/, and occasionally edits to
CORPUS/ or ARMAMENTARIUM/. Those sessions run inside the
originating project's repo, not inside LOGOS, so LOGOS never sees a
Stop hook fire — there's nothing to trigger a commit or push.
This daemon fills that gap: it watches LOGOS, attributes each write to its originating project, and commits under that project's identity so the resulting history reads correctly.
Quick context
flowchart LR
P1[Project A session]
P2[Project B session]
LOGOS[(/NEXUS/LOGOS/ working tree)]
W{{LOGOS Watcher daemon}}
FG[(forgejo NEXUS/LOGOS)]
P1 -- writes CHRONICON/<uuid>.toml --> LOGOS
P2 -- writes HYPOMNEMATA/B/note.md --> LOGOS
LOGOS -- inotify --> W
W -- commit as Project A --> LOGOS
W -- commit as Project B --> LOGOS
W -- push origin main --> FG
Documentation
- Architecture — pipeline stages, modules, state machines, sequence diagrams.
- Protocol — attribution rules, sidecar TOML format, quarantine semantics.
- Operations — systemd unit, logs, alarm recovery, push-failure handling.
- Testing — test layout, what each package contributes, how to add tests, how to diagnose flakes.
Build
./build # build + test
./build publish # produce ./out/Watcher/Watcher
Requires Docker. Everything else (the .NET SDK, NuGet packages, build
tooling) runs inside a pinned container; nothing else to install. Full
contract: /NEXUS/LOGOS/CORPUS/conventions/build-system.md
(cross-repo link; opens in forgejo).
The systemd user unit at systemd/nexus-logos-watcher.service
launches the container-built binary at out/Watcher/Watcher. Build
once with ./build publish before systemctl --user start.
Because the unit is a user service (not a system service), it
runs only while the installing user has at least one active login
session — ssh, console, graphical, or RDP (xrdp). It does not start
at machine boot unless loginctl enable-linger <user> is set. Full
lifecycle, including how to make it run continuously, is in
docs/operations.md → When the watcher runs.