zara-sysadmin: Linux/NixOS expert-system plugin #6

New issue

Closed

opened 2026-08-31 18:42:29 +00:00 by lost-rob0t · 0 comments

lost-rob0t commented

2026-08-31 18:42:29 +00:00

(Migrated from github.com)

Plugin

zara-sysadmin

Goal

Build a reusable Linux/NixOS operations expert for Zara. This should reason from structured system facts and diagnostics, not just feed arbitrary shell output to an LLM.

Initial domain tools

service status/start/stop/restart;
journal/log queries;
process inspection;
port/listener inspection;
disk/filesystem usage;
memory/CPU/load inspection;
network/interface/route/DNS inspection;
systemd unit inspection;
Nix generation/profile inspection;
bounded Nix build/rebuild/check operations;
package/config diagnostics where safely observable.

Use zara-shell as a lower-level execution dependency where appropriate, but keep the public sysadmin interface typed and task-oriented.

Expert-system direction

Ship Prolog rules/facts for symptom -> hypothesis -> diagnostic -> verification chains. Prefer zara-expert for reusable expert hosting when available.

Examples:

failed service -> inspect unit result and recent journal;
running service + closed port -> inspect listener/socket/config;
DNS failure -> distinguish resolver, route and upstream failures;
failed Nix switch -> retain build/activation evidence and generation state.

Safety

read-only diagnostics by default;
mutations are explicit operations;
bounded logs/output/runtime;
no silent privilege escalation;
never report remediation as successful without post-change verification.

Acceptance criteria

typed system diagnostic tools exist
Prolog rules can derive at least several diagnostic paths from structured facts
remediation actions have explicit verification steps
NixOS/systemd support is first-class
failure evidence is preserved structurally
tests use fake command/system backends and need no root/network
README documents supported platforms, permissions and expert rules
registry validation, plugin tests and nix flake check pass

## Plugin `zara-sysadmin` ## Goal Build a reusable Linux/NixOS operations expert for Zara. This should reason from structured system facts and diagnostics, not just feed arbitrary shell output to an LLM. ## Initial domain tools - service status/start/stop/restart; - journal/log queries; - process inspection; - port/listener inspection; - disk/filesystem usage; - memory/CPU/load inspection; - network/interface/route/DNS inspection; - systemd unit inspection; - Nix generation/profile inspection; - bounded Nix build/rebuild/check operations; - package/config diagnostics where safely observable. Use `zara-shell` as a lower-level execution dependency where appropriate, but keep the public sysadmin interface typed and task-oriented. ## Expert-system direction Ship Prolog rules/facts for symptom -> hypothesis -> diagnostic -> verification chains. Prefer `zara-expert` for reusable expert hosting when available. Examples: - failed service -> inspect unit result and recent journal; - running service + closed port -> inspect listener/socket/config; - DNS failure -> distinguish resolver, route and upstream failures; - failed Nix switch -> retain build/activation evidence and generation state. ## Safety - read-only diagnostics by default; - mutations are explicit operations; - bounded logs/output/runtime; - no silent privilege escalation; - never report remediation as successful without post-change verification. ## Acceptance criteria - [ ] typed system diagnostic tools exist - [ ] Prolog rules can derive at least several diagnostic paths from structured facts - [ ] remediation actions have explicit verification steps - [ ] NixOS/systemd support is first-class - [ ] failure evidence is preserved structurally - [ ] tests use fake command/system backends and need no root/network - [ ] README documents supported platforms, permissions and expert rules - [ ] registry validation, plugin tests and `nix flake check` pass