Agent - Open Items

Only keep unfinished Agent work here.

Active

retention refinement for local backups/checkpoints
runtime install verification improvements
catalog hash validation
runtime removal / upgrade handling
runtime update process for dev containers
PATH normalization
shell profile consistency
runtime PATH injection
graceful shutdown verification
process reattachment on agent restart
SSH server install in dev provisioning
structured logging for Loki
provisioningComplete state surfacing in /status
game/dev template cleanup: remove node-exporter, keep Alloy preinstalled, and keep /etc/default/alloy fixed to /etc/alloy/config.alloy on port 12345
Alloy config label cleanup for game/dev containers: keep the required label contract minimal and consistent with dashboards/discovery (vmid, instance, container_type, optional metadata) and remove redundant labels if they add no value
verify automatic Alloy config refresh path on metadata changes after initial POST /config
resolve dev interactive shell boundary behavior: current live validation indicates PTY/dev console sessions can traverse upward from /home/dev/workspace, so confirm whether shell confinement is intended and, if so, implement it at the correct layer

fold Forge + NeoForge installer flow into one shared helper without changing external behavior
fold devcontainer runtime installers into a shared helper with thin per-runtime wrappers
split internal/http/agent.go by responsibility without changing package/API shape
add only small concrete HTTP helpers for repeated method/config/JSON response patterns
fold repeated addon/config checks into one helper such as HasAddon(...)
stop tracking runtime-generated files if they are not intentionally versioned
keep websocket library migration separate from the first cleanup wave; only revisit after behavior-preserving dedupe/split work is complete

prefer behavior-preserving folding over broad refactors
merge repeated flows, not concepts
reduce local complexity and file sprawl before introducing new abstractions
keep helpers small and concrete
treat Go/runtime/toolchain upgrades as separate validation work, not part of cleanup-only changes

do not add agent-side remote/offsite backup sync if PBS/platform layer is the intended durability path
agent backup scope is local, app-aware rollback
do not assume dev-container backups are agent-owned; current likely direction is platform/LXC-snapshot ownership for dev backups unless that decision changes

readiness-aware start/restart unification
guarded operation locking
crash recovery path consolidation
local Minecraft backup CRUD
pre-restore checkpoint hardening
code-server desired-state reconciliation during dev provisioning
code-server manual start endpoint uses service start path, not installer path
live backup/restore validation on real Minecraft servers
current Minecraft runtime provisioning split (vanilla internal profile vs normal fabric) and Forge/NeoForge first-start handling
dev shell / PTY behavior relative to the intended workspace boundary model

API response-shape normalization for frontend consumers
portal UX/polish
PBS/offsite backup strategy
dev-container snapshot orchestration / retention policy if that ownership stays outside the agent