system-check

system-check

Purpose

Give a fast, readable snapshot of the development environment's health and flag whether any remediation is needed. Combines standard Linux commands (free, uptime, df, ps) with a targeted look for leftover Claude-spawned processes, and decides on its own whether to escalate to /cleanup-claude-procs.

When to Activate

• User says "system check", "health check", "how's the system looking", "is everything OK"
• Before/after a long-running task when the user wants to confirm the box is healthy
• After multiple Claude sessions have been run and the user suspects cruft

Workflow

1. Gather core metrics (run in parallel)

free -h
uptime
df -h / /tmp 2>/dev/null | grep -v tmpfs
# Top 5 memory consumers
ps -eo pid,user,%mem,%cpu,etimes,command --sort=-%mem | head -6
# Claude-family process count (excluding this session's tree)
pgrep -af 'claude-code|@anthropic-ai/claude-code|playwright-mcp|chromium.*--headless' | wc -l
# Load vs. core count
nproc
# Local IP address
hostname -I | awk '{print $1}'

2. Interpret thresholds

Evaluate each signal and classify OK / WARN / ALERT:

Signal	OK	WARN	ALERT
Available memory	> 25%	10–25%	< 10%
Swap used	< 25%	25–60%	> 60% with free RAM < 20%
Load avg (1-min) / cores	< 0.7	0.7–1.5	> 1.5
Disk / usage	< 80%	80–90%	> 90%
Stray claude/playwright procs (ppid=1)	0	1–2	3+
Uptime	any	> 30 days	n/a

3. Decide on escalation

• ALERT on stray procs → automatically invoke /cleanup-claude-procs (dry-run first, then confirm with user before killing).
• WARN on stray procs → report them and suggest /cleanup-claude-procs, don't auto-run.
• ALERT on memory + stray procs present → same as above but lead with the memory concern.
• ALERT on disk → list largest dirs under $HOME and /tmp with du -sh (top 5).
• OK across the board → one-line "all green" summary.

4. Report format

Use a compact, scannable format:

SYSTEM CHECK — 2026-04-15 09:31
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
IP:      192.168.1.100
Memory:  3.0 GiB available / 4.0 GiB total    [OK]
Swap:    640 MiB / 2.0 GiB used               [OK]
Load:    0.23 0.10 0.03 (4 cores)             [OK]
Disk /:  42% used                             [OK]
Uptime:  1 day, 11h
Stray claude procs: 0                         [OK]

→ All green. Nothing to do.

For a WARN/ALERT, add a Recommendations section with the specific next action and offer to run it.

5. Offer follow-ups

Depending on findings, suggest:

• /cleanup-claude-procs — for stray processes
• du -sh ~/* | sort -h — for disk pressure
• sudo reboot — if uptime > 30 days AND swap pressure high
• Nothing — if everything is green

Examples

/oh-my-claudecode:system-check
/oh-my-claudecode:system-check --verbose

Notes

• Thresholds are tuned for low-memory hosts (4 GB RAM default).
• Never kills anything on its own — escalation to /cleanup-claude-procs still goes through that skill's own confirmation flow.
• If pgrep shows Claude procs but they all belong to the current session (check PPID chain via pstree), don't count them as stray.
• Keep the report to one screen. Verbose mode (--verbose) may add per-process detail and full df output.