From linux-system-optimisation
Run a system benchmark tailored to the detected hardware on a Linux desktop. CPU (sysbench / stress-ng), GPU (glmark2 / vkmark / clpeak — chosen by GPU vendor), disk I/O (fio against the benchmark folder's filesystem), memory (sysbench memory + mbw), and a quick desktop-responsiveness probe. Saves timestamped JSON + Markdown summary to the user-defined benchmark folder. Triggers on "benchmark my system", "run benchmark", "rebench", or as a follow-up to `onboard`.
How this skill is triggered — by the user, by Claude, or both
Slash command
/linux-system-optimisation:benchmarkThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Single-shot benchmark run. Reads the detected `system_profile` from config, selects appropriate tools, runs them with conservative durations (default 30s per CPU/memory test, 1-minute fio job), then writes results.
Single-shot benchmark run. Reads the detected system_profile from config, selects appropriate tools, runs them with conservative durations (default 30s per CPU/memory test, 1-minute fio job), then writes results.
${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/linux-system-optimisation/config.json
If missing → call onboard first. The benchmarks_dir field tells you where to save results.
Inside benchmarks_dir:
<ISO-timestamp>/
├── results.json # machine-readable
├── summary.md # human-readable, ranked findings + suggested next steps
└── raw/ # raw stdout from each tool
Pick per detected hardware. Skip a group with a clear reason if a tool isn't applicable.
sysbench cpu --threads=$(nproc) --time=30 run — events/sec.stress-ng --cpu $(nproc) --cpu-method=matrixprod --metrics-brief --timeout 30s — bogo-ops.cat /proc/cpuinfo | grep MHz and sensors to see thermal/throttle behaviour.glmark2, vkmark, nvidia-smi --query-gpu=power.draw,temperature.gpu,clocks.gr --format=csv -l 1 during run.glmark2, vkmark, radeontop -d- (sample) during run.glmark2, intel_gpu_top -J (sample) during run.sysbench memory --memory-block-size=1M --memory-total-size=10G runmbw -q 1024 (1 GB block, MB/s)benchmarks_dir's filesystem (so the test is on the same disk class the user actually uses).
fio --name=randread --rw=randread --bs=4k --size=1G --runtime=60 --time_based --ioengine=libaio --direct=1 --output-format=json--rw=randwrite. Clean up the test file after.vmstat 1 5 and pidstat 1 5 while running a brief CPU+memory mixed load (stress-ng --cpu 2 --vm 2 --vm-bytes 1G --timeout 5s).{
"timestamp": "<ISO-8601>",
"host": { "kernel": "...", "distro": "...", "de": "...", "session": "..." },
"cpu": { "sysbench_events_per_sec": 0, "stress_ng_bogo_ops": 0, "max_freq_mhz": 0, "throttled": false },
"gpu": { "vendor": "...", "glmark2_score": 0, "vkmark_score": 0, "max_temp_c": 0, "max_power_w": 0 },
"memory": { "sysbench_mib_per_sec": 0, "mbw_mib_per_sec": 0 },
"disk": { "device": "...", "fs": "...", "randread_iops": 0, "randwrite_iops": 0 },
"responsiveness": { "load_avg_under_stress": 0, "context_switches_per_sec": 0 }
}
Brief, scannable. Lead with anything red-flagged: thermal throttling under CPU load, GPU power-cap hits, disk IOPS far below class expectations, swap thrash. Each finding pairs with a suggested action that optimise can act on.
Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub danielrosehill/claude-code-plugins --plugin linux-system-optimisation