メニュー
Repeatable macOS/Apple-Silicon benchmark + diagnostics scored PASS/WARN/FAIL with saved reports. Use when baselining a Mac, verifying it performs to spec, or tracking CPU/memory/disk/thermal health.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/laurigates/claude-plugins --skill macos-performance-benchmarkコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Install, configure and troubleshoot MCP servers. Use when adding/enabling servers, editing .mcp.json, fixing OAuth, or when a server runs stale code after an upstream fix.
FinOps snapshot — org billing, workflow stats, cache usage. Use when you want a high-level view of CI spending or workflow health before diving deeper.
GitHub Actions billing, workflow efficiency, and waste analysis at org or repo level. Use when investigating CI/CD costs, wasted runs, or optimizing triggers.
SKILL.md を表示中
| name | macos-performance-benchmark |
| description | Repeatable macOS/Apple-Silicon benchmark + diagnostics scored PASS/WARN/FAIL with saved reports. Use when baselining a Mac, verifying it performs to spec, or tracking CPU/memory/disk/thermal health. |
| args | [mode] |
| argument-hint | mode — diagnose | bench | full | report (default full) |
| allowed-tools | Bash(bash *), Bash(uname *), Bash(sudo bash *), Read, Grep, Glob |
| created | "2026-07-03T00:00:00.000Z" |
| modified | "2026-08-07T00:00:00.000Z" |
| reviewed | "2026-08-07T00:00:00.000Z" |
A repeatable, threshold-scored performance suite. Runs diagnostics and
benchmarks for CPU/thermals, memory/swap, disk/storage, and startup/background
load, writes a timestamped run under ~/.cache/macos-perf/, and generates a
markdown report with PASS/WARN/FAIL verdicts against tunable baselines.
| Use this skill when... | Use something else when... |
|---|---|
| You want a repeatable baseline — "is this Mac performing to spec?" scored PASS/WARN/FAIL | The Mac is hot/slow right now and you need live attribution — use macos-performance-triage |
| Tracking CPU/memory/disk/thermal health across runs, comparing before/after a change | The machine actually hung or panicked — use macos-incident-postmortem |
| Producing a saved report to keep or share | A security extension (Kandji ESF, XProtect, EDR) is the hot process — use endpoint-security-cpu |
| Benchmarking NVMe/AES/SHA/memory throughput against thresholds | You're reclaiming disk space — use macos-disk-usage |
This skill is the proactive baseline companion to the reactive
macos-performance-triage playbook: run this to know the machine's normal;
run triage when something is wrong now.
macOS-only. sysctl, pmset, powermetrics, macmon, diskutil, and the
Apple-Silicon counters are Darwin-specific. run.sh refuses on non-Darwin.
test "$(uname -s)" = "Darwin" || { echo "macos-plugin: not Darwin, refusing"; exit 1; }
Parse $ARGUMENTS for the run mode (default full):
| Mode | What runs | Time | sudo |
|---|
diagnose | 4 diagnostics + report | ~25s | no |
bench | 3 benchmarks + report | ~3 min | no |
full | diagnose + bench (default) | ~5 min | no |
diag-cpu / diag-memory / diag-disk / diag-startup | one diagnostic | <10s | no |
bench-cpu / bench-memory / bench-disk | one benchmark | ~1 min | no |
report | regenerate the latest run's report | instant | no |
report-list | list saved runs with PASS/WARN/FAIL counts | instant | no |
baseline-show | print this machine's recorded benchmark baseline | instant | no |
baseline-reset | clear the baseline (next bench run re-establishes it) | instant | no |
Add sudo in front for thermal-pressure data (powermetrics); everything else,
including CPU power via macmon, runs without sudo.
Run the bundled suite and surface the verdict. The scripts self-detect tools and skip gracefully when one is absent.
bash "${CLAUDE_SKILL_DIR}/scripts/run.sh" $ARGUMENTS
For thermal-pressure data (optional), prefix with sudo:
sudo bash "${CLAUDE_SKILL_DIR}/scripts/run.sh" diag-cpu
run.sh prints the generated report and its path
(~/.cache/macos-perf/<timestamp>/report.md). Summarize the Overall Status
and every WARN/FAIL, mapping each to the process or subsystem responsible. For a
hot process or hung machine, hand off to the reactive skills named in the
"When to Use" table.
bash "${CLAUDE_SKILL_DIR}/scripts/run.sh" report-list
Each saved run keeps its summary.tsv (one status row per check) and per-script
logs, so before/after comparisons and regressions over time are trivial.
Benchmarks are self-calibrating — no fixed thresholds ship in the repo (they'd
be wrong on every machine but the author's). The first benchmark run records
each score as this machine's baseline (best-seen), stored in
~/.cache/macos-perf/baseline.env. Later runs compare against it:
The degrade bands are env-overridable (MACOS_PERF_BENCH_WARN_DEGRADE,
MACOS_PERF_BENCH_FAIL_DEGRADE). Inspect or clear the baseline with the
baseline-show / baseline-reset modes.
Multi-core scaling efficiency is reported, never scored. The ratio
(multi-core throughput ÷ n × single-core) assumes every core is equivalent —
which is false on Apple Silicon's P+E topology, where a healthy 10P+4E machine
measures 24–27% simply because 4 of its 14 cores run at a fraction of P-core
throughput. No single floor is meaningful across symmetric and asymmetric
machines, so the number is emitted as an info line (with the P:E core split when
hw.perflevel* is available) and never contributes a PASS/WARN/FAIL. The
absolute multi-core throughput is still scored, against this machine's own
self-calibrating baseline.
Diagnostics keep absolute thresholds — disk-free %, RAM, memory pressure,
launch-item counts are health/hygiene checks, not performance scores, so they
don't self-calibrate. Those defaults suit a modern Mac and are individually
MACOS_PERF_*-overridable — see scripts/config.sh for the
full list, or drop a thresholds.local.sh beside it:
MACOS_PERF_RAM_MIN_GB=16 bash "${CLAUDE_SKILL_DIR}/scripts/run.sh" diagnose
Results default to ~/.cache/macos-perf/; override with MACOS_PERF_RESULTS_DIR.
macmon (Rust, sudo-free) is preferred for CPU power/thermals, matching the
macos-performance-triage toolkit; powermetrics (sudo) is the fallback and the
only source of thermal pressure. Install the sudo-free path with:
brew install macmon jq
| Context | Command |
|---|---|
| Fast health check, no sudo | bash "${CLAUDE_SKILL_DIR}/scripts/run.sh" diagnose |
| Full benchmark baseline | bash "${CLAUDE_SKILL_DIR}/scripts/run.sh" bench |
| Machine-readable verdicts | `awk -F'\t' '$1~/^(PASS |
| List runs + counts | bash "${CLAUDE_SKILL_DIR}/scripts/run.sh" report-list |
| Show / reset baseline | bash "${CLAUDE_SKILL_DIR}/scripts/run.sh" baseline-show |
| Retune a diagnostic threshold | MACOS_PERF_<KEY>=<value> bash "${CLAUDE_SKILL_DIR}/scripts/run.sh" <mode> |
macos-performance-triage — reactive live triage (this skill's companion)macos-incident-postmortem — after a hang/panicmacos-disk-usage — disk-space forensics and reclamation