Help us improve
Share bugs, ideas, or general feedback.
From argus-jvm-harness
Monitors a running JVM process over a time window for GC regression, heap leaks, thread growth, and pause-time spikes. Produces a severity-ranked report with recommended JVM flags via the Argus CLI.
npx claudepluginhub rlaope/argus --plugin argus-jvm-harnessHow this skill is triggered — by the user, by Claude, or both
Slash command
/argus-jvm-harness:argus-jvm-harnessThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
A thin Claude Code wrapper around `argus harness <pid>`. The CLI does the work; this skill picks the right invocation, bootstraps the CLI when needed, and summarises the resulting session in natural language.
JVM performance tuning - GC optimization, profiling, memory analysis, benchmarking
Diagnoses memory leaks in processes (JVM, Node.js) by profiling allocations, analyzing heap dumps, and tracing code paths. Use when RSS grows monotonically or OOM kills occur.
Detects memory leaks in Node.js, Python, and JVM apps by analyzing event listeners, closures, unbounded caches, and retained references. Use for troubleshooting memory growth.
Share bugs, ideas, or general feedback.
A thin Claude Code wrapper around argus harness <pid>. The CLI does the work; this skill picks the right invocation, bootstraps the CLI when needed, and summarises the resulting session in natural language.
The user says any of: "monitor my JVM", "watch this process", "check why GC keeps spiking", "is there a leak in pid X", "tune the JVM at ", "harness", or anything implying a time-window health check rather than a single-snapshot one. For one-shot questions ("how does my heap look right now?"), prefer argus doctor <pid> instead.
A target PID. If the user hasn't given one, ask them or run argus ps first to list candidates.
Locate the CLI. Run command -v argus. If absent, install it once with the official installer (no flags, public host):
curl -fsSL https://raw.githubusercontent.com/rlaope/Argus/master/install.sh | bash
then re-resolve PATH with source ~/.zshrc or use the absolute path ~/.argus/bin/argus.
Pick a profile.
--profile=quick (2 s tick, default 1 min): use for fast spot-checks and demos.--profile=deep (10 s tick, default 30 min): default for real investigations.
The user's words usually decide: "quick look" → quick; "monitor for 10 minutes / for a while / overnight" → deep.Run with a JSON sink so the model can read findings programmatically:
argus harness <pid> --profile=deep --duration=<user-specified-or-30m> --out=/tmp/argus-harness.json --format=json
Stream stdout to the user — the live tick line is informative — but use the JSON file as the source of truth for the summary.
Summarise. After the command exits, read /tmp/argus-harness.json and report:
severity, title, and detail, ordered as the JSON returns them (engine already sorted by severity).hits count for any finding that fired more than once — that is the harness's signal of a persistent problem versus a transient spike.suggestedFlags across all findings, in a single block the user can paste into their JVM args.recommendations array already contains them).Do not modify the JVM. This skill is observation + recommendation only. Never invoke argus vmset, argus gcrun, argus heapdump, or argus jfr start automatically. If the recommendations suggest one of those, surface it as a suggestion and let the user run it.
--duration from the profile if the user did not specify one. Don't pick numbers out of thin air.findings[].severity, findings[].title, findings[].detail, findings[].recommendations[], findings[].suggestedFlags[], and counts.{critical,warning,info}.argus harness --help