Describe the incident. The plugin selects the skills.
This is the BTrace plugin marketplace for Claude Code, Codex and Pi. Each plugin has one shared implementation. Ask in operational language. You do not need to know BTrace syntax. The plugin plans the probe, attaches it, reads the output and removes it.
// An example plan. The skill composition is from the plugin README: slow endpoint in one Kubernetes pod.
{{ hostNote }}
Five plugins. One repository.
Select a plugin to see what it contains and what to ask it.
> {{ pSel.name }}{{ pSel.title }}
{{ pSel.desc }}
The situation selects the skills.
btrace-observability bundles eleven skills. Select a situation. The skills that the plugin combines for it come on.
{{ sk.name }}
{{ sk.desc }}
A report, not a guess.
The analysis skills require evidence for each finding. The probe skills keep the scope narrow and remove the probe at the end.
> evidence // jafar-perf
- [{{ ev.n }}]{{ ev.t }}{{ ev.d }}
> scope and safety // btrace-observability
- [{{ sa.n }}]{{ sa.t }}{{ sa.d }}
History from JFR. Proof from a live probe.
Install jfr-analyzer next to btrace-observability. The jfr-btrace-interop skill sends the target, the time window and the hypothesis from one to the other.
{{ h.d }}
Write the skill once.
Put portable instructions and scripts in plugins/<name>/skills/. Add only the minimum metadata that each host needs. Skills copy the BTrace knowledge they use. A pre-push gate checks that copy against a BTrace checkout.
