tracelisted
Install: claude install-skill x-mesh/xm
<Purpose>
x-trace tracks xm tool executions across two artifacts:
- **Per-session traces** in `.xm/traces/*.jsonl` — agent call trees, estimated tokens, cost, elapsed time, and a git snapshot per session. These power the timeline / cost / replay / diff views, which the LLM renders by parsing JSONL.
- **A cross-tool activity ledger** in `.xm/last.json` — which tool last ran, on which commit, and how far HEAD has moved since. This is maintained by the `x-trace` CLI (`record` / `last` / `status` / `since` / `doctor`), not by LLM parsing.
No external dependencies.
</Purpose>
<Use_When>
- User wants to trace or observe multi-agent execution
- User says "trace", "execution log", "check cost", "token usage", "show timeline"
- User wants to compare two runs ("diff", "compare before and after")
- User wants to replay a previous execution ("replay", "reproduce")
- Other xm skills want to record agent calls for observability
</Use_When>
<Do_Not_Use_When>
- Simple single-step tasks with no agent fan-out
- Cost tracking for non-xm workflows
- Real-time monitoring (x-trace is post-hoc, not live)
</Do_Not_Use_When>
# x-trace — Agent Execution Tracing
## Model Routing
| Subcommand | Model | Reason |
|------------|-------|--------|
| `show`, `list`, `cost`, `diff` | **haiku** (Agent tool) | Read-only log parsing and display |
| `record`, `last`, `status`, `since`, `doctor` | **haiku** (Agent tool) | Deterministic CLI output — script decides, model relays |
| `replay` | **sonnet** | Requi