Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Trace agent execution by collecting spans and building a trace tree for a task
.claude/skills/ruvnet-observe-trace/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | -61% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-06 | ✓→✗ | ▼ Worse | -59% | 0% |
Collect distributed trace spans for a task and build a visual trace tree showing the execution flow, timing, and bottlenecks.
When you need to understand how a task was executed across agents -- which spans ran, how long each took, where bottlenecks occurred, and how agents coordinated.
mcp__plugin_ruflo-core_ruflo__memory_search --namespace observability (or memory_list) to retrieve all spans matching the <task-id>. The memory_* tool family routes by namespace; agentdb_hierarchical-* does NOT (it routes by tier working|episodic|semantic), so use memory_* here. See ruflo-agentdb ADR-0001 §"Namespace convention".parentSpanId references, with the root span at the topmcp__plugin_ruflo-core_ruflo__agentdb_context-synthesize to combine span metadata into a narrative summary of the execution flowbashnpx @claude-flow/cli@latest memory search --query "trace spans for task TASK_ID" --namespace observability
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-15 | fail→pass | 9,565 | 1,891 | -80% | 1 | 1 | 0% | 1,894 | 734 | -61% | 0 | 0 | — |
case-09 | pass→pass | 4,444 | 1,471 | -67% | 1 | 1 | 0% | 850 | 612 | -28% | 0 | 0 | — |
case-01 | fail→fail | 17,250 | 4,397 | -75% | 1 | 1 | 0% | 3,618 | 731 | -80% | 0 | 0 | — |
case-02 | fail→fail | 22,900 | 4,507 | -80% | 1 | 1 | 0% | 5,093 | 719 | -86% | 0 | 0 | — |
case-03 | fail→fail | 16,973 | 4,145 | -76% | 1 | 1 | 0% | 3,569 | 735 | -79% | 0 | 0 | — |
case-14 | fail→fail | 8,502 | 1,804 | -79% | 1 | 1 | 0% | 1,496 | 738 | -51% | 0 | 0 | — |
case-04 | fail→pass | 8,626 | 3,183 | -63% | 1 | 1 | 0% | 1,422 | 969 | -32% | 0 | 0 | — |
case-05 | pass→pass | 15,451 | 12,844 | -17% | 1 | 1 | 0% | 2,680 | 2,707 | +1% | 0 | 0 | — |
case-06 | pass→fail | 11,881 | 11,269 | -5% | 1 | 1 | 0% | 2,394 | 978 | -59% | 0 | 0 | — |
case-07 | pass→pass | 13,699 | 12,122 | -12% | 1 | 1 | 0% | 2,634 | 2,860 | +9% | 0 | 0 | — |
case-08 | fail→pass | 11,703 | 4,182 | -64% | 1 | 1 | 0% | 1,249 | 1,219 | -2% | 0 | 0 | — |
case-10 | pass→pass | 13,566 | 2,656 | -80% | 1 | 1 | 0% | 2,544 | 786 | -69% | 0 | 0 | — |
case-11 | pass→pass | 15,267 | 14,116 | -8% | 1 | 1 | 0% | 2,819 | 3,147 | +12% | 0 | 0 | — |
case-12 | pass→pass | 12,864 | 2,009 | -84% | 1 | 1 | 0% | 2,130 | 724 | -66% | 0 | 0 | — |
case-13 | pass→pass | 13,855 | 5,380 | -61% | 1 | 1 | 0% | 2,171 | 1,215 | -44% | 0 | 0 | — |
case-16 | pass→pass | 11,718 | 2,272 | -81% | 1 | 1 | 0% | 2,191 | 719 | -67% | 0 | 0 | — |
case-17 | pass→pass | 4,693 | 1,585 | -66% | 1 | 1 | 0% | 872 | 702 | -19% | 0 | 0 | — |
case-18 | pass→pass | 4,554 | 1,257 | -72% | 1 | 1 | 0% | 837 | 557 | -33% | 0 | 0 | — |
case-19 | fail→pass | 10,207 | 3,972 | -61% | 1 | 1 | 0% | 1,751 | 1,209 | -31% | 0 | 0 | — |
case-20 | pass→pass | 9,190 | 1,738 | -81% | 1 | 1 | 0% | 1,624 | 680 | -58% | 0 | 0 | — |
case-21 | pass→pass | 13,306 | 5,120 | -62% | 1 | 1 | 0% | 2,242 | 1,292 | -42% | 0 | 0 | — |
case-22 | pass→pass | 6,324 | 3,347 | -47% | 1 | 1 | 0% | 1,134 | 1,057 | -7% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +14 percentage points is the difference between those two pass rates over the 18 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.