Analyze Chrome, Chromium, Electron, React DevTools, or Perfetto-compatible JSON traces and audit user-reported profiling findings without loading large artifacts into context; prove trigger-to-render/layout chains, separate measured facts from source inference, find exact code choke points, classify forced layout and render fanout, implement semantically safe fixes, and verify behavior plus repository budgets.
SkillScore68Maintained
Automatic
◆Frontend / UI pack
Last verified 8/29/2026
What it changes
Passed 12 of 18 test cases with the skill, and 8 of 18 without it.
Without skill45%
With skill68%
Checks it passed that the model alone missed
The command passes --top 40 to limit peak output items.
The command passes --min-ms 16 to filter sub-frame events.
The launch environment sets VITE_TUTTI_WHY_DID_YOU_RENDER to 0.
and 9 more in the full report
6 fixed9 passed either way6 still failing1 broke
Pass-rate lift
+23 pp
Tokens
+84%
Agent turns
0%
Response time
-23%
Model gemini-3.6-flashMeasured 8/29/2026v115 of 22 cases passed with skill · 18 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 6
Prompt
Where in the project repository are testing conventions documented, and what CLI tool/script lists performance agent GUI scenarios?
What was checked, without and with the skill
Checked
Without skill
With skill
The response cites docs/conventions/testing.md for repository testing standards.
Missed
Passed
The response specifies running pnpm perf:agent-gui -- --list-scenarios to list available scenarios.
Missed
Passed
case-18 · gemini-3.6-flash · 8/29/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md187 lines2 filesincludes scripts
Analyze Performance Traces
End with an evidence chain from trigger to exact source. Fix the earliest proven cause. Never turn an API name or an inclusive duration into a causal claim without checking execution context.
Choose the evidence mode
State which mode applies before making findings:
Trace-backed: a trace is supplied or discoverable. Durations, processes, threads, and event ordering may be reported from it.
Reported-finding audit: the user supplies durations or a chain but not the artifact. Treat those details as leads. Confirm current source paths and trigger conditions; label the durations, thread, and invalidation scope unverified.
Source-only: no trace-derived lead exists. Report hypotheses, not measured bottlenecks.
Do not block a reported-finding audit merely because the original trace file is absent. If the user asks for safe implementation after the source chain is proven, proceed within authorization; require a comparable post-change trace before claiming millisecond or frame-rate improvement.
If neither a trace nor a concrete lead exists, ask for the smallest reproducible capture unless the user explicitly wants source-only analysis.
Start safely
Read repository instructions, nearest area instructions, relevant architecture docs, current diff, and validation policy. Record Git baseline before editing.
Discover trace artifacts without printing them wholesale. For each candidate:
ls -lh TRACE.json
head -c 512 TRACE.json
tail -c 512 TRACE.json
Record trace revision/build, production versus development mode, profiling hooks, source maps, screenshots, React tracks, renderer count, and capture window. Separate profiler startup and instrumentation overhead from product work.
Prefer a repository-owned performance runner when it reproduces the same interaction. List scenarios first; a convenient but different scenario is not proof. In this repository, inspect docs/conventions/testing.md and use pnpm perf:agent-gui -- --list-scenarios when relevant.