just-vibeGeneral · apply by default
/perf
Measure a performance problem and improve its cause
Use for measured performance repair; architecture scaling work needs workload-level evidence. Once the bottleneck is located in one layer, prefer react-rerenders, vite-bundle, vercel-performance, db-explain or ml-inference-perf.
Make it your own.
In Claude Code, use the slash command and add your context. In Codex, select perf from the just-vibe skill picker, then send the same brief.
Version 0.11.0 also supports /jv perf, /just-vibe perf and /jv:perf in Claude. See shortcut setup and context examples.
/just-vibe:perf Measure and fix slow product filtering; preserve search results./just-vibe:perf Optimize a slow search without changing ranking or using a warmer cache baseline./just-vibe:perf Inspect source without profiler data; provide hypotheses and a bounded measurement plan.What the agent does
- Define a repeatable workload and metric and record the baseline conditions.
- Profile the path to isolate the dominant resource or wait before changing code.
- After the change, compare latency, resource use and correctness under the same conditions.
Inputs
- slow operation, workload, performance objective, and measurement budget.
Optional context: scope, references, constraints, successCriteria, environment, mode, budget.
Scope
- Reads
- Measured bottlenecks and focused optimization; no blanket rewrites or unbounded production load.
- Writes
- Apply: only the requested local changes and relevant isolated verification. Inspect/plan requests remain inspection/planning. External actions require their exact action and target in session authorization.
- Mode
- Apply; slow operation, workload, performance objective, and measurement budget.
- Prerequisites
- Resolve the user brief and inspect the relevant project or supplied evidence. External capabilities are optional unless the selected action actually needs them.
Expected output
- Patch with the workload, sample size, before/after measurements, correctness checks and tradeoffs.
How the work is checked
- Improvement persists under comparable workloads; optimized behavior remains correct at a relevant boundary.
When to stop or clarify
- If measurement is unavailable, provide hypotheses without numerical claims. Stop when noise or resource limits prevent trustworthy conclusions.
Handling missing context
- Infer
- Resolve the named files, existing scripts, current task and earlier corrections from the conversation and repository.
- Assume
- Use the narrowest interpretation that completes a reversible local task; state a consequential assumption once.
- Ask
- Ask when competing targets or incompatible success conditions would change the result; continue independent inspection first.
Technical guidance
- Evidence
- Obtain a reproducible workload, baseline, resource measurements and an explicit user-impact metric.
- Method
- Profile the critical path before optimizing and compare under matched revision/data/cache conditions.
- Pitfall
- Lower allocation or fewer renders may not improve observed latency; averages hide regressions in tails or errors.
- Check
- Repeat the same workload, retain correctness checks and report distributions, tradeoffs and measurement uncertainty.
Situational decisions
When a faster implementation changes output quality or consistency: Reject the comparison or disclose the product tradeoff for an explicit decision.
The coding agent follows this workflow using its available tools. Installation does not grant service access or guarantee an outcome. Read the compatibility notes.