benchmark

Benchmark

Compare local code against real-world open-source implementations. Find high-quality references, analyze differences, and produce actionable improvement recommendations.

Workflow

Step 1: Scope

Identify what to benchmark from the user's request or argument:

• Pattern: a specific code pattern (e.g. "repository pattern", "middleware chain")
• Architecture: a structural approach (e.g. "hexagonal architecture", "monorepo layout")
• Feature: a concrete feature implementation (e.g. "auth with JWT refresh", "i18n setup")
• Library usage: how a library is used in practice (e.g. "Prisma with soft deletes", "Zustand store patterns")

Read the local implementation to understand:

1. Which files implement the pattern (use Glob + Grep)
2. The current approach, structure, and key decisions
3. Stack details (language, framework, libraries) to target relevant references

Ask the user to confirm scope if ambiguous (AskUserQuestion).

Step 2: Search for References

Use WebSearch to find 3–5 high-quality references. Search strategies:

"<pattern> site:github.com <stack>" — direct GitHub repos
"<pattern> best practices <framework> <year>" — articles and guides
"<pattern> example implementation <language>" — tutorials with code
"awesome-<topic> github" — curated lists

Filter criteria:

• Stars/popularity: prefer repos with >500 stars or articles from known sources
• Recency: prefer references updated within the last 2 years
• Stack match: same language/framework as the local codebase
• Production quality: skip toy examples, prefer real-world usage

Collect for each reference:

• URL
• Repo name or article title
• Why it's relevant
• Star count / authority signal

Step 3: Analyze References

Use WebFetch on the top 3 matches to extract:

• Approach: high-level strategy and design decisions
• File structure: how they organize the feature
• Key patterns: naming, error handling, testing, configuration
• Trade-offs: what they optimize for (simplicity, performance, extensibility)