Junhyun Lim
Community@ensj · Santa Clara
Junhyun Lim. Jimmy. Ensj. I'm a student studying computer science and mathematics!
Agent Skills by Junhyun Lim
Showing 12 vetted skills indexed across 1 GitHub repositories.
ensjclassify-then-schema
Classify input text and generate a JSON Schema for a structured response.
ensjresearch-orchestration
Orchestrates a five-step pipeline that plans, searches, verifies, and synthesizes cited research briefs.
ensjresearch-lean-linear
Generates cited research briefs by planning queries, searching the web, and distilling sources.
ensjcompare-compositions
Compares composition methods on a quality-cost frontier and writes composition-profile.json.
ensjeval-sweep
Runs adaptive variant sweeps on atomic skills with statistical verdicts and provenance-checked scoring.
ensjrun-composition
Runs typed skill compositions end-to-end and measures predicted versus actual performance deltas.
ensjeval-writing-styles
Sweeps prompt writing styles over shared eval cases and reports a quality-versus-token Pareto frontier.
ensjassemble-tier
Assembles the cheapest model-tier assignment for a composition that ties the all-opus baseline.
ensjpackage-composition
Packages a measured prompt composition into a portable pure-prose SKILL.md with frozen prompts and evals.
ensjmodel-ladder
Finds the cheapest model tier where an atomic skill produces a stable winner.
ensjtier-ladder
Certifies the cheapest model tier that ties the top tier for an atomic skill.
ensjtest-packaged-skills
Validates packaged composition skills structurally and runs live end-to-end dispatch tests.
Frequently Asked Questions About Junhyun Lim
FAQPage SchemaWhat tasks can I accomplish with ensj's skill-eval-framework skills?▼
You can run variant eval sweeps, compare composition topologies on a quality-cost frontier, test writing styles, find the cheapest viable model tier, assemble cost-optimized model-heterogeneous compositions, and package or end-to-end test compositions as portable SKILL.md files.
Who should use these skills?▼
Engineers and researchers building or benchmarking LLM skills who need rigorous, seed-driven evaluation of prompt variants, model tiers, and composition topologies. It suits anyone optimizing the quality-versus-cost tradeoff of skill pipelines before shipping them.
How do the evaluation and tiering skills work in practice?▼
Skills invoke frozen pattern scripts like patterns/eval-sweep.js and patterns/tier-ladder.js over shared cases and seeds. Results are scored against rubrics, and outputs such as composition-profile.json, tier-profile, and model-profile.json are written for downstream decisions.
What are the prerequisites for using these skills?▼
You need the skill-eval-framework environment with its frozen patterns scripts, defined eval cases, and access to the haiku, sonnet, and opus model tiers. Compositions must have frozen winning prompts before packaging or tier assembly.
How does the framework control evaluation cost?▼
It uses adaptive budget-driven seed rounds that stop once a precision-width rule decides an outcome, plus tier ladders and cheapest-tier assembly that downgrade steps only when they tie the all-opus baseline, minimizing spend without sacrificing measured quality.