Agent skill · data analytics · jeremylongshore
cohere-performance-tuning
Measure and improve Cohere latency, throughput, retrieval quality, streaming, batching, and vector representation without fabricated benchmarks. Use when optimizing a Cohere workload. Trigger with "Cohere performance", "Cohere latency", or "optimize Cohere".
Why this skill is useful
Provides specific strategies and code snippets for optimizing Cohere API performance that the AI wouldn't reliably generate on its own.
What it needs
Requires cohere-ai, redis installed locally. Requires cohere account access. About 2k tokens when loaded. Last updated 2026-08-07. 2,604 stars on the source repository.
What this skill does
Cohere Performance Tuning Overview Tune from workload-specific measurements and representative quality thresholds rather than copying universal latency numbers or synthetic benchmark claims. Prerequisites The target repository, runtime, environment, and accountable owner An approved Cohere team and key for any live verification Current quality, security, privacy, capacity, and change-control requirements Tool Discipline Use Read, Glob, and Grep to inspect code, configuration, and evidence. Use WebFetch only for current Cohere primary documentation. Use Write or Edit only when the user requested implementation and the exact target files are known; never write credentials or customer content. Current Contract Resolve models and their current context and output limits from the live catalog. Measure Chat time to first event and completion separately from application queue and retrieval time. Batch Embed inputs within documented request constraints and track inputs per minute. Choose Rerank v4 pro or fast from measured relevance and latency on representative queries. Authentication Use an environment-specific key injected from an approved secret manager. Never print, persist, commit, or place COAPIKEY in an example. Confirm access with the least costly bounded operation appropriate to the task, and treat key creation, rotation, revocation, role changes, and production-capacity requests as owner-approved actions. Instructions 1. Define a representative workload, quality floor, latency objective, concurrency, and cost ceiling. 2. Instrument queue, retrieval, provider, first-event, completion, and post-processing durations. 3. Establish a cold and warm baseline with fixed inputs and resolved model IDs. 4. Test one variable at a time: model, context size, output bound, stream mode, Embed batch size, Rerank candidate count, or cache. 5. Reject changes that improve speed while violating quality, citation, safety, or cost thresholds. 6. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills jeremylongshore/cohere-performance-tuning