Fk Benchmarker — Development agent for Claude Code
Owns the measurement protocol - noise floor, warm-up, clock ramp, repeats, CUDA sync, peak VRAM, launch counts - and answers "is this speedup real?".
How to install Fk Benchmarker
Installs to ~/.claude/agents/kadirnar-fast-kernel-fk-benchmarker.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/kadirnar/fast-kernel/HEAD/.claude/agents/fk-benchmarker.md -o ~/.claude/agents/kadirnar-fast-kernel-fk-benchmarker.md Restart Claude Code, or start a new session, for it to be picked up.
What Fk Benchmarker does
name: fk-benchmarker description: Owns the measurement protocol - noise floor, warm-up, clock ramp, repeats, CUDA sync, peak VRAM, launch counts - and answers "is this speedup real?". Use when results look noisy or contradictory. tools: Read, Glob, Grep, Bash model: inherit
Explain and, if needed, re-run measurements with `fast-kernel eval --force --repeats N` or `fast-kernel baseline --force` (only when the baseline is suspect). Check GPU contention (`nvidia-smi`), clock state, thermal
Alternatives in Development
- Kernel Cuda Specialist 14.5k ★
- Peer Reviewer Demographics — A simulated peer reviewer specializing in population representativeness, demographic measurement, and populati 134 ★
- Book Disruptor — Chaos agent for the book pipeline 114 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Repo Sweeper
Read-only mechanical sweeps — PR and CI status across many PRs, worktree/branch leftovers, issue states, path
Docs Consistency Reviewer
Read-only cross-cutting doc consistency auditor for m3l-common. Checks version-floor consistency, submodule co
Comfy Runner
Exécute des runs ComfyUI longs ou multiples (sweeps, seeds, vidéo) via bin/studio et rapporte un résumé compac
Source Acquirer
Dispatched by acquire-sources for large worklists (≥ ~8 sources). Runs the resolve→curl→validate download batc
General Skills Executor
Generic isolated runner for noisy "fire-and-forget" skills whose raw output would pollute the main context — c
Performance Benchmarker
Use for designing and running performance benchmarks with actionable reporting.
Related Skills
Emailwarm
Claude skill for email warmup: warm a new sending domain so mail stops landing in spam. Audits SPF/DKIM/DMARC,
Correct Reference Now
Correct the reference clock when the model-known date diverges from the seeded floor
Late CLI
High-performance AI agent for long-horizon tasks and local models. Built on empirical research. 200k+ tokens o
