AU

autoresearch-speed

Automates throughput optimization research for emulator performance.

Install

mkdir -p .claude/skills/autoresearch-speed && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/13349" && unzip -o skill.zip -d .claude/skills/autoresearch-speed && rm skill.zip

Installs to .claude/skills/autoresearch-speed

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Unified Super Mario Bros emulator speed-improvement workflow for this repo. Use when Codex is asked to optimize, profile, benchmark, or coordinate self-improvement research for Super Mario Bros NES throughput, including N-agent proposal and implementation waves, Modal benchmark tournaments, stale candidate replay, merge adjudication, or cleanup of speed-research worktrees.
375 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)
Advanced

Key capabilities

  • Optimize emulator speed around env_steps_per_sec.
  • Profile code performance for throughput gains.
  • Benchmark throughput gains without regressions.
  • Iterate on speed improvements with cumulative rounds.
  • Manage mutable artifacts outside the repository.
  • Report artifacts and next baseline upon completion.

How it works

The skill runs an optimization loop for Super Mario Bros NES emulator throughput, using a fixed baseline for comparison and specific commands for profiling, screening, and accepting changes.

Inputs & outputs

You give it
Code changes to the emulator, especially in src/emulator.rs, src/vec_env.rs, and src/py_api.rs
You get back
Report with phase, inputs, outputs, decision, and next action

When to use autoresearch-speed

  • Optimize emulator speed
  • Profile code performance
  • Benchmark throughput gains

About autoresearch-speed

Runs iterative improvement loops for emulator environment steps per second. It profiles and benchmarks changes to ensure speed gains without regressions in accuracy.

Unified Super Mario Bros emulator speed-improvement workflow for this repo. Use when Codex is asked to optimize, profile, benchmark, or coordinate self-improvement research for Super Mario Bros NES throughput, including N-agent proposal and implementation waves, Modal benchmark tournaments, stale ca

When not to use it

  • When the task is not related to Super Mario Bros NES throughput optimization.
  • When the task involves feature or accuracy regressions.
  • When the task requires creating or switching branches, forking workers, or using cloud services without explicit approval.

Limitations

  • It only counts valid env_steps_per_sec gains.
  • It requires a concrete speed mechanism and cheap falsification path for each candidate.
  • It does not permit changes to API shape, deterministic lanes, observation bytes, rewards, resets, terminations, infos, state loading, or benchmark workload.

How it compares

This skill provides a structured, iterative workflow for emulator speed optimization with strict guardrails and reporting, unlike ad-hoc performance tuning.

Compared to similar skills

autoresearch-speed side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
autoresearch-speed (this skill)01moReviewAdvanced
python-performance-optimization272moNo flagsIntermediate
python-repl64moReviewBeginner
optimizing-performance12moReviewIntermediate

Try saying

Example prompts that trigger this skill in your AI assistant.

Search skills

Search the agent skills registry