Skill Creator Agnostic

This skill is a thin overlay on Anthropic's skill-creator. It does not replace the upstream workflow; it adds cross-runner and repo-level guardrails so skill work stays reliable across Codex, Copilot, Opus, and similar agents.

Before benchmarking, read references/benchmark-contract.md.

On Windows or when running from PowerShell, also read references/windows-powershell-benchmarking.md.

Critical

• Start from Anthropic's skill-creator workflow. Use this skill to add environment and repo guardrails, not to fork or replace the upstream skill.
• Do not edit third-party skills such as Anthropic's skill-creator to encode repo-specific behavior. Keep those rules in repo-managed files and companion skills instead.
• Do not assume any specific runner CLI exists. Choose the benchmark runner from what is actually available in the current environment.
• When the chosen runner supports sub-agents, background agents, or equivalent parallel executions, use that capability for MEASURED benchmarks by default instead of running evals serially.
• Keep all eval workspaces under a temp root such as $env:TEMP/<skill-name>-workspace/, never inside the source repo.
• For repo-managed skills, keep skills/<name>/, ~/.claude/skills/<name>/, and ~/.agents/skills/<name>/ in sync before calling the work done.
• Every repo-managed skill must keep a per-skill evals/evals.json.
• If an eval entry declares files, treat those paths as skill-relative fixtures and stage them into the temp workspace for both benchmark configurations.
• Benchmark directories must follow iteration-N/eval-name/{config}/run-N/ exactly, and the eval directory itself must start with eval-; do not flatten files directly under with_skill/ or without_skill/.