Skill Creator

The skill's core function is to create and modify `SKILL.md` files and evaluation data (`evals.json`, `eval_metadata.json`). These artifacts serve as knowledge base components for other agents. If an attacker can inject malicious instructions or data into these created skill definitions or evaluation files, it could lead to poisoning the RAG or knowledge bases of agents that consume them.

Write the SKILL.md
Save test cases to evals/evals.json
Update the eval_metadata.json files and evals/evals.json with the assertions

The skill explicitly instructs to make skill descriptions 'a little bit "pushy"' to combat 'undertriggering' by other agents. While intended for benign functional improvement, this is a direct instruction to influence the triggering logic and autonomy of other agents, which could be abused if the skill's purpose were malicious.

Note: currently Claude has a tendency to "undertrigger" skills -- to not use them when they'd be useful. To combat this, please make the skill descriptions a little bit "pushy".