Adds a user-invoked `wtf` skill for when a reply, plan, or document did
not make sense. With no argument it explains the agent's last message;
with a file, link, passage, or pointer it explains that. It explains
rather than rewrites, keeps caveats and failures, and adds no claims.
Manual-only on every host: `disable-model-invocation: true` plus a Codex
`agents/openai.yaml` with `allow_implicit_invocation: false`. The name is
exempt from the ce- prefix because `/wtf` is what a confused user types.
Trevin Chow
91de6e84ee