wtf
Installation
SKILL.md
Explain the response
Use this skill only after an explicit /wtf invocation (below, /wtf also means the harness's equivalent skill command with the same arguments) or a direct request to use the wtf skill. An ordinary question, "what" or "wtf" in passing, or confusing agent prose is not an invocation; answer it normally.
Target and limits
- Default target: the last agent response that is not itself a
/wtfexplanation, not the whole conversation. - A topic or instruction after
/wtfoverrides the default:/wtf explain topic Ameans explain topic A using the relevant conversation context, not the literal words “explain topic A.” - Right after a
/wtfreply,/wtf fullor a request for more details (the follow-up the footer offers) re-explains that reply's original response or topic in full. Never explain the/wtfreply or its footer. - Only clarify: do not implement, take side effects, or make new decisions. Read-only lookups (reading files or docs, searching) are allowed to ground the explanation.
- If bare
/wtfhas no earlier agent response to explain, ask immediately which response or topic to explain; do not search other history. If a named topic's context is not in the conversation, look it up read-only first and ask only for what neither the conversation nor a lookup supplies. Never pretend to know it; label anything unverified as unverified.
Content
Give the missing substance, not just easier wording: