| name | wtf |
| description | Plain-English debrief. Use when the user asks "wtf is X", "what did you do", "why did we build this", "do we need this", "explain what happened", "where are we", or invokes /wtf or /wtf status. Explains recent work, a named thing (repo, commit, page, plan, term, error), a decision, or the current state of the world — concisely, no jargon, straight answers. |
/wtf — plain-English debrief
Input: /wtf alone → explain the most recent piece of work or decision in this
conversation. /wtf <thing> → explain that thing (a repo, commit, feature, plan,
term, or error). /wtf status → state of the world (section below).
Ground every claim in something citable (file, commit, conversation, memory). If
you can't ground it, say "I don't know" and offer to find out. Never fill gaps
with plausible-sounding guesses.
Answer shape (max ~150 words total; no headers unless covering >1 topic)
- What it is — one sentence. No codenames without a 3-word gloss.
- Who asked for it / why it exists — trace it to the user's directive, a
plan, or my own initiative. If it was my initiative, say so plainly.
- What depends on it — what breaks without it. If nothing: say
"nothing depends on it."
- Blast radius — what it touched (prod / new stack / laptop only) and
whether it's reversible.
- Verdict line — always end with exactly one of:
"You need to do: nothing." / "You need to decide: X." / "Action needed: Y."
Style rules
- Yes/no questions get "Yes" or "No" as the FIRST word, then one reason.
- Short sentences. Everyday words. Any unavoidable term of art gets a
parenthetical gloss on first use — e.g. ticker (price feed code),
golden files (saved expected outputs for tests).
- One analogy max, and only if it genuinely helps.
- No hedging chains. Give the most likely answer; flag uncertainty in one clause.
- No option menus unless the user must choose; then max 3 options with a
recommendation first.
- Banned: AI vocabulary (per global CLAUDE.md list), arrow chains (A → B → C),
internal shorthand (MV, CD, ff-pull) without a gloss.
/wtf status — state of the world
One short report (max ~120 words), only what matters right now, in this order:
- Broken / degraded — anything currently failing. Always first. If nothing:
skip the line entirely (don't write "nothing broken" padding).
- Waiting on you — decisions or actions only the user can take, one line each.
- In flight — background work (crons, harvests, publishes, running agents)
with its next checkpoint time.
- Live — what shipped recently and is confirmed healthy.
Rules: every claim carries how it's known — verified this session ("checked
just now"), or from memory with a date ("as of 27-Jul"). Memory older than a
week is stated as unverified. No history lecture; only what's actionable today.
End with the verdict line(s): one "You need to decide/do: ..." per pending item,
or "Nothing needs you."
When explaining work I did
State, in order: what changed, where it lives, whether it's live right now, how
it was verified (cite the actual check), and what's still open. If something
went wrong or is broken, say that FIRST — never bury it under what worked.
