bro wtf + retrospect
Capture frustration verbatim; turn it into retros and prevention beads.
bro wtf <complaint>
Captures the user's frustration verbatim as a wtf bead — timestamp and
git snapshot included. The complaint is data; the retro comes later.
bro self retrospect
Turns wtf beads into structured outcomes — a retro record plus one
prevention bead per action, linked discovered-from, with the wtf
answered.
| Command | What it does |
|---|---|
bro self retrospect record <plan.toml> | Validate a TOML plan and fan it out: retro bead + prevention beads, wtf answered |
bro self retrospect status | Exit gate — non-zero while a wtf is unanswered |
bro self retrospect schema | Print the commented TOML template |
bro self retrospect list | Retros and open wtfs |
The agent can't self-declare "sorry, fixed" — retrospect status holds
the gate until every wtf has a recorded answer.
The retro-on-stop guard
The wtf path is incident-driven; the session-end retro is every-session.
The retro skill carries the checklist — loose ends, hand-rolled
mechanics that should become beads or bro commands, call economy, bro self learn capture, bro review debt collect, state assertions — and a
guard def fires the one-line nudge at stop:
{
"guard": {
"defs": [
{
"name": "retro-on-stop",
"when": { "on": ["stop"] },
"say": "session ending — run the retro checklist"
}
]
}
}The guard is the doorbell; the skill's checklist is what "ran the retro"
actually means. Scale to the session — a one-shot Q&A skips with a
line; findings land as bd items, never as prose the user must convert.