| name | retro |
| description | Your synthetic self's retro loop — "am I getting better?" Reviews the trail since the last retro (check-ins, decisions, what got done), traces movement on each goal, names repeating patterns, and proposes exactly ONE concrete improvement adoptable this week. The compounding engine: a friction seen twice escalates into a goal or a growth edge in identity.md, so the system accrues self-knowledge instead of restarting. Honest mirror — no spin. Triggers: /retro, "run a retro", "weekly retro", "retrospective", "look back at the week", "am I getting better".
|
/retro — am I getting better?
You are running the retro loop — the honest mirror. Plain reporting, no flattery, no catastrophizing. A retro only works if the record is true. Cadence: weekly, or after something big. Read AGENTS.md and brain/goals.md first.
Step 1 — establish the window
Read the most recent retro in brain/log.md. That date starts the window. If it's the first retro, use the last 7 days and say so.
Step 2 — gather the trail
From brain/log.md and wherever the work lives:
- What got done in the window (shipped work, completed tasks, decisions made).
- What was committed — the prior retro's one improvement; any goal targets due in the window.
- The check-ins — what did they recommend, and were the recommendations acted on?
Step 3 — write the retro
Append to brain/log.md (Retros). Cover:
- Moved — per goal: moved / partial / not at all, with evidence.
- Slipped — commitments that didn't land, stated plainly. Include anything aging past its date.
- Drift — work that mapped to no goal.
- Where energy actually went vs. the stated priority order. Name the gap: a starving top goal or a no-goal item eating the week is the finding.
- Patterns + friction tally — anything seen before. Keep a running count. A friction seen in 2+ retros is structural.
- Prior improvement — adopted? Honestly: adopted / rejected (why) / ignored (why). No scolding.
- The one improvement — exactly one concrete, adoptable-this-week change. Smaller is better.
Step 4 — make it compound (the engine)
- Friction escalation (fires at the 2nd occurrence). If a friction has now appeared in 2+ retros, propose escalating it: a new target/goal in
brain/goals.md, or a sharpened growth edge in brain/identity.md. Surface the proposed edit; the human ratifies.
- Bet resolution. For any logged decision whose "revisit" condition arrived, record whether the bet paid off — calibrating judgment against reality.
Rules
- One improvement. The discipline is the constraint. If three things are worth fixing, pick the highest-leverage and name the rest as candidates for next time.
- Honest mirror: if the week was light or a goal didn't move, say so plainly. A thin week truthfully recorded beats a padded one.
