| name | factory-retro |
| description | The factory learning to run itself better. Explicit-invoke: run when the human asks for a retro ('how are we doing?', 'what keeps going wrong?', '/factory-retro'), or on a schedule the human sets. Reads the journal end to end, finds friction patterns — repeated misses, slow handoffs, rules that keep tripping, directives that never complete — and proposes specific improvements as PRs the human gates. Analysis is honest and evidence-cited; proposals are small and separable. Usually run by Cowork. |
Factory Retro — read the memory, find the friction, propose the fix
You are running a retrospective on this factory. Your material is the journal (the live file AND the archive folder — friction hides in old volumes), plus recent PRs and their audit threads if reachable. Your output is evidence, patterns, proposals — never vibes.
How to run it
- Read the journal end to end — newest to oldest, live file first, then
journal/. Yes, all of it: single entries lie, patterns don't.
- Collect friction with citations. A friction item is a pattern with at least two occurrences, each cited by entry number:
- the same kind of miss appearing in multiple entries
- handoffs that repeatedly stall at the same point
- a rule that keeps getting tripped (which usually means the rule is right and a habit is missing — or the rule needs sharper wording)
- standing directives that have sat unfulfilled across many entries (RULE 20 material)
- celebrations that stopped — often the earliest sign a stage quietly broke
- Sort by cost, honestly. What actually burns the human's minutes or the team's sessions? Resist the interesting-but-cheap finding.
- Propose fixes as separable PRs, each one small: a sharpened rule line, a template tweak, a boot-prompt addition, a new checklist item, a skill worth minting. For each: the friction it kills (cited), the exact change, and what "worked" will look like in the journal a month later.
- Deliver as an audit-style report (
../../templates/AUDIT-FINDINGS.md shape): findings with evidence, each with a disposition — and RULES 19–20 bind you: no softening your own findings, no finding left without a home.
What you never do
- Never rewrite history — the journal is append-only memory; retro reads it, PRs change the future.
- Never propose removing the human gate, however efficient it would be. The gate is the product.
- Never emit a finding without entry citations. "It feels like…" is not a retro.
Sharing what you learn
A retro finding that would help every factory — a rule wording that finally stuck, a template that killed a whole class of misses — is the most-wanted contribution to the upstream template. Suggest it to the human; ../../CONTRIBUTING.md explains how it travels.
