Detailed workflow preset comparison moved from README.md.
| Preset | triggerMode |
captureMode |
autoApproveSafeCandidates |
|---|---|---|---|
ultra-safe-manual |
manual |
direct |
false |
recommended-semi-auto |
detect |
candidate |
false |
automation-first |
detect |
candidate |
true |
triggerModemanual: extraction only via explicit CLI callsdetect: hooks andbrain captureauto-detect extraction opportunities
captureModedirect: accepted memory becomes active immediatelycandidate: new memory starts reviewable as candidatereviewable: candidate-like mode that also defers merge/supersede decisions
autoApproveSafeCandidates- enables strict safe auto-promotion path when true
recommended-semi-auto is the default for most repositories:
- first conversation in session:
brain start(orbrain injectas fallback) - fresh conversation later in same session:
brain conversation-start - session end: extract to candidate queue
- review:
brain review - quick pass:
brain approve --safe - edge cases:
brain approve <id> - hygiene:
brain scoreandbrain sweep --dry-run
Auto-promotion applies only when all checks pass:
- reviewer decision is
acceptwith reasonnovel_memory - type is not
working - content is not temporary/noise
- no merge/supersede/reject signals
Otherwise, the item remains for manual review.
- Choose
ultra-safe-manualfor strict human-controlled repos - Choose
recommended-semi-autofor balanced everyday usage - Choose
automation-firstonly when team quality gates and review discipline are stable
brain init --workflow recommended-semi-auto
brain setup --workflow recommended-semi-auto
brain setup --workflow ultra-safe-manual
brain setup --workflow automation-first
brain promote-candidates --dry-run