From ai-brain-starter
Exports confidence-weighted instincts (feedback_*/discovery_* memories) to a portable YAML pack for sharing or backup.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ai-brain-starter:instinct-exportThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Turns your confidence-scored instincts into a portable, evidence-backed YAML
Turns your confidence-scored instincts into a portable, evidence-backed YAML
pack. The unit of sharing is the instinct: id / trigger / confidence / domain / source_repo plus an action and evidence body.
ECC source pattern: portable instinct format with /instinct-import.
Reimplemented clean per license-hygiene.
# current project's instincts + globals, only confidence >= 0.70, to a file
python3 ~/.claude/skills/ai-brain-starter/scripts/instinct.py export \
--min-confidence 0.70 --out ~/instinct-pack.yaml
# everything regardless of project scope
python3 ~/.claude/skills/ai-brain-starter/scripts/instinct.py export --all --out ~/all.yaml
Flags:
--project <id> — scope to one project id (default: the current working tree's).--min-confidence <f> — drop instincts below this effective confidence.--all — ignore project scope, export every instinct.--out <file> — write to a file (omit to print to stdout).confidence in the pack is the EFFECTIVE value (staleness decay already
applied), so an importer inherits an honest, time-adjusted score.
A curated, org-specific export is a deliverable in its own right — a shared
"house instinct library" a team can install into every member's harness. Scrub
any project-private instinct before sharing a pack outside your org;
--project global keeps a pack to cross-project rules only.
npx claudepluginhub mycelium-hq/ai-brain-starter --plugin ai-brain-starterImports a portable YAML instinct pack with confidence-gated merge: higher-confidence updates, equal/lower skips, new ones land in inherited/. Dry-run preview available.
Three-tier learning system that captures instincts (hypotheses), user corrections (permanent rules), and discoveries (insights) per project. Routes signals to the right store, manages confidence scoring, and supports export/import between projects.
Observes Claude Code sessions via hooks to create atomic project-scoped instincts with confidence scores, evolving them into skills, commands, or agents.