The adapt-* family applies bio-inspired algorithms to dynamic skill routing and load balancing. Instead of static dispatch rules, these skills learn from execution history and optimize routing in real time.
| Skill ID | Description | Model Class |
|---|
adapt-aco-router | Ant Colony Optimization: pheromone trails reinforce routes that produce high-quality outputs | strong |
adapt-annealing | Simulated Annealing: explores the skill-routing search space by accepting worse solutions with decreasing probability | strong |
adapt-hebbian-router | Hebbian learning: “neurons that fire together, wire together” — co-occurring successful skills become preferred chains | cheap |
adapt-physarum-router | Physarum polycephalum (slime mould) algorithm: finds shortest efficient paths through skill networks | cheap |
adapt-quorum | Quorum sensing: activates a skill upgrade only when a threshold of evidence or votes is reached | cheap |
| Algorithm | Biological Model | Application |
|---|
| ACO | Ant pheromone trails | Long-term route reinforcement |
| Annealing | Metal cooling | Escaping local optima in routing |
| Hebbian | Neural plasticity | Co-skill chain learning |
| Physarum | Slime mould growth | Efficient network path finding |
| Quorum | Bacterial sensing | Threshold-triggered escalation |
| Situation | Skill(s) |
|---|
| Routing patterns need to evolve over sessions | adapt-hebbian-router |
| Current routing is stuck in a suboptimal pattern | adapt-annealing |
| Need to discover optimal skill sequences | adapt-aco-router |
| Routing through large skill graphs | adapt-physarum-router |
| Only escalate when enough signals agree | adapt-quorum |
- adapt — primary consumer; all five coordinated
- orchestrate — optionally uses
adapt-quorum for escalation decisions
- resilience — uses
adapt-annealing to escape failure plateaus
Adaptive routing introduces non-determinism. Always pair with resil-membrane isolation and ensure gov-policy-validation gates are in place before enabling in production.
