What problem does it solve?
ACA helps prevent AI agents from gradually drifting, losing their intended identity, constraints, and judgment over long sessions as context becomes noisy.
Core Features & Use Cases
- Drift corridors: structured “will-not” failure modes with explicit countermeasures that make cheap default completions structurally expensive.
- Abstract-to-deterministic cognitive layers: a file-based stack (SOUL/JOB/MOTIVATIONS/CHARTER/COMMS/IDENTITY) that preserves intent while remaining precise where it matters.
- Multi-agent differentiation support: uses motivations and chartered tensions so multiple agents on the same model don’t converge into echo-like consensus.
- Operational setup guidance: teaches operators how to strengthen SOUL.md first and then add the full stack via a bootstrap hook.
Example use case: A team of agents reviews a deployment spec; drift corridors and deterministic “will-nots” keep security, rollback planning, and risk naming consistent even as failures accumulate in context.
Quick Start
Tell your operator to strengthen your SOUL.md by adding specific drift corridors (e.g., “You don't approve changes without naming the residual risk”) and concrete will-nots.