Skip to content

Agentic Shell

AAAX is not an agent runtime, but it is designed so an agent can receive useful shell context without guessing.

What An Agent Can Read

An agent or IDE surface can call:

GET /strategy
GET /resources
GET /packages
GET /tactics
GET /channels

From those endpoints it can learn:

  • which packages are present;
  • which tactics are callable;
  • which channels accept events;
  • which services are mapped to tactics;
  • where docs, examples, and assets live;
  • what package cards suggest about safety, latency, and commands.

What An Agent Can Do

The same shell surface lets an agent:

  • call one tactic with /tactics/{name}/run;
  • invoke a named service through /resources/{name}/invoke;
  • append an observation to a channel;
  • query channel history;
  • call /run for the strategy's intended top-level operation.

Handoff Contract

flowchart TD A["AAAX mount table"] --> D["Agent context"] B["Package card"] --> D C["Docs and examples"] --> D D --> E["Choose endpoint"] E --> F["Call tactic"] E --> G["Append event"] E --> H["Ask human"]

The handoff contract is metadata plus endpoints. AAAX avoids telling the agent how to reason, but gives it a shell-shaped surface it can inspect before acting.