Category 06 of 12
Agents & Cognition
The Runtime Abstraction Layer, affordance-based capability routing, model routing, the agentic tool loop, specialists, swarms, and delegation.
Pages in this category
How AdapterManager drives Claude Code, Codex, Cursor, Hermes, Antigravity, and HTTP runtimes through one normalized task/result contract.
Every workspace agent is ranked ready, offline-capable, enablable, or incapable against a node's declared requirements — never a hardcoded agent name.
How each task is classified and routed to the cheapest sufficient model tier, unless an explicit pin or override wins.
Sessions persist across turns, but the prompt is reconstructed each time — persona, task, plan, and fresh observations reassembled, not resent stale.
Inside a run, an agent node wields its full Agentis toolbox — role manifest, universal floor, bridged MCP tools, and the agentis.* surface — as one streamed activity feed.
Every settled tool result carries a compass.next suggestion — guidance that lives in results, not just in prompts.
Bypass mode: auto-approval, a root-power briefing, and env-fix resume that re-runs the unchanged graph after fixing missing config.
Platform, custom, generated, and community specialist roles — and why a specialist is a role, not a fixed agent.
Curated facts, preferences, rules, and vision-derived design patterns distilled into retrievable atoms the engine injects at dispatch.
How a subdomain of work gets deterministically mapped to the specialist accountable for it, with every hand-off recorded and auditable.
Fan a task out to worker sessions, each in its own worktree, then merge — the two cooperation shapes and when to reach for each.
How an agent hands a bounded task to a resolved teammate and awaits the result, with budget and attribution threaded through so no spend is anonymous.
The autonomous orchestrator's command-and-control loop and liveness heartbeat.