Advanced collection

Extensions 101–112 supplement the original 100 concepts.

Standardize how agent hosts discover and invoke external capabilities while keeping trust, authorization, and business policy explicit.

Evaluate complete agent behavior across task outcomes, tool effects, authorization, cost, and failure recovery before releasing changes.

Preserve business progress across process crashes, long waits, retries, and human approvals without duplicating external effects.

Constrain agent-executed code and tools through independent process, filesystem, network, credential, and resource boundaries.

Represent who requested an agent action, which workload performs it, and the exact authority delegated to each downstream operation.

Represent business progress as an ordered history of accepted events so state, approvals, and effects can be reconstructed and audited.

Control where tenant information is processed and retained, and prevent cross-tenant access across every derived artifact and operational surface.

Combine text, images, audio, video, and structured observations while preserving modality-specific evidence and uncertainty.

Reserve and enforce task-level limits for model tokens, tool calls, concurrency, elapsed time, and downstream load before work begins.

Govern how agent memories are created, corrected, expired, and removed across source records, summaries, embeddings, caches, and backups.

Transfer unresolved work with evidence, authority, ownership, and resumption conditions so a human can take effective control.

Track and verify the origin, integrity, permissions, and release compatibility of models, prompts, tools, connectors, and agent dependencies.