Agent Systems
Harness Engineering & Sandboxes
Reproducible environments, structured tools, isolation, traces, and evaluation hooks.
10 chapters · 5-7 hours reading · 3-4 hours labs · 6-10 hours project
Chapters
- 1The harness as an evidence-producing boundaryMental model
- 2Reproducible fixtures and environment identityReproducibility
- 3Typed tool contracts, authorization, and bounded observationsInterfaces
- 4Containers, application kernels, and microVM isolationIsolation
- 5Runner architecture and lifecycle state machineArchitecture
- 6Traces, event records, and replay semanticsObservability
- 7Budgets, termination, idempotency, and recoveryControl
- 8Fault injection and adversarial resilience experimentsFailure engineering
- 9Evaluation hooks and evidence-based verdictsEvaluation
- 10Operating a harness as security-critical infrastructureProduction