Skip to content

Online Companion

The online companion holds material that should remain versioned, testable, and independently updateable outside the print manuscript.

The book keeps the argument, decision criteria, and minimal contract forms. This section keeps complete CLI walkthroughs, runtime configurations, trace and event catalogs, evaluation datasets, sources, and reference walkthroughs.

Main routes

  • Runtime configs and MCP boundary
  • Runtime CLI
  • Eval datasets
  • Traces and events
  • Templates
  • Checklists
  • Changelog
  • Errata
  • Example artifacts: artifacts/trace-demo.jsonl, artifacts/trace-failed-tool-timeout.jsonl, artifacts/trace-post-dispatch-timeout.jsonl, artifacts/session-failed-tool-timeout.json, artifacts/eval-failed-run-timeout.json, artifacts/eval-unknown-effect-reconciliation.json
  • Filled examples: examples/capability-contract-support-ticket.md, examples/release-decision-record-support-ticket.md, examples/incident-record-support-ticket-timeout.md, examples/production-readiness-support-ticket.md, examples/context-manifest-support-ticket.yaml, examples/threat-map-negative-tests.yaml, examples/slo-card-support-ticket.yaml, examples/adlc-transition-support-ticket.yaml, examples/readiness-rubric-support-ticket.yaml
  • Complete reference package walkthrough
  • Complete source list

Practice path for a safe agent

Readers who want to reproduce the material should start with the complete reference package walkthrough. It connects chapters to agent_runtime_ref files, CLI commands, companion artifacts, and tests. The shortest practical path is inspect-agent for inventory, simulate-run for a controlled run, dump-events and inspect-trace for evidence, inspect-approvals for the human gate, export-eval-dataset for evaluations, and check-rollout and check-controls for the release decision.

What belongs here

  • Complete YAML configurations and review forms.
  • CLI commands and expected JSON surfaces.
  • Trace and event catalogs plus validation-message catalogs.
  • Evaluation datasets, verifier contracts, and rollout judgment examples.
  • Long source catalogs, changelog, errata, and update rules.

What stays in the book

  • Why an architectural decision is needed.
  • Which risk it controls.
  • Who owns the action and its evidence.
  • How a team knows that runtime, policy, trace, evaluation gate, and rollout are ready.

Chapter map

This map connects every print chapter to the corresponding topic in the online book and the closest executable or reference artifact. Link names matter more than internal site numbering: if the online structure changes, readers can still find the topic and its verifiable artifact instead of guessing which web chapter has the same number.

Print chapter Online topic Practice and reference
1 Why an agent needs a platform Reference package
2 Execution forms, manager and handoffs Practical request patterns
3 Safe agent architecture Technology stack map
4 Trust boundaries Policy templates
5 Policy layer and capability catalog Policy bundle schema
6 Tool gateway and approvals Approval schema
7 Memory risks Memory evaluation patterns
8 Memory types and lifecycle Memory retrieval schema
9 Context retrieval and compaction Continuity envelope
10 Execution model and tools Configuration reference
11 Sandbox and MCP, MCP and A2A boundary Tool failure recovery
12 Retries, limits, and rollback Tool failure recovery
13 Traces and events Trace and event catalog
14 Agent-system SLOs Filled SLO card
15 Evaluations and regression gates Evaluation datasets
16 End-to-end evidence spine Change review and rollout schema
17 Platform and product teams Capability contract template
18 Supported golden paths Filled golden-path contract
19 Agent inventory and registry Registry operations handbook
20 From SDLC to ADLC, change management ADLC transition example
21 Provenance and trusted artifacts Lifecycle artifact schema
22 Observability and detection telemetry Trace schema
23 Goal misalignment and insider risk, behavioral evaluations Negative threat-model scenarios
24 Assurance and response Incident response playbook
25 Decommissioning and replacement Lifecycle artifact schema
26 Baseline runtime CLI reference
27 Policies and capability catalog Executable trajectory-policy scenarios
28 Production launch checklist Production readiness checklist