Chapter 14 of 15

Chapter 14 of 15

Chapter 14: Configuration Reference

Outline

14.1 Why Configuration Belongs in a Security Book

  • Configuration is not deployment trivia; it is executable security policy
  • OpenClaw behavior changes materially with routes, tool exposure, and approval rules
  • The appendix exists to make operational truth legible

14.2 Source Hierarchy and Precedence

  • Versioned defaults, deployment overlays, workspace instructions, runtime settings, and task-scoped overrides
  • Why precedence must be explicit rather than guessed
  • Keeping live runtime state separate from static configuration

14.3 Model and Provider Routing

  • Local configuration as the only trustworthy statement of model availability
  • Provider allowlists, route selection, and capability-specific model choices
  • Avoiding “documentation says it exists” authority drift

14.4 Tool Exposure and Capability Policy

  • Tool availability, elevated actions, approval requirements, and sandbox choices
  • Per-tool boundaries as part of configuration, not only runtime behavior
  • Why hidden defaults are dangerous

14.5 Agent Instructions, Memory, and Workspace Context

  • System persona, workspace guidance, human notes, and durable memory serving different roles
  • Separating behavioral guidance from operational truth
  • Preventing prompt context from becoming accidental authority

14.6 Secrets, Environment Variables, and External Endpoints

  • Secrets as separate security subjects, not prompt material
  • Endpoint configuration, service URLs, and provider credentials
  • Rotation, scoping, and visibility discipline

14.7 Runtime Overrides, Sessions, and Delegated Work

  • Session-level overrides, subagent scope, thread binding, and cron parameters
  • When narrower overrides are helpful and when they become governance drift
  • Carrying policy without smearing authority across runtimes

14.8 Validation, Drift Detection, and Change Control

  • Verifying live schema, route ownership, health, and dependency readiness before acting
  • Treating configuration changes as reviewed artifacts
  • Detecting stale assumptions before they become incidents

14.9 Common Configuration Anti-Patterns

  • Ambient credentials, undocumented overrides, duplicate ingress ownership, and implied availability
  • Why “temporary” config often becomes production truth
  • Recovery patterns that restore clarity

14.10 Reference Checklist

  • The minimum questions operators should answer before trusting a deployment
  • Concise closing guidance for future maintenance

Chapters 4 to 15

Confirm your email to keep reading

This chapter is part of the full book. Enter your email and open the confirmation link to unlock all 15 chapters in the reader, plus a Markdown download of the whole book. Chapters 1 to 3 stay free.

We'll email a confirmation link that unlocks all 15 chapters, plus occasional AI Deploy product updates. You can unsubscribe anytime.

Privacy

Chapter 14: Configuration Reference | AI Agent Harness Book