Home / CCA-F / Claude Code / Architecture and Setup — Detail & Examples
DWG 3.1 — DETAIL & EXAMPLES

Architecture and Setup — Detail & Examples

Part of Claude Code · overview: domain-3.html
In this page

The agentic loop traced against a real debugging task, and a look at why permission scoping in settings.json is a real security boundary, not a convenience toggle.

3.1aThe loop, applied to a dev task

ACTRead the failing test file and the source file it covers.
OBSERVEThe test expects a sorted list; the function returns unsorted.
DECIDEAdd a sort call before return; re-run the test to confirm.
ACTEdit the source file, then run the test suite.
OBSERVETest passes; no other tests broke.

This is the exact act/observe/decide loop from Domain 1, just scoped to a filesystem and shell instead of an abstract task.

3.1bPermissions as a real boundary

settings.json in practice
  • A read operation (viewing a file) is typically safe to auto-allow.
  • A destructive or irreversible shell command (e.g. one that deletes files) is exactly the kind of action that should require explicit approval, not run silently.
  • Getting this scoping wrong in either direction is a real cost: too permissive and a mistaken edit can cause real damage; too restrictive and every trivial action needs a manual approval, defeating the point of automation.

Claude Certified Architect — Study NotesDWG 3.1