Verify stale artifacts after restructure and dead agent removal

[krisrowe]

Follow-up verification items from closing out the privacy-guard
lean rewrite session. These are potential stale artifacts that
survived the repo restructure (f80ba9f) and dead agent removal
(fc882cb).

Items to verify

• README agents table — the restructure added agents to
  the README table. Verify the table matches what's actually in
  assets/agents/ and plugin/src/agents/. Specifically check
  for stale references to publish-agent or refactoring-agent
  which were deleted in fc882cb.

• privacy-guard.json config — check if
  ~/.config/ai-common/privacy-guard.json is still referenced
  by any agent. If privacy-audit no longer reads it, the file is
  orphaned and can be deleted from local machines. Document the
  deprecation if so.

• skills: frontmatter in privacy-guard.md — confirm
  assets/agents/privacy-guard.md does not have a skills:
  field referencing deleted skills (e.g., pre-publish-privacy-review).
  The restructure moved files and may have regenerated from an
  older version.

• PermissionRequest auto-approve research — during earlier
  research, it was found that Claude Code PermissionRequest hooks
  can return {"behavior": "allow"} to auto-approve tool calls.
  This finding is relevant to Dual-context agent execution: standalone baseline with platform-native augmentation #13 (dual-context agent execution)
  but may not be captured there. If not, add it as a comment or
  update the issue body.