theorydeltaclaim-fidelity audits
built 2026-07-17dossiers: 5last verified 2026-07-09independent · evidence-traced · no vendor influence
← back to Claude Code dossier

CLAIM 7 / Claude Code

"Each subagent runs in its own context window with a custom system prompt, specific tool access, and independent permissions"

as labeled

Observed

Confirmed: per-subagent context isolation, tool allowlists and denylists (tools, disallowedTools frontmatter), per-agent model and permissionMode all behave as documented (backing block re-validated against docs and releases 2026-06-08). The flip side of "independent permissions" — non-inheritance of parent deny rules — is scored at row 1.

Evidence

theorydelta.com · 2026independent · evidence-backed · every claim sourced or labelledabout ·glossary ·rss ·mcp ·llms.txt