Claude Code 2.1.318 defaults new managed agents to auto permission policy
Anthropic's changelog says the managed-agents-onboard command now gives new agents' built-in tools the auto permission policy by default, with an opt-out.
Claude Code 2.1.318 changes the permission policy a new managed agent starts with. The changelog, dated November 4, 2026, says the /claude-api managed-agents-onboard command now gives a new agent's built-in tools, and a quickstart template's MCP tools, the auto permission policy by default. Anthropic labels that policy "Judge risky actions" in the entry. The change is listed under Changed.
The changelog does not describe the review mechanism, so the entry does not say how a call is classified as risky. It does say the new default matches the Claude Console. A user who wants no approval prompts can say "run without asking" to opt out. That phrasing is the documented route to the opt-out, but the entry does not say whether it works per agent or per session.
The entry is one line in a long release. A separate change to the same command's templates appears in version 2.1.319, not 2.1.318. That release says the templates were updated to match the Claude Console: a Performance tuner template was removed, and web search and fetch were turned off in five templates whose prompts don't use them. Those are template edits. The 2.1.318 entry is the one that changes permission defaults.
What changed
New managed agents created by the managed-agents-onboard command now get the auto permission policy, labelled 'Judge risky actions', by default for built-in tools and quickstart template MCP tools, with an opt-out.
Type: Breaking change · Applies to: Claude Code 2.1.318, /claude-api managed-agents-onboard, auto permission policy
Why it matters
Teams onboarding new managed agents will have risky actions judged automatically unless they opt out, so the permission default now matters during setup.
What you can do
Check the permission policy on any agents created with the onboarding command before they run tools against real systems.
Original source
Where we saw it
Claude Code changelog
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works