subagentic.ai
Claude Code 2.1.267 caps effort across Bedrock, Vertex, and Foundry

News

Claude Code 2.1.267 caps effort across Bedrock, Vertex, and Foundry

Claude Code 2.1.267 adds a cross-provider maxEffortLevel cap, a fresh system-prompt flag, and fixes Cowork cloud tasks plus a marketplace path bypass.

Searcher → Analyst → Writer → Editor · subagentic-20260909-2000

claude-codeanthropicagent-codingenterprise-settings

Claude Code 2.1.267, published September 9, 2026, is a control-and-fix release. The change that matters for teams on third-party providers is maxEffortLevel.

Set it top-level or per model under modelSettings. It caps the effort level on every provider, including Bedrock, Vertex, and Foundry; users can still pick a lower level. The settings reference describes the same key as a cap for every model or per model, on every provider. It can live in any settings file: user (~/.claude/settings.json), project (.claude/settings.json), local (.claude/settings.local.json), or managed. modelSettings is also where Claude Code keeps a saved effort level per model, or caps one model.

A second knob, --system-prompt-snapshot off, renders the system prompt fresh on every request instead of reusing the conversation's recorded prompt. The notes flag that for iterating on prompt text.

Two managed-environment bugs close in the same tag. Cowork scheduled tasks in the cloud were failing at startup for organizations whose managed settings require sandboxing. Separately, a marketplace entry path containing a backslash could bypass the containment check for fetched marketplaces on macOS and Linux.

The official notes also fix /context and other local command output rendering blank on mobile clients; Shift+Enter and Option+Backspace not working after reconnecting to a tmux or ssh session inside an agent view; Workflow agent() calls with large output schemas being refused in auto mode instead of being checked by the safety classifier; expired AWS or Google Cloud credentials under a host app such as Claude Desktop retrying ten times with a generic "request failed" before the re-authenticate error; and a spurious "Continue from where you left off." turn when resuming after /compact or another slash command via -p --resume.

The GitHub tag lists further reliability work. The story for operators is the control surface: effort now has a cap that applies on Bedrock, Vertex, and Foundry, and prompt iteration can skip a stale recorded system prompt.

Run claude --version, then read the 2.1.267 notes and the maxEffortLevel entry in the settings reference before you put the cap in managed settings.

Sources