Release archive

Claude Code v2.1.283: Auto Mode Default on Third-Party Providers, New Model Controls, and /doctor prompt-audit

Claude Code v2.1.283 starts sessions in auto mode by default on third-party providers or with telemetry off, adds availableModelsMatch and deniedModels managed settings, and adds /doctor prompt-audit.

September 25, 2026 • 10 min read • By ClaudeWorld

Claude Code v2.1.283 is a large release. Its headline changes are a new default permission mode for interactive sessions on third-party providers or with telemetry off, two managed settings that give admins tighter control over which models can be used, and a /doctor prompt-audit check for prompting patterns written for older models. It also reverts the claude-ai name reservation introduced in 2.1.282 and brings fixes for MCP, plugins, startup latency, VS Code and cloud sessions.

What’s New

Auto mode by default on third-party providers or with telemetry off

Interactive sessions on third-party providers, or with telemetry off, now start in auto mode when no permission mode is configured. permissions.defaultMode still overrides it.

Model controls: availableModelsMatch and deniedModels

With the new availableModelsMatch managed setting set to "exact", an availableModels entry allows only the model version it names, so new releases stay blocked until they are listed. The new deniedModels managed setting blocks specific models even when availableModels allows them.

/doctor prompt-audit

/doctor prompt-audit (also /checkup prompt-audit) audits your CLAUDE.md files, skills, agents and commands for prompting patterns written for older models. On Claude Code configuration, its report leads with stale paths, stale commands and contradicting instruction files, and it keeps thinking keywords that Claude Code documents.

claude-ai name reservation reverted

The 2.1.282 reservation of the claude-ai name is reverted: skills, commands, workflows and MCP servers’ skills and prompts with that name load again, and Skill(claude-ai:*) rules are ordinary prefix rules. Separately, Skill(anthropic-skills:<name>) deny rules now also block that skill when Claude Desktop delivers it as a plugin, and Skill(skill:<name>) denies match the skill’s alias and display name.

Gateway and telemetry additions

Gateway hint headers now include x-claude-code-prompt-id, so LLM gateways can group the requests that serve one user prompt; opt in with CLAUDE_CODE_GATEWAY_HINT_HEADERS=1. The Claude apps gateway gains an opt-in load_test_mode block, which builds and signs requests without sending them upstream and returns a canned reply so a deployment can be load tested, plus a mantle upstream provider for Amazon Bedrock’s Mantle endpoint. With OTEL_LOG_TOOL_CONTENT=1, the tool.output OpenTelemetry span event now also carries MCP tool, WebFetch and WebSearch outputs.

System prompt text and files together

--system-prompt and --append-system-prompt now accept their text and -file forms together; the file’s text comes first.

Fixes and Improvements

  • Windows: The PowerShell tool no longer lets cmd /c rd, rmdir, del or erase delete drive roots, the home folder and other folders that Remove-Item refuses.
  • MCP: Progress notifications are no longer discarded once a long-running MCP tool call moves to the background; the background task now shows the latest progress.
  • MCP: A brief HTTP 404 from a stateless remote MCP server (for example a proxy mid-redeploy) no longer leaves that server unusable for the rest of the session while it is still shown as connected, and stdio MCP servers are no longer left running when the session ends while they are still starting.
  • MCP: Images returned by MCP tools are now also saved to a file, so Bash, Read and other tools can open them, and /context now counts MCP server instructions as their own row.
  • MCP: claude mcp add, add-json and remove no longer report success when the user or local config file could not be written, for example inside a sandbox.
  • Models: The /model picker no longer shows a hardcoded Haiku version and price when ANTHROPIC_DEFAULT_HAIKU_MODEL pins a different model, and DISABLE_PROMPT_CACHING_HAIKU now takes effect when Haiku is the session’s main model.
  • Models: The /model picker’s Opus row and the Default model’s name drop “(1M context)” where Opus already has a 1M context window; the window itself is unchanged.
  • Plugins: claude plugin validate now fails plugin or marketplace names that Claude Code cannot install, and plugins whose outputStyles, themes, monitors or lspServers paths are missing or point outside the plugin directory.
  • Plugins: claude plugin uninstall no longer removes the other of two installed plugins whose ids differ only in case, and installed_plugins.json is no longer rewritten, losing records, when it holds a record this version cannot read.
  • Plugins: User-installed plugins and marketplaces no longer fail to load with “cache-miss” after the home or config directory was moved, for example in bind-mounted devcontainers.
  • Sandbox: Managed sandbox settings are no longer ignored entirely when one nested value is invalid; the invalid value now fails closed and the rest of the block still applies.
  • Remote Control: Remote Control is now available on paid plans when telemetry is turned off with DISABLE_TELEMETRY or DO_NOT_TRACK.
  • Performance: First-reply and first-request latency are improved, and on startup claude -p and Claude Code Remote no longer load the interactive UI, while the auto-mode classifier’s rules and the Artifact tool load on first use instead of at launch.
  • Lists: /help, /hooks, /copy, /memory, /rewind, /diff, /plugin and other pickers gain page keys, mouse wheel and clicks, and /tasks rows now show a status icon, the name and whole facts.
  • Prompt suggestions: Suggestions in the terminal now appear less often after 20 in a row go unused; using one brings them back.
  • Self-hosted runner: Lifecycle hooks’ git now skips a repository’s Git LFS pre-push hook, ignores a writable system core.hooksPath, and does not sign commits without --configure-git. Handling of GIT_SSL_CAINFO and GIT_SSL_NO_VERIFY under Anthropic-managed git also changed: the runner’s own git always verifies Anthropic’s git route, and warning lines say what applies where.
  • VS Code: The permission mode indicator no longer shows Default while the session keeps running in auto or bypass mode after an automatic switch out of it failed (the switch is now retried until it lands), and a session teleported from the web no longer drops messages sent while Claude was working.
  • Cloud sessions: Cloud sessions no longer occasionally redo an already-finished step, such as posting a duplicate comment or push, after recovering from a server-side restart.
  • Claude Tag: A new “Channels Claude can search” admin setting limits Claude’s Slack search to public channels it has been added to, set per organization, workspace or channel.
  • Code Review: A review that stopped at its time limit with nothing verified now shows as incomplete, isn’t charged, and is retried once.

This release contains 94 changes in total; see the full changelog for the complete list.

How to Update

  1. Check your installed version: claude --version
  2. Update in place: claude update
  3. If you use a third-party provider or run with telemetry off and have not configured a permission mode, interactive sessions now start in auto mode; set permissions.defaultMode if you want a different default.