Claude Code release notes: 2.1.216 resumes the right agent

Claude Code release notes: 2.1.216 resumes the right agent

Parking agent work is useful only if the same agent comes back. Claude Code 2.1.216 fixes a bug where resumed background sessions could revert to the default agent. The agent's own prompt and tool restrictions are now restored. The release also fixes isolation for parallel Git checkouts and does a better job of preserving human answers when a session has been idle.

Source: Claude Code changelog for 2.1.216.

Claude Code 2.1.216 keeps the role and tool boundaries

A background agent is an independent Claude Code session that can keep working while you do something else. A subagent is a specialized assistant inside one session, with its own prompt and limited tool access.

Version 2.1.216 fixes two concrete problems in these workflows:

  • A resumed background session gets the intended agent's prompt and tool restrictions instead of the default agent's.
  • A subagent in an isolated Git checkout can no longer redirect Git to the shared project checkout through git -C, --git-dir, or the GIT_DIR and GIT_WORK_TREE environment variables.

A Git worktree is a separate checkout on its own branch. It lets several Claude Code sessions work in parallel without colliding with each other's changes.

Sources: Claude Code 2.1.216 on GitHub. Anthropic's guide to parallel sessions and worktrees.

Human answers now survive the pause

The same release fixes two breaks in the conversation. Free-text answers to AskUserQuestion now receive neutral wording, so an answer asking Claude to wait or explain is not treated as permission to continue. Claude Code on the web should also stop repeating the same question or dropping the answer after a few minutes of inactivity.

Long sessions also get a performance fix for normalization work that previously grew quadratically with the number of turns. That can reduce slow resumes in conversations that span several work sessions.

Source: Official release notes for Claude Code 2.1.216.

What the change means for a real Claude workflow

Consider an analysis agent with read access and an implementation agent with scoped write access. When work pauses, the analysis agent should not wake up as a broad default agent, and the implementation agent should not land in the wrong checkout. Version 2.1.216 removes known failure modes that could break that handoff.

This week's test is simple: resume a real background task and check the agent definition, allowed tools, current worktree, and any questions that still need a human decision. In integrated workflows, keep secrets in environment variables or a secret manager, then control writes with scoped permissions, approval gates, and a short run log.

Source: Anthropic's documentation on subagents and tool boundaries.

Try this prompt this week

Human step: Update through your approved distribution path, run claude --version to confirm that you are on 2.1.216, and resume a background task that already has its own agent definition and worktree.

Source: Version check in the Claude Code changelog.

Read CLAUDE.md, the relevant agent definition, and the current Git status.
Identify the role, tool access, and worktree this task is meant to use.
Compare that with the session and checkout you are working in now.
Run only existing read-only checks that support the comparison.
Report mismatches, pending questions, and decisions that need approval.
Make no changes. Finish with a short resume receipt.

Check three things in the response:

  • The agent definition and current checkout are named.
  • Verified facts are clearly separated from assumptions.
  • An unresolved question or unexpected write permission stops the workflow.

Put the review at the restart point

Pause before the next write or pull request, compare the resume receipt with the intended agent role, and continue once the checkout and permissions match. This is practical Tool Forge work: making a Claude workflow easy to resume without losing speed or control.

The npm registry confirms that 2.1.216 is the current published package version.

Source: Claude Code 2.1.216 in the npm registry.

FAQ

What changes for resumed background agents in Claude Code 2.1.216?

The resumed session restores the intended agent's prompt and tool restrictions instead of falling back to the default agent.

What is the difference between a background agent and a subagent in Claude Code?

A background agent is a separate session monitored in Agent View. A subagent works inside an existing session with its own context, prompt, and tool access.

Does Claude Code 2.1.216 fix anything beyond session resume?

Yes. The release also fixes dropped answers in Claude Code on the web, misleading wording after free-text AskUserQuestion responses, and a long-session performance slowdown.

The Forge newsletter

Get new articles in your inbox

Pick the topics you care about. No noise, at most one email a week.

Get new articles in your inbox

We follow GDPR. Unsubscribe anytime.