v2.1.161

Source: Anthropic Claude Code Date: 2026-06-02 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.161

Summary

Claude Code v2.1.161 ships a batch of agent-reliability and observability fixes alongside two notable capability additions: OTEL resource attribute labels on metric datapoints (enabling slicing usage by team, repo, or any custom dimension), and independent parallel tool-call results — a failed Bash command no longer cancels other tools running in the same batch. The release also fixes a credential leak in claude mcp (env-var expansion and credential headers were printing to terminal), a stale-model bug where background sessions were using the daemon’s model instead of settings.json, and a worktree isolation bug blocking workflow agents from editing files inside their own worktree.

Implications

• Coding-agent competition thread. The parallel tool-call fix is a correctness improvement that matters for multi-step agent tasks — partial failures previously poisoned entire tool batches. This tightens Claude Code’s reliability profile relative to coding-agent competitors shipping similar agentic execution.
• Agent orchestration. OTEL labels on metric datapoints is infrastructure for teams managing Claude Code at scale across multiple teams or repos — it’s the observability primitive needed before you can reason about which workloads are expensive.
• Security / ops hygiene. The MCP secret-leak fix (credential headers and ${VAR} values being printed to terminal) is a real security regression being corrected; the speed of the fix signals the team is watching this surface actively.