[BUG] Repeated malformed tool-call markup (stray count before <invoke>) within one session — commands silently dropped

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

▎ What happened: In a single long interactive session, the assistant repeatedly emitted malformed function/tool-call markup — a stray count token appeared right before the <invoke ...> opening tag (e.g. count<invoke name="Read">…). Each time,
▎ the call failed to parse ("Your tool call was malformed and could not be parsed. Please retry.") and the intended command (Bash / Read / Agent) never executed.
▎
▎ Frequency: ~6 times in one session, across multiple tool types (Bash, Read, Agent). Never observed before in normal use.
▎
▎ Observed pattern: Failures clustered on tool calls preceded by a longer free-text preamble; retries with a shorter preamble sometimes succeeded.
▎
▎ Impact: Wasted tokens/quota, broke a multi-step workflow, forced repeated retries.
▎
▎ Environment: Claude Code (CLI) · model: Claude Opus · macOS (Apple Silicon).
▎
▎ Expected: Tool-call markup should be well-formed every time; this malformation should not recur within a session.

What Should Happen?

What happened: In a single long interactive session, the assistant repeatedly emitted malformed function/tool-call markup — a stray count token appeared right before the <invoke ...> opening tag (e.g. count<invoke name="Read">…). Each time,
▎ the call failed to parse ("Your tool call was malformed and could not be parsed. Please retry.") and the intended command (Bash / Read / Agent) never executed.
▎
▎ Frequency: ~6 times in one session, across multiple tool types (Bash, Read, Agent). Never observed before in normal use.
▎
▎ Observed pattern: Failures clustered on tool calls preceded by a longer free-text preamble; retries with a shorter preamble sometimes succeeded.
▎
▎ Impact: Wasted tokens/quota, broke a multi-step workflow, forced repeated retries.
▎
▎ Environment: Claude Code (CLI) · model: Claude Opus · macOS (Apple Silicon).
▎
▎ Expected: Tool-call markup should be well-formed every time; this malformation should not recur within a session.

Error Messages/Logs

Steps to Reproduce

What happened: In a single long interactive session, the assistant repeatedly emitted malformed function/tool-call markup — a stray count token appeared right before the <invoke ...> opening tag (e.g. count<invoke name="Read">…). Each time,
▎ the call failed to parse ("Your tool call was malformed and could not be parsed. Please retry.") and the intended command (Bash / Read / Agent) never executed.
▎
▎ Frequency: ~6 times in one session, across multiple tool types (Bash, Read, Agent). Never observed before in normal use.
▎
▎ Observed pattern: Failures clustered on tool calls preceded by a longer free-text preamble; retries with a shorter preamble sometimes succeeded.
▎
▎ Impact: Wasted tokens/quota, broke a multi-step workflow, forced repeated retries.
▎
▎ Environment: Claude Code (CLI) · model: Claude Opus · macOS (Apple Silicon).
▎
▎ Expected: Tool-call markup should be well-formed every time; this malformation should not recur within a session.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.146

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_