[BUG] Tool calls repeatedly emitted with bare <invoke> tags (missing antml: namespace), causing parse failures

Open 💬 2 comments Opened Jul 3, 2026 by dahee90522

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?

In the latter half of a session, the assistant's tool calls were repeatedly emitted using bare <invoke> / <parameter> tags instead of the correct antml:invoke / antml:parameter namespaced form.

As a result, the tool calls failed to parse and were silently dropped (surfacing as "?" with no execution), so no work could be performed.

Even after the user pointed this out many times and the assistant acknowledged and promised to fix it, the same malformed-format error recurred 20+ times.

Notably, some tool calls in the same session parsed and executed fine (e.g. git status, eslint), while others in the same context broke — so the failure was intermittent / state-dependent rather than a consistent formatting mistake.

Impact: work was blocked and the session became effectively unusable.

What Should Happen?

All tool calls should be emitted in the correct antml:invoke / antml:parameter namespaced format and parse/execute reliably every time.

Error Messages/Logs

Tool calls with bare <invoke> / <parameter> tags silently failed to parse — they surfaced as "?" in the UI with no execution and no explicit error message.

Steps to Reproduce

  1. Start a Claude Code session in the VS Code extension environment.
  2. Have the assistant perform a long, multi-step editing task with many Bash and Edit/Read tool calls.
  3. Observe that, partway through, some tool calls begin to be emitted with bare <invoke> / <parameter> tags (missing the antml: namespace) and silently fail to parse.
  4. Point out the error and ask the assistant to fix it.

Expected: the assistant corrects the format immediately.
Actual: the malformed format keeps recurring (20+ times) even after acknowledgment; intermittent/state-dependent.

Claude Model

Opus 4.8 (claude-opus-4-8)

Is this a regression?

Not sure / unknown

Claude Code Version

2.1.191 (Claude Code)

Platform

Claude API (Claude Code)

Operating System

macOS 26.2 (build 25C56)

Terminal/Shell

VS Code extension environment (zsh)

Additional Information

  • /bug was unavailable in this environment, so filed manually.
  • Some tool calls in the same session parsed fine while others broke, pointing to an intermittent / state-dependent parsing failure rather than a consistent formatting mistake.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗