All subagent (Agent/Task) tool calls fail with `[Tool result missing due to internal error]` while main-thread tool calls succeed
Summary
Every tool call issued from within a spawned subagent (the Agent tool / Task subagent) returns [Tool result missing due to internal error] — including the most trivial calls (Bash echo HARNESS_OK, Bash git status -s, Read). Meanwhile, tool calls on the main conversation thread work normally in the very same session.
The result is that subagent delegation is completely unusable: any subagent that tries to call a tool immediately fails on its first tool use.
Environment
- Claude Code: 2.1.172
- Node: v22.16.0
- OS: Linux 6.8.0-117-generic
- Subagent launch mode:
Agenttool withrun_in_background: true
Reproduction
- In a session where the main thread's tools work fine, spawn a subagent via the
Agenttool. - Have the subagent perform any single tool call, e.g.
Bashrunningecho HARNESS_OK. - The subagent receives
[Tool result missing due to internal error]instead of the command output.
This was reproduced 5 consecutive times over ~30 minutes, across 3 different subagent types (a custom security-engineer agent ×4 and the built-in general-purpose agent ×1). All 5 failed on the very first tool call. Throughout the same period, main-thread Bash / Read / gh calls succeeded every time.
It is not specific to any command or repository — true, echo, and git status -s all fail identically, and Read fails as well.
Expected behavior
Subagent tool calls should execute and return their results, exactly as main-thread tool calls do.
Actual behavior
Every subagent tool call returns the literal string:
[Tool result missing due to internal error]
while no error is surfaced for the main thread.
Impact
Subagent / Task delegation is fully broken. Any workflow that fans work out to subagents cannot make a single tool call, so delegated tasks abort at their first step.
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗