[BUG] Claude Code trusts subagent file-write reports without verifying files exist on disk
Resolved 💬 3 comments Opened Mar 19, 2026 by prince-hyly Closed Mar 23, 2026
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?
Claude Code trusts the agent's text output as ground truth. This leads to:
- False confirmation of success to the user
- Blocking legitimate re-runs based on stale/false reported state
- No warning when files don't exist
What Should Happen?
After any subagent reports writing files, Claude Code should verify the files exist on disk before telling the user the operation succeeded and making decisions based on that reported state.
Error Messages/Logs
Steps to Reproduce
- Run /plan --create <spec-path> using the tdd-l1-spec-generator plugin
- The orchestrator subagent reports: "All 6 files written successfully" with file paths listed
- Claude Code presents the plan as complete and ready
- User deletes the plan folder (or it was never written)
- Claude Code later blocks the user from re-running --create, saying "a fully finalized plan already exists at
plans/20260319-sub-intent-classifier"
- No such directory exists on disk
Claude Model
Sonnet (default)
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
2.1.76
Platform
Anthropic API
Operating System
macOS
Terminal/Shell
Terminal.app (macOS)
Additional Information
_No response_
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗