STUPID-2026-0086
Claude Code silently deleted its own shared task-store files ~5 seconds after a teammate agent finished the last task, four times, with no Task tool call in the logs (GitHub #88346)
Why 5/10
The agent wiped its own task-coordination files; no user code or data affected.
- Scope
- 0/4
- Time, tokens or the agent's own working state only
- Recoverability
- 3/3
- Permanently lost or irreversible, including data that was exposed
- Did it happen
- 2/2
- Happened in real use
- Beyond the user
- 0/1
- Only the user's own non-production work
The score is the sum of these four factors. How each factor is scored · scored 2026-10-29
Instruction given
No explicit instruction tied to the deletion — a lead session had created tasks in a shared task store and a teammate agent was working through them normally; no command to clear, archive, or resync the store was ever given.
Expected behavior
Task JSON files under a shared team/lead task store should persist until a user or an explicit Task tool call removes them. Completing the highest-numbered task in the store should not trigger any deletion of the store itself.
Actual behavior
Across four separate occurrences on Claude Code versions 2.1.234-2.1.237, task files under `~/.claude/tasks/session-<id>/` disappeared roughly five seconds after a teammate agent completed the highest-numbered task in a store it shared with a lead session. The reporter's own transcript review found no Task tool_use of any kind in any session during the deletion windows. Two distinct patterns occurred: the entire store emptied with `.highwatermark` rewritten to match the just-completed task's ID, or a single file was removed while the watermark was left stale. All four affected stores were shared team/lead task directories.
Damage
Task-tracking state for shared multi-agent sessions was wiped outright — not moved to trash, no backup — losing the record of what had been assigned and completed across a lead/teammate task store. No source code or application data was affected; the loss was confined to Claude Code's own internal task-coordination files.
Full write-up
On August 20, 2026, a Claude Code user filed GitHub issue #88346 documenting a reproducible data-loss bug in the CLI's multi-agent "teammate" feature: task JSON files stored under ~/.claude/tasks/session-<id>/ for a shared lead/teammate task store vanished roughly five seconds after a teammate agent completed the highest-numbered task in that store — with no corresponding Task tool call anywhere in the session transcripts. The reporter documented four occurrences across versions 2.1.234 through 2.1.237 on Ubuntu Linux, in two distinct failure shapes: a full-store wipe with the .highwatermark marker rewritten to the completed task's ID, or a single file silently removed while the marker was left stale. Filed with forensic timestamps and explicit reproduction steps (create tasks in a lead session, spawn a teammate sharing the store, have it complete the last task, watch the store empty within seconds), the issue was labeled bug, has repro, and data-loss, and remained open with no maintainer response as of publication. No source code or user files were lost — only the shared task-tracking state itself, which broke continuity for teams relying on it to coordinate work across lead and teammate agents.
Classification
- Agent
- Claude Code
- Failure mode
- Destructive Action
- Root cause
- Other
- Domain
- Other
- OWASP Agentic Top 10 (our mapping)
- ASI02 Tool Misuse and Exploitation list
- MIT AI Risk Repository (our mapping)
- 7.3 Lack of capability or robustness taxonomy
- Source
- Github Issue
Related incidents
Cite this record
StupidLLM. "STUPID-2026-0086: Claude Code silently deleted its own shared task-store files ~5 seconds after a teammate agent finished the last task, four times, with no Task tool call in the logs (GitHub #88346)." AI Agent Incident Database, 2026. https://www.stupidllm.com/incident/STUPID-2026-0086/ (accessed 2026-11-22). CC BY 4.0.
← STUPID-2026-0085By IDSTUPID-2026-0087 →
Get told when an agent breaks something
We document AI agent failures daily, severity-scored against a published scale. Each Monday, if any scored 7 or above that week — deleted data, leaked secrets, broken production — you get one email with the sources. When nothing did, you get nothing.
This database is callable over MCP — query it from inside your agent.