Usage limits should not apply to recovery from agent-caused data loss

Problem

When a Claude Code subagent destroys user work (see #43138 and #43142), the user is forced to spend their usage limit recovering what was lost. This means the user is effectively penalized twice:

1. First penalty: Hours of work destroyed by a bug in the Agent tool (stale file overwrite + git deny bypass)
2. Second penalty: The recovery process consumes the same usage limits, reducing what the user can accomplish in their billing period

What happened

During a long refactoring session, a subagent silently reverted a 19,000-line file to a state from hours earlier, destroying:

• 6+ new i18n section blocks
• ~30 dead keys removed across 50 language blocks
• Quote normalization (39 values)
• Multiple key renames and moves
• App.jsx reference updates

The user now has to spend their remaining usage limit step-by-step rebuilding everything that was lost — not doing new work, just getting back to where they were before the bug hit.

Suggestion

When Anthropic confirms that work was lost due to a platform bug (agent stale writes, permission bypass, etc.), the usage consumed during recovery should be credited back or exempted. Users should not bear the cost of recovering from bugs they didn't cause.

At minimum:

• Acknowledge that agent-caused data loss creates a double-penalty situation
• Consider a credit/refund mechanism for sessions where significant recovery work was needed due to platform bugs
• Prioritize fixing the root causes (#43138, #43142) so this scenario doesn't repeat

Related issues

• #43138 — Agent tool: subagent can overwrite file with stale content
• #43142 — Agent tool: subagent bypasses git deny rules in settings.local.json

---
🤖 Generated with Claude Code