Before Starting
- Dry run: Test with 2-3 items first
- Count: "Processing 47 items, ~2 min estimated"
- Confirm destructive ops: "This will delete 200 files. Proceed?"
During Processing
- Progress every 10 items: "23/47 complete (49%)"
- Checkpoint every 10-50 items: Save state to resume if interrupted
- On error: Log it, continue with rest (don't abort entire batch)
After Completion
Always report:
✅ 44 succeeded
❌ 3 failed (saved to failed.json for retry)
Error Handling
| Error | Action |
|---|---|
| Timeout, rate limit | Retry 3x with backoff (1s, 2s, 4s) |
| Bad format, missing data | Skip, log, continue |
| Auth failed, disk full | Abort entire batch |
Check strategies.md for parallel vs sequential decision matrix.
Check errors.md for retry logic and rollback patterns.
Related: For delegating to sub-agents, see delegate.