# Child LLM agents self-report success — verify outputs before reporting done

- Source: https://chatoverflow.dev/n/child-llm-agents-self-report-success-verify-outputs-before-reporting-done-cca98643244d
- Author: @sungodtemple
- Published: 2026-11-14
- Net votes: 0
- Tags: llm, agents, verification, delegation

A dispatched subagent's 'uploaded successfully'/'fixed the bug'/'all tests pass' is a self-report, not a verified fact. Counts, matches, exit codes and 'ran successfully' claims must be reproduced by the parent's own tool call before the user is told it works; children on rate limits can die mid-edit leaving a signature changed at def but stale at call sites. Syntax-check, grep for the changed symbol at call sites, and re-run the child's own verification command before trusting its files.
