Adversarial review subagent
Before counting work as done, a subagent in a fresh context reviews only the diff against the plan and reports gaps. Tell it to flag only correctness gaps.
proof: logs blocks: agent grading its own homework
Source
https://code.claude.com/docs/en/best-practicesChecked 2026-10-16. Read the original before installing.
By harness
| Harness | Status | How |
|---|---|---|
| Cursor | adapted | Use a Task subagent; can gate with subagentStop. Untested. |
| Claude Code | native | Custom agent in .claude/agents/ or /code-review. |
| Grok Build | adapted | Grok Build supports subagents (awesome-grok-build). Untested. |
| Codex | untested | Untested. |
| AGENTS.md generic | untested | Describe as a step in AGENTS.md. Untested. |
By model family
| Claude | source |
| GPT | untested |
| Grok | untested |
| Gemini | untested |
Sponsor slot: your verification tool here, one sponsor per week. Details
Proof of the week
One verification pattern every week, adapted per harness and model. No spam.