Prompt turns Opus 5.5 and Jev into a parallel bug-fixing coding agent
A copy-paste JSON prompt that runs three Opus patch attempts in parallel, filters them by tests, and has Jev pick the final patch from the passing candidates based on test evidence.
THIS PROMPT TURNS OPUS 5.5 + JEV INTO A PARALLEL CODING AGENT Give this to your AI agent before its next bug fix: { "goal": "Find a verified patch for one failing test.", "base": "Pin one commit and create a separate Git worktree for each worker.", "workers": ["minimal_fix", "edge_case", "refactor"], "limits": {"parallel": 3, "retries": 1}, "checks": ["targeted test", "regression suite"], "workflow": [ "Run three Opus 5.5 patch attempts in parallel.", "Capture each diff, test result, and failure log.", "Exclude failed candidates. Let Jev choose from passing patches using test evidence and patch scope.", "Replay the selected patch from the pinned base in a clean worktree." ], "output": ["selected.patch", "test-evidence", "candidate comparison", "decision record", "rollback plan"], "stop": "Report and stop if no patch passes, Jev is unavailable, or clean replay fails.", "gate": "Prepare for human review. Do not merge or deploy." } Opus 5.5 drafts three fixes → tests filter the patches → Jev picks from the ones that pass → a clean worktree checks the winner again. Two patches may pass the same tests. Jev still has to choose one - and you can inspect the evidence behind that choice. Keep this prompt for your next hard bug ↓
This post quotes another post: “GPT-6 SOL + Jev self-improving coding agent prompt”