提示词：Opus 5.5 + Jev 组成并行修 bug 编码 agent
给出可直接复制的 JSON 提示词，让三个 Opus 并行出补丁、测试过滤，再由 Jev 依据测试证据挑选最终补丁。
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 ↓
这条是引用着另一条写的：「GPT-6 SOL + Jev 自修复编码 Agent 提示词」