一键导入
codex-sol-fusion
Use when the user explicitly requests Sol Fusion, Codex Sol/Luna orchestration, or a cost-quality split between GPT-5.6 Sol and Luna.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use when the user explicitly requests Sol Fusion, Codex Sol/Luna orchestration, or a cost-quality split between GPT-5.6 Sol and Luna.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | codex-sol-fusion |
| description | Use when the user explicitly requests Sol Fusion, Codex Sol/Luna orchestration, or a cost-quality split between GPT-5.6 Sol and Luna. |
Run a host-native Codex workflow in which Sol owns judgment and Luna handles bounded work. Do not start hidden nested Codex processes. Agent model pins are exact only after the plugin's namespaced profiles have been installed.
The explicit $codex-sol-fusion token also activates this plugin's otherwise inert ledger hooks for the current session. Do not claim the hooks affect unrelated sessions.
| Role | Agent | Model and effort | Use |
|---|---|---|---|
| Gatherer | csf_luna_gatherer | Luna, low | Narrow read-only evidence |
| Worker | csf_luna_worker | Luna, max | Isolated implementation and tests |
| Reviewer | csf_luna_reviewer | Luna, high | Routine read-only review |
| Verifier | csf_sol_verifier | Sol, max | Independent high-risk closure |
The invoking root agent is the chair. Use the Sol policy for an exact gpt-5.6-sol runtime; otherwise use the lean Luna policy without claiming that the runtime model changed. Lifecycle hooks are intentionally not used for chair injection because Codex cannot reliably distinguish root and child sessions there.
.workflow/LEDGER.md with stable checkboxes.fork_turns: "none"; include the assigned ledger IDs, boundaries, acceptance criteria, and required evidence in the task message instead of sharing the chair's full history..workflow/scratch/.csf_sol_verifier for independent closure when the risk warrants it.Mark a waived item only as [~] deferred: reason. Never mark an item complete from a worker's narration alone; verify the current artifact. Keep external actions within the user's explicit authority.