菜单
Manage Re:ZERO warning memory. Use when witch warnings pass, when processing .rezero/memory/rem.md, or when resolved warnings should be removed.
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
Initialize a repository for Re:ZERO Loop. Use for `/rezero init` in Pi, `/rezero:run init` in Claude Code, or `$rezero:run init` in Codex: detect project type, set up witch evaluation tools, create memory ignore rules, and record unavailable tools.
Orchestrate the Re:ZERO Loop. Use for /rezero requests in Pi: split large work, let Subaru implement, call Seven Witches, send warnings to Rem, reset on fail, and commit on pass.
Run the Re:ZERO Loop. Use for `/rezero:run` in Claude Code or `$rezero:run` in Codex: initialize if needed, plan tasks, let Subaru implement, run Seven Witches review, remember warnings, reset on failures, and commit on pass.
Run the Seven Witches evaluation for a Re:ZERO attempt. Use after Subaru implements and verifies a task to get independent pass/warning/fail verdicts.
Subaru's single-task Re:ZERO implementation loop. Use inside rezero after task splitting: implement, verify, survive witches, reset on fail, commit on pass.
Split a large Re:ZERO request into small ordered tasks with clear done criteria. Use before implementation when the request spans multiple features, files, or risks.
| name | rezero-rem |
| description | Manage Re:ZERO warning memory. Use when witch warnings pass, when processing .rezero/memory/rem.md, or when resolved warnings should be removed. |
| license | MIT |
Goal → preserve non-blocking witch warnings until they are resolved and accepted.
When a route passes with warnings:
.rezero/memory/rem.md..rezero/memory/rem.md in the assistant chat.## Warning <number>
- Task: <request>
- Witch: <name>
- Warning: <short issue>
- Evidence: <command/output/file>
Treat Rem fixes as normal Re:ZERO attempts:
.rezero/memory/rem.md.rezero-witches.fail..rezero/memory/rem.md when no warnings remain..rezero/memory/rem.md is created or appended, the newly written content must be printed in the assistant chat before continuing.