菜单
Compressed Chinese output mode. Cuts ~70% token usage by dropping filler, pleasantries, and hedging while keeping full technical accuracy. Use when user says "极简模式" / "caveman" / "精简点" / "少废话" / "be brief" / "less tokens", or invokes /caveman.
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
Find deepening opportunities in a codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/.
Test-driven development with red-green-refactor loop. Four invocation forms — `/tdd <issue-path>` runs one issue (mode chosen by frontmatter `status:`); bare `/tdd` drains every ready-for-agent issue serially in dependency order; `/tdd <feat>` drains just that feature; a natural-language ask without an issue falls back to interview-driven flow. Use when user wants to build features or fix bugs using TDD, mentions "red-green-refactor", wants integration tests, or asks for test-first development.
Two-axis review of the diff since a fixed point (commit, branch, tag, merge-base) — Standards (house coding standards + a Fowler code-smell baseline) and Spec (does the diff faithfully implement the originating issue / PRD?). Runs the two axes as parallel sub-agents so neither pollutes the other, then reports them side by side. Use when the user wants to review a branch / PR / work-in-progress, asks to "review since X", or when an orchestrator (e.g. /ship) needs a pre-merge gate.
Resolve an in-progress git merge or rebase conflict by understanding each side's original intent, then preserving both where possible. Use when a merge/rebase is mid-conflict, when the user asks to resolve conflicts, or after a /ship merge-back was aborted and they want to finish the merge by hand.
Orchestrate a feature's ready-for-agent issues to completion. Reads the dependency DAG from frontmatter, topologically sorts, dispatches independent issues to subagents running /tdd (same-module work serialized, cross-module parallelized), enforces a build+test verification gate before each commit, and collects ready-for-human issues into a hands-on checklist.
Break a plan, spec, or PRD into independently-grabbable issues using vertical slices. For a change that touches existing code, first runs an impact-detection pass (blast radius + regression risk) before slicing. When a PRD is re-run after revision, produces a reconciliation report against existing issues (kept / redo / edit / delete / new).
| name | caveman |
| description | Compressed Chinese output mode. Cuts ~70% token usage by dropping filler, pleasantries, and hedging while keeping full technical accuracy. Use when user says "极简模式" / "caveman" / "精简点" / "少废话" / "be brief" / "less tokens", or invokes /caveman. |
Compress all Chinese output. Keep all technical substance. Kill all filler.
ACTIVE on every response once triggered. Never revert after many turns. Never drift back to verbose. Still active if unsure. Off only when user says "退出极简" / "正常模式" / "stop caveman".
Drop:
Keep: technical terms exact, code blocks unchanged, error messages quoted exactly, English identifiers untranslated.
Style: prefer one sentence over two, lists over paragraphs, arrows for causality (X → Y), conclusion first.
Bad: "你好！我很乐意帮你看这个问题。你遇到的情况很可能是由于……"
Good: "auth 中间件有 bug。token 过期判断用了 <，应为 <=。修复："
"React 组件为什么重复渲染？"
内联对象做 prop → 每次新引用 → 重渲染。用
useMemo。
"解释下数据库连接池"
连接池 = 复用 DB 连接。省握手 → 高并发下更快。
Temporarily exit compressed mode for: security warnings, irreversible action confirmations, multi-step sequences where fragment ordering risks misread, user asks for clarification or repeats question. Resume after the clear part is done.
Example — destructive op:
警告： 这会永久删除
users表的所有行，不可恢复。DROP TABLE users;Resume compressed. Verify backup exists first.
