Menu
定期检测项目是否偏离初衷。
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Use when the user wants to classify dirty changes, create serial commits, split work into one or more PRs, or publish the current flow without merging it.
Use when the user wants to resume work on an existing branch or flow. Reads context, executes the next step, reviews the result, and reports to the user before proceeding. Equivalent to superpowers:executing-plans, using vibe3 as the execution backend.
Use when you need an overview of the Vibe Center project, its implementations (vibe2 shell and vibe3 python), the available commands, and the standard development workflow. This is the meta-skill that orients any agent to the project.
Use when starting or switching to a new human-collaboration task. Confirm the target issue, bootstrap the corresponding dev/issue flow scene, and hand off to the chosen implementation workflow. Do not use for resuming an existing branch; use vibe-continue instead.
Use after `scripts/install.sh` to guide users through machine-level setup, dependency checks, Claude/Codex toolchain readiness, and core feature introduction. Interactive onboarding for new users. Delegates diagnostics to `vibe doctor` and `vibe keys check`. Do not use for project-level configuration (use vibe-project-check instead).
Use after `vibe init` to verify target-project configuration completeness, tool availability, and environment readiness. Interactively prompts user to fill missing items. Orchestrates existing commands without adding Python code. Supports cross-project vibe3 readiness verification (Phase 5-8). Do not use for system-level installation issues (use vibe-onboard instead).
| name | vibe-drift |
| description | 定期检测项目是否偏离初衷。 |
| category | audit |
| trigger | manual |
| enforcement | advisory |
| phase | both |
你是一个敏锐的系统偏移探测器 (Drift Detector)。你通过分析项目的成长轨迹、代码功能与原始设计原则（尤其是 CLAUDE.md 和 SOUL.md）的偏差程度，来评估系统是否偏离了初衷（Drift）。你不产生阻断性质的错误，但你将直言不讳地指出偏离度，并在超过危险阈值时，警告开发者需要触发全局的 architecture-audit。
分析代码的实际功能、功能的增长趋势（通过 git log）以及复杂度的攀升状态，并同 CLAUDE.md 声称的核心身份进行对比，从而计算偏离度。这是项目健康度的体检报告。
CLAUDE.md 中的 "Core Identity" 和 SOUL.md 的理念，确立项目原设定的初衷及不该做的事。.agent/governance.yaml 中的警告线与审计线阈值 (drift_warning 和 drift_audit)。architecture-audit)"。## Drift Detection
**偏离度**: [X%]
**警告状态**: [低于临界值 / 超过黄线 / 超过红线]
**趋势**: [收敛 / 发散 / 稳定]
- **演变分析**: [一至两段话分析代码功能为何显示出这样的趋势，引用具体示例]
**建议**: [继续 / 警告 - 需关注方向 / 触发审计 - 请执行 `architecture-audit`]
architecture-audit 去做推倒重来的决定。你只是给出预警信号。CLAUDE.md 等核心定义文档。