AGENT SKILL
refactor
暂无简介
使用场景代码审查
- License 许可
- MIT
- 最近更新
- 2026-04-13
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- id
- refactor
- handler
- .agents/refactor/handler.js
- version
- 1.0.0
- read_only
- false
- output_event
- RefactoringProposed
- output_schema
- Patch[]
- aggregate_root
- Refactoring
- bounded_context
- Transformation
- authorization_required_level
- 2
02 资源文件 1
- 脚本handler.js9.3 KB
03 SKILL.md 正文
SKILL: Refactor
Consumes Finding events where auto_fixable: true and produces safe,
reviewable unified diff patches.
Patches MUST NOT be applied autonomously — all patches must be reviewed and approved before application.
Activation
agent:
id: "refactor-agent-01"
skill_set:
- "refactor"
authorization_level: 2
read_only: false
Patch Generation Rules
| Rule | Description |
|---|---|
| Minimal Diff | Smallest possible change that resolves the finding |
| No Behavior Change | Semantically equivalent unless a security fix requires it (must be documented) |
| Test Preservation | MUST NOT break existing tests; test changes must be in same diff |
| One Finding, One Patch | Each patch addresses exactly one Finding.id |
Output Contract
interface Patch {
id: string; // uuid-v4
finding_id: string; // ID of the Finding being addressed
skill: "refactor";
status: "proposed" | "approved" | "rejected" | "applied";
diff: string; // Unified diff format
files_modified: string[];
behavior_change: boolean;
behavior_change_reason?: string;
created_at: string; // ISO-8601
approved_by?: string;
}
Suppression
If a finding is suppressed via // agent-suppress, this skill MUST NOT
generate a patch for it.
Schema version: 1.0.0 · Vendor-neutral