code-impact-analysis
Safely analyze code impact before refactors, behavior changes, API changes, and cross-module edits.
ソース情報
- リポジトリ
- 777genius/universal-plugins-for-ai-agents
- ソースの最終更新活動
- 2026年5月22日 15:02
- 検出された SKILL.md の言語
- 英語
- スター
- 6
- フォーク
- 0
インストール方法
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
ソースファイルを確認
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
SKILL.md を表示中
SKILL.md
ソースの指示 · 読み取り専用プレビュー- name
- code-impact-analysis
- description
- Safely analyze code impact before refactors, behavior changes, API changes, and cross-module edits.
- execution_mode
- docs_only
- supported_agents
- ["codex","claude"]
# Code Impact Analysis
## What it does
Builds a verified impact set before changing shared behavior, public interfaces, state stores, routes, API DTOs, database models, feature flags, or memory/transcription flows.
## When to use
Use this before any refactor or behavior change where missing one caller could break the product.
## How to run
1. Locate the definition with Serena or `rg`.
2. Find exact callsites with `rg`.
3. Find type-aware references with Serena.
4. Ask CGC for callers, callees, and complexity.
5. Use Semble to find similar patterns that exact search may miss.
6. Decide the minimum edit set.
7. Run focused tests, typecheck, or app verification.
Useful commands:
```bash
rg -n "symbolName\\(" .
uvx --from semble==0.2.0 semble search "similar usage of symbolName" .
cgc analyze callers symbolName --context "$(basename "$PWD")"
cgc analyze calls symbolName --context "$(basename "$PWD")"
cgc analyze complexity --limit 20 --context "$(basename "$PWD")"
```
## Constraints
- If `rg` and Serena disagree, inspect both results. Generated code, dynamic dispatch, and framework magic can explain the difference.
- If CGC disagrees with `rg` or Serena, treat CGC as incomplete until verified.
- If Semble finds a pattern outside exact symbol search, inspect it before concluding the impact set is small.
- If a change crosses modules or user-facing workflows, add or update tests near the affected behavior.
- Stop and ask for direction only when unrelated dirty user changes block a safe edit, the change requires a product decision, or no reasonable verification is possible.
GitHubで見る