anthropics / anthropics/claude-code-action
`/install-github-app` and `docs/usage.md` examples lack `author_association` guard — self-trigger loop and denial-of-wallet exposure at scale
- 主要語言
- TypeScript
- 星號
- 8.9k
- 分支
- 2.1k
- 平均合併
- 3 天 9 小時
- 30 天內合併 PR
- 10
描述
### Problem
The workflow generated by `/install-github-app` and the example in [`docs/usage.md`](https://github.com/anthropics/claude-code-action/blob/main/docs/usage.md) do not include an `author_association` filter or a `sender.type != 'Bot'` guard in the `if:` condition. This causes two issues:
1. **Self-trigger loop :** The agent's reply comment can contain `@claude`, re-triggering the workflow indefinitely.
2. **Denial-of-wallet surface:** On public repos, any user can comment `@claude` to trigger the workflow. The action's internal `checkWritePermissions` gate (in [`src/github/validation/permissions.ts`](https://github.com/anthropics/claude-code-action/blob/main/src/github/validation/permissions.ts)) blocks non-collaborators from write actions, but the workflow still runs and may consume API tokens before that gate fires at the tool-execution layer.
### Current pattern (from `docs/usage.md`)
The [`docs/usage.md`](https://github.com/anthropics/claude-code-action/blob/main/docs/usage.md) example does not even include an `if:` condition:
```yaml
name: Claude Assistant
on:
issue_comment:
types: [created]
jobs:
claude-response:
runs-on: ubuntu-latest
steps:
- uses: anthropics/claude-code-action@v1
with:
anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
```
The `/install-github-app` generated workflow adds a content check but no actor filter:
```yaml
if: |
(github.event_name == 'issue_comment' && contains(github.event.comment.body, '@claude')) || ...
```
### Suggested addition
```yaml
if: |
contains(fromJSON('["OWNER","MEMBER","COLLABORATOR"]'), github.event.comment.author_association) &&
github.event.sender.type != 'Bot' &&
(github.event_name == 'issue_comment' && contains(github.event.comment.body, '@claude')) || ...
```
This prevents both self-trigger loops (bot comments are excluded) and non-collaborator triggers (only OWNER/MEMBER/COLLABORATOR can invoke).
貢獻指南
研究方向
Start with docs/usage.md and trace the /install-github-app entry point to the generated workflow source. Compare both workflow conditions with the permissions check in src/github/validation/permissions.ts, then verify that authorized human comments can trigger while bot and non-collaborator comments cannot. Done means both the documentation example and generated workflow include the intended guards.
由索引模型根據 Issue 內容生成。
評估
- 技術堆疊
- github-actions, typescript
- 領域
- ci-cd, documentation, security
- Issue 類型
- 缺陷
- 難度
- 3/5
- 預估耗時
- 1-2 天
- 活躍度
- 冷清
- 描述清晰度
- 基本清楚
- 新手友好度
- 68/100