技能 cc-pr-review-ci
📦
cc-pr-review-ci
內容修訂 r1 高風險 ⚙️ 外部命令📁 檔案系統存取🔑 環境變數
Review TypeScript and React Pull Requests
Pull request reviews can miss issues across security, tests, performance, and maintenance. This skill organizes twelve review domains and prepares findings for GitHub comments.
支援: Claude Codex Code(CC)
安裝前請停止並請求確認。
請審查安裝計畫，並在變更檔案前取得使用者明確同意。
使用我的智慧代理安裝
將此請求複製給你的智慧代理，其中包含權威 Skill 頁面和清單。
智能體請求
Review the Skillstore skill "cc-pr-review-ci" from https://skillstore.io/skills/adrianbrowning-cc-pr-review-ci.md and its manifest at https://skillstore.io/api/skills/adrianbrowning-cc-pr-review-ci/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.你的智慧代理仍應顯示其計畫，並依安全政策請求必要確認。
Agent 可讀資源
當 AI Agent、爬蟲或腳本需要乾淨脈絡、而不是讀取完整頁面時，請使用這些連結。
測試它
正在使用「cc-pr-review-ci」。 Review local changes that add tests with mocks returning the asserted values.
預期結果:
- High priority: A test asserts its mock output without exercising production behavior.
- Suggested fix: Execute the production path and assert observable results.
- Review decision: Approved with suggestions unless a critical issue is also found.
正在使用「cc-pr-review-ci」。 Review a workflow supporting both pull requests and manual runs.
預期結果:
- High priority: Pull request fields are used during manual runs, where those fields are unavailable.
- Suggested fix: Select valid metadata for each supported event.
- The review artifact includes a finding identifier, location, explanation, and repair prompt.
安全審計
高風險v1 • 2026/10/6 開啟版本化報告
Most static flags identify Markdown, JavaScript strings, legitimate credential use, or ordinary repository review operations. Confirmed risks include unvalidated shell interpolation and predictable temporary review input. Semantic findings identify unauthenticated finding suppression, marker-based dismissal of other reviewers, and approval binding to a potentially unreviewed head commit.
15
已掃描檔案
1,310
分析行數
13
複核項目
0
已忽略的誤報
已確認的安全問題 (3)
高
Unauthenticated Dismissals Suppress Security Findings
Any negative reaction or human reply containing dismissal language qualifies without an authorization check. Matching findings are then skipped silently, allowing participants to hide unresolved issues.
Fetched replies omit author identity, and reactions are treated as counts. The explicit suppression rule cannot distinguish maintainer decisions from other participants.
高
Spoofable Marker Permits Dismissal of Other Reviewers
Existing reviews are selected only by a body marker, without checking author identity. The script attempts dismissal before successfully posting a replacement, potentially removing legitimate blocking reviews.
The query selects review IDs solely by body text, and the retrieved state is ignored. Impact depends on the token having review dismissal permission.
高
Approval Can Target an Unreviewed Head Commit
The script fetches the current PR head and attaches the earlier review artifact to that commit. A push between analysis and posting can receive approval without being reviewed.
The loaded artifact is not compared with an analyzed commit SHA. The API payload uses only the head fetched during posting.
能力複核項目 (13)
這些是真實的本機能力，對此技能而言可能是預期行為，因此需要複核，但不會按已確認的惡意行為計分。
高
Synchronous process execution
execSync(
execSync runs a shell command containing unvalidated repository and PR environment values. Controlling those values permits command injection with inherited GitHub credentials.
高
Synchronous process execution
execSync(
The review-list command interpolates unvalidated environment values into a shell. Shell metacharacters can execute additional commands if those values are attacker-controlled.
高
Synchronous process execution
execSync(
The state query executes interpolated repository and PR values through a shell without validation. Environment input control creates a command injection path.
高
Synchronous process execution
execSync(
The dismissal command includes unvalidated environment values in a shell string. An injected command runs within the credentialed posting process.
高
Synchronous process execution
const result = execSync(
Passing review content through standard input is appropriate, but the command itself interpolates unvalidated environment values. Those values can inject shell commands.
高
Ruby/shell backtick execution
`gh api repos/${owner}/${repo}/pulls/${PR_NUMBER} --jq '{sha: .head.sha}'`,
This template literal is passed to execSync and contains unvalidated environment values. Shell execution makes injected metacharacters dangerous, unlike ordinary JavaScript strings.
高
Ruby/shell backtick execution
`gh api repos/${owner}/${repo}/pulls/${PR_NUMBER}/reviews ` +
This command fragment enters execSync with unvalidated repository and PR values. Attacker control of those environment values can inject shell commands.
高
Ruby/shell backtick execution
`gh api repos/${owner}/${repo}/pulls/${PR_NUMBER}/reviews/${reviewId} --jq '.state'`,
The state-query template literal enters execSync with unchecked environment values. Shell metacharacters in those values can execute additional commands.
高
Ruby/shell backtick execution
`gh api repos/${owner}/${repo}/pulls/${PR_NUMBER}/reviews/${reviewId}/dismissals ` +
The dismissal endpoint interpolates unchecked repository and PR values before shell execution. Input control can cause credentialed command injection.
高
Ruby/shell backtick execution
`gh api repos/${owner}/${repo}/pulls/${PR_NUMBER}/reviews ` +
The review-posting endpoint is built from unchecked environment values and passed to execSync. Shell injection remains possible despite safe standard-input payload handling.
中
Temp directory access
const REVIEW_PATH = process.env.REVIEW_OUTPUT_PATH ?? "/tmp/review.json";
The default artifact path is predictable in a shared temporary directory, with no ownership or symlink checks. Shared-runner attackers could substitute review input.
中
Temp directory access
**Never post comments directly.** Write `review.json` to the path given in `$REVIEW_OUTPUT_PATH` (de
The instructions default to a predictable shared temporary artifact path. Without private storage or symlink protection, another local process could replace or redirect it.
中
Temp directory access
- **PR run**: write to `$REVIEW_OUTPUT_PATH` (default `/tmp/review.json`)
The workflow instructs writing to a predictable shared temporary path by default. Shared hosts require private artifact storage to prevent substitution or symlink redirection.
風險因素
⚙️ 外部命令 (50)
post-review.js:16 post-review.js:47 post-review.js:140 post-review.js:150 post-review.js:154 post-review.js:177 post-review.js:38 post-review.js:48 post-review.js:73 post-review.js:74 post-review.js:75-80 post-review.js:80 post-review.js:85-86 post-review.js:86-87 post-review.js:87-88 post-review.js:88-89 post-review.js:89-91 post-review.js:91-92 post-review.js:92-126 post-review.js:126-128 post-review.js:128-129 post-review.js:129-130 post-review.js:130-131 post-review.js:131-132 post-review.js:132-133 post-review.js:133-141 post-review.js:141-142 post-review.js:142-151 post-review.js:151-155 post-review.js:155-156 post-review.js:156-178 post-review.js:178-179 post-review.js:179-184 SKILL.md:8 SKILL.md:10 SKILL.md:20-23 SKILL.md:23-27 SKILL.md:27-30 SKILL.md:30-34 SKILL.md:34-37 SKILL.md:37-39 SKILL.md:39 SKILL.md:42-45 SKILL.md:45 SKILL.md:49-53 SKILL.md:53-55 SKILL.md:55 SKILL.md:56 SKILL.md:57-58 SKILL.md:58-60
📁 檔案系統存取 (11)
審計單位: codex
分享與引用此報告
分享版本化評估報告、中性徽章、嵌入卡片與引用。Skillstore 只報告證據，不判斷該 Skill 是否安全。
複製報告連結
https://skillstore.io/skills/adrianbrowning-cc-pr-review-ci/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_reportMarkdown 徽章
[](https://skillstore.io/skills/adrianbrowning-cc-pr-review-ci?utm_source=security_passport_badge)HTML 徽章
<a href="https://skillstore.io/skills/adrianbrowning-cc-pr-review-ci?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/adrianbrowning-cc-pr-review-ci/security.svg" alt="Skillstore security assessment" loading="lazy"></a>嵌入卡片
<iframe src="https://skillstore.io/embed/skills/adrianbrowning-cc-pr-review-ci.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>學術引用 (APA · BibTeX · CFF)
APA 引用
adrianbrowning. (2026). cc-pr-review-ci security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/adrianbrowning-cc-pr-review-ci/audits/1BibTeX 引用
@techreport{adrianbrowning-adrianbrowning-cc-pr-review-ci-2026,
author = {adrianbrowning},
title = {cc-pr-review-ci security audit report (audit version 1)},
institution = {Skillstore},
year = {2026},
number = {1},
url = {https://skillstore.io/skills/adrianbrowning-cc-pr-review-ci/audits/1},
note = {Author version unspecified}
}CITATION.cff
cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "cc-pr-review-ci security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
- name: "adrianbrowning"
date-released: "2026-10-06"
url: "https://skillstore.io/skills/adrianbrowning-cc-pr-review-ci/audits/1"
identifiers:
- type: other
value: "skillstore:adrianbrowning-cc-pr-review-ci:audit:1"
description: "Skillstore immutable audit report identifier"
Skillstore 評分
評分依據 證據置信度: 中 45
架構
85
可維護性
87
內容
65
社群
83
規範符合度
你可以打造什麼
Check Changes Before Opening a Pull Request
Review local TypeScript and React changes against origin/main and inspect the structured results before requesting team feedback.
Prepare Consistent Repository Reviews
Use domain checklists to prepare actionable findings with stable identifiers and proposed fixes for human assessment.
Integrate Review Artifacts into CI
Separate AI review generation from credentialed posting after hardening the script and configuring permissions in an external workflow.
試試這些提示
Review Local Changes
Review my current changes against origin/main using cc-pr-review-ci. Print the structured review locally and do not post comments.
Review a Pull Request
Review PR [number] across all twelve domains. Prepare the review artifact, report skipped domains, and leave GitHub posting to the separate script.
Validate Tests and CI
Review PR [number] across all domains, emphasizing test validity and CI permissions. State missing dependencies and explain each proposed fix.
Reassess a Changed Pull Request
Re-review PR [number] after recent commits. Verify authorized dismissals, reassess changed finding locations, and report suppressions for human approval before posting.
最佳實務
- Harden and validate the posting script, bind artifacts to reviewed commits, and use private per-job artifact storage.
- Keep AI review credentials read-only and grant the separate posting step only the GitHub permissions it requires.
- Review findings and suppressions manually, and provide optional testing skills and duplication reports when those domains matter.
避免
- Run the current posting script with untrusted environment values or a broadly privileged token.
- Treat missing optional domains or AI approval as evidence that tests passed or security is guaranteed.
- Accept arbitrary reactions as authorized dismissals or post an old review after the PR head changes.
常見問題
What languages and frameworks does this skill target?
It targets TypeScript and React pull requests, with additional checklists for security, testing, CI, scope, documentation comments, and maintainability.
Can I use it without a pull request?
Yes. It compares changes against origin/main and prints the structured review to standard output.
Does the reviewing agent publish comments?
The instructions prohibit direct posting. A separate Node.js script uses GitHub CLI to publish the review; an external workflow must enforce permissions.
Does it execute tests or generate duplication reports?
Test validity review is static. Duplication review consumes an existing jscpd report rather than generating one.
Which optional skills are needed?
Test validity requires test-validity-review. Testing can use testing-best-practice, but retains a built-in checklist when that skill is unavailable.
Is the posting script ready for untrusted CI inputs?
No. Add shell-free execution, artifact validation, authorized dismissal checks, and reviewed-commit verification before using it with untrusted inputs.