使用方法
- 复制安装命令,在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/woohyun212/security-skill --skill secure-code-review开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。
What this skill does
Performs a structured security-focused code review across 10 domains: Input Validation, Authentication & Session Management, Authorization & Access Control, Injection Prevention, Cryptography, Error Handling & Logging, Data Protection, Dependency Security, Configuration Security, and Race Conditions. Maps findings to CWE identifiers and severity levels, and produces a report with remediation guidance and proof-of-concept notes.
When to use
- During pull request review for security-sensitive code changes (auth, payments, file handling, crypto)
- Before a production release to verify security controls are in place
- When onboarding a legacy codebase with no prior security review history
- When a bug bounty report or vulnerability disclosure references a code-level flaw requiring root cause analysis
Prerequisites
- No external tools required (checklist-based; use with source code access)
- Read access to the source code under review
- (Optional) Static analysis output from tools such as Semgrep, CodeQL, or Bandit to supplement manual review
- CWE reference: https://cwe.mitre.org
Inputs
| Item | Description | Example |
|---|---|---|
REPO_PATH | Path to the source code under review | /src/api/ |
LANGUAGE | Primary language(s) of the codebase | Python, TypeScript, Go |
REVIEW_SCOPE | Files, modules, or PR diff to review | auth/, payments/handler.py, PR #342 |
APP_TYPE | Application type | web-api / cli / library / mobile |
DATA_SENSITIVITY | Highest data classification handled | PII / PCI / internal / public |
Workflow
Step 1: Define Review Scope
Before starting the checklist, bound the review to avoid scope creep and ensure thoroughness within the defined boundary.
## Review Scope
- **Repository / Path**: <REPO_PATH>
- **Language(s)**: <LANGUAGE>
- **Files reviewed**: <list each file or PR diff>
- **Review date**: <YYYY-MM-DD>
- **Reviewer**: <name or agent>
- **Total lines reviewed**: <approximate>Prioritize review order:
- Authentication and authorization code
- External input handling (HTTP request parsing, file uploads, deserialization)
- Database access and query construction
- Cryptographic operations
- Configuration loading and secret handling
- All remaining files in scope
Step 2: Apply the 10-Domain Security Checklist
Work through each domain systematically. Mark each item as [x] (pass), [ ] (fail / finding), or [-] (not applicable). For each [ ] item, create a finding entry in Step 3.
Reference: See REFERENCE.md for the full 10-domain security checklist with CWE references, severity classification table, and finding detail examples.
Work through each domain systematically. Mark each item as [x] (pass), [ ] (fail / finding), or [-] (not applicable). For each [ ] item, create a finding entry in Step 3.
Step 3: Classify Findings by Severity
Reference: See REFERENCE.md for the severity classification table (CRITICAL/HIGH/MEDIUM/LOW/INFO with criteria and examples).
Step 4: Generate Report
Produce the final report using the templates below.
Summary
## Secure Code Review — Summary
- **System / PR**: <SYSTEM_NAME or PR link>
- **Files reviewed**: <N files, ~X lines>
- **Review date**: <YYYY-MM-DD>
- **Reviewer**: <name or agent>
### Finding Counts
| Severity | Count |
|----------|-------|
| CRITICAL | 0 |
| HIGH | 2 |
| MEDIUM | 3 |
| LOW | 4 |
| INFO | 2 |
| **Total** | **11** |Reference: See REFERENCE.md for finding detail examples and positive observations/recommendations templates.
Done when
- All 10 domains have been worked through for the defined review scope
- Every checklist failure (
[ ]) has a corresponding finding entry with file, line, CWE, impact, PoC, and remediation - All findings are assigned a severity level (CRITICAL / HIGH / MEDIUM / LOW / INFO)
- The summary table accurately reflects finding counts
- Positive observations and recommendations sections are complete
- The report is saved or delivered to the requesting team
Failure modes
| Issue | Cause | Solution |
|---|---|---|
| Review scope too broad to complete in one session | Large codebase or PR | Prioritize Domains 1–5 (highest risk) first; schedule remaining domains as a follow-up |
| CWE mapping unclear | Unfamiliarity with CWE taxonomy | Search https://cwe.mitre.org using the vulnerability keyword; use the closest parent CWE if an exact match is not found |
| Finding severity difficult to calibrate | Unclear exploitability | Default to the higher severity when in doubt; document the uncertainty in the finding |
| Static analysis tool produces excessive false positives | Overly broad ruleset | Tune the ruleset; use manual review to confirm or dismiss each flagged item |
| Source code unavailable (black-box context) | Reviewing compiled or obfuscated output | Switch to the owasp-check skill for behavioral testing; note the limitation in the report |
Notes
- This skill is intended for manual security review. It complements but does not replace automated SAST/DAST tooling.
- For dependency-specific findings, use the
dependency-auditskill to get detailed CVE data and upgrade paths. - Findings marked CRITICAL should block merge or deployment until remediated. HIGH findings should be resolved before the next release unless a documented, time-bound exception is approved.
- CWE IDs provide a stable reference that maps to NVD/CVE data, OWASP guidance, and many SAST tool outputs — always include them in findings shared with development teams.
- When reviewing authentication or cryptographic code, consider pairing with a second reviewer; these domains have the highest consequence for subtle mistakes.