Dengbao Code Audit

Purpose

Use this skill to perform a source-level security audit for internal systems, especially systems that may be deployed on the public internet and need to prepare for 等保三级-style checks.

This skill does not replace formal MLPS/等保测评. It produces an engineering-focused code audit report with evidence, risk grading, and remediation guidance.

Required Output

Save the final report as Markdown. Prefer:

reports/等保代码安全检测报告.md

If a report path is requested by the user, use that path.

Workflow

1. Identify the project root and scan scope.
   • Include source code, configs, deployment manifests, CI/CD files, scripts, README/docs, and environment templates.
   • Exclude dependency/vendor/build directories such as node_modules, .git, dist, build, target, .venv, venv, coverage.
   • If the system is public-facing, raise the risk level of auth, permission, data leakage, and exposed service findings.
2. Run the bundled inventory script when useful:

python3 <skill-dir>/scripts/inventory.py <project-root> --output <workdir>/dengbao-inventory.json

3. Run the bundled pattern scanner for first-pass evidence:

python3 <skill-dir>/scripts/grep_rules.py <project-root> --output <workdir>/dengbao-findings.json

4. Read the most relevant code paths instead of relying only on grep output.
   • Trace request entry points to auth middleware/guards/interceptors.
   • Check whether permission decisions happen on the backend, not only in frontend routes/buttons.
   • Check whether audit logs contain actor, IP, time, operation, object, result, and failure reason for important operations.
   • Check whether sensitive data is encrypted or desensitized at storage, transmission, logs, export, and backup boundaries.
   • Check deployment files for public ports, admin panels, debug flags, TLS, domain/certificate references, and database/cache exposure.
5. Create a draft report with the helper if desired:

python3 <skill-dir>/scripts/summarize_findings.py \
  --inventory <workdir>/dengbao-inventory.json \
  --findings <workdir>/dengbao-findings.json \
  --output <project-root>/reports/等保代码安全检测报告.md

6. Replace script-only conclusions with reviewed findings.
   • A regex hit is evidence, not proof.
   • If implementation cannot be confirmed from source, mark it as 待确认 and list the material needed.
   • Do not report secrets verbatim. Redact tokens, passwords, private keys, session IDs, and connection strings.

What To Inspect

Load these references as needed:

• references/checklist.md: category-by-category audit checklist and evidence requirements.
• references/severity-rubric.md: risk grading rules and public-facing adjustment.
• references/framework-patterns.md: framework-specific code paths and common risky patterns.
• references/remediation-playbook.md: standard remediation guidance and acceptance criteria.
• references/report-template.md: final Markdown report structure.

Finding Requirements

Each finding must include:

• Risk level: 高危, 中危, 低危, or 待确认.
• Category: one of 登录认证, 权限控制, 安全审计, 数据安全, 接口安全, 传输安全, 运维暴露面, 备份恢复, 发布变更, 文档材料.
• Evidence: file path and line number when available.
• Impact: what can go wrong in this system.
• Remediation: concrete code/config/process change.
• Acceptance criteria: how the team can verify the fix.

Reporting Rules

• Lead with an executive summary and top risks.
• Group detailed findings by risk level, then by category.
• Include a coverage table for all 10 check categories, even if the conclusion is 待确认.
• Include a prioritized remediation roadmap: 立即整改, 近期整改, 持续治理.
• Separate code-confirmed issues from documentation/process gaps.
• Be practical. Avoid vague statements like "strengthen security" without a testable action.

High-Priority Red Flags

Always check carefully for:

• Public endpoints without backend authentication.
• Authorization enforced only by frontend menus/routes.
• Admin or privileged APIs without MFA or stronger controls.
• Hardcoded credentials, JWT secrets, access keys, database URLs, or private keys.
• SQL/NoSQL query concatenation with user-controlled input.
• Command execution, file path construction, template rendering, or SSRF using user input.
• File upload without extension, MIME, size, content, and storage path controls.
• Sensitive data in logs, exports, responses, localStorage, or backups.
• Open CORS, debug mode, permissive security headers, insecure cookies.
• Docker/Kubernetes/Nginx configs exposing SSH, database, Redis, MQ, admin consoles, or debug ports.