security-audit（安全审计）

Security Audit Workflow

Run a comprehensive security audit that combines automated static analysis, threat modeling, and multi-perspective council review. This skill produces a prioritized audit report with actionable remediation steps.

Scope Exclusions

[!IMPORTANT] This audit covers application-level security only. Production infrastructure (TLS termination, reverse proxy, network segmentation, firewall rules, DNS) is managed by a separate project and is out of scope. Do not flag missing TLS, reverse proxy configuration, network-level MITM risks, or production deployment topology as findings.

Step 1: Define Audit Scope

Ask the user:

• Scope: Full codebase or specific area? (e.g., apps/api/src/auth/, apps/web/src/)
• Trigger: What prompted this audit? (routine, pre-release, security incident, new feature, dependency update)
• Focus areas: Authentication, API security, data protection, frontend security, or all?

CHECKPOINT: Confirm the audit scope and focus areas with the user before proceeding.

Step 2: Automated SAST Scanning

Invoke /security-scanning:security-sast on the defined scope.

Scan for:

• Injection: SQL injection, NoSQL injection, command injection, LDAP injection
• XSS: Reflected, stored, and DOM-based cross-site scripting
• CSRF: Missing CSRF protections on state-changing endpoints
• Authentication: Weak password policies, broken auth flows, session fixation
• Secrets: Hardcoded API keys, passwords, tokens, connection strings
• Dependencies: Known vulnerabilities in npm packages (CVEs)
• Deserialization: Insecure deserialization patterns
• Prototype pollution: JavaScript-specific object manipulation attacks

Collect every finding with:

• Severity: Critical / High / Medium / Low / Info
• Category: OWASP Top 10 mapping
• Location: File path and line number
• Description: What the vulnerability is
• Evidence: The specific code pattern that triggered the finding
• Remediation: How to fix it

Step 3: Security Hardening Review

Invoke /security-scanning:security-hardening for a comprehensive hardening review:

HTTP Security

• Security headers (HSTS, CSP, X-Frame-Options, X-Content-Type-Options)
• CORS configuration (allowed origins, methods, headers)
• Cookie security flags (HttpOnly, Secure, SameSite)

API Security

• Rate limiting on all public endpoints
• Input validation completeness (every user input validated)
• Output encoding to prevent injection
• API key and token management
• Request size limits

Authentication and Authorization

• Password hashing algorithm (bcrypt, argon2)
• JWT configuration (algorithm, expiration, refresh)
• Session management (timeout, invalidation, rotation)
• Role-based access control enforcement
• OAuth/OIDC configuration (if applicable)

Data Protection

• PII encrypted at rest
• Sensitive data encrypted in transit (TLS 1.2+)
• Logging does not include sensitive data
• Error messages do not leak internal details
• Database connection uses SSL

Environment Variables

• No secrets hardcoded in code or logs
• Startup validation rejects weak/default secrets in production

Step 4: STRIDE Threat Modeling

Invoke /security-scanning:stride-analysis-patterns to systematically model threats:

Spoofing

• Can user identities be faked? (auth bypass, token theft)
• Can service identities be spoofed? (service-to-service auth)

Tampering

• Can request data be modified in transit?
• Can database records be altered without authorization?
• Can client-side data (localStorage, cookies) be tampered?

Repudiation

• Are all security-relevant actions logged?
• Can a user deny performing an action?
• Is there an audit trail for data changes?

Information Disclosure

• Can sensitive data leak through error messages?
• Can unauthorized users access other users' data?
• Are there timing attacks or side-channel leaks?

Denial of Service

• Can the system be overwhelmed by excessive requests?
• Are there resource-intensive endpoints without rate limiting?
• Can large payloads cause memory exhaustion?

Elevation of Privilege

• Can a regular user access admin functionality?
• Can horizontal privilege escalation occur (access another user's resources)?
• Are there IDOR (Insecure Direct Object Reference) vulnerabilities?

Document each threat with:

• Likelihood: High / Medium / Low
• Impact: Critical / High / Medium / Low
• Risk Score: Likelihood x Impact
• Existing Mitigations: What's already in place
• Gaps: What's missing

Step 5: Attack Tree Analysis

For the top 3 highest-risk threats identified in STRIDE:

Invoke /security-scanning:attack-tree-construction to build attack trees showing:

• Attack goal: What the attacker wants to achieve
• Attack paths: Different ways to reach the goal
• Sub-goals: Intermediate steps required
• Required resources: Attacker skill level, tools, access needed
• Existing defenses: Current mitigations along each path
• Weakest links: Where defenses are thinnest

Step 6: Architecture Council Security Review

Activate a subset of the Architecture Council (defined in .claude/councils/architecture-council.md) for security review:

Model Selection: See the Model Selection section in README.md for mapping agent model specs to Task tool parameters.

Security Engineer (Lead) — consult: security-scanning

• Validate automated findings (identify false positives)
• Prioritize remediation based on actual risk
• Assess overall security posture
• Assessment: Strong / Adequate / Needs Improvement / Critical Risk

Principal Engineer — consult: full-stack-orchestration

• Assess architectural implications of required remediations
• Identify systemic patterns that create vulnerabilities
• Recommend architectural changes for defense-in-depth
• Assessment: Architecturally sound / Needs refactoring / Fundamental issues

Backend Specialist — consult: backend-development

• Evaluate backend-specific security patterns
• Assess API security implementation quality
• Review database access patterns for injection risks
• Assessment: Well-secured / Some gaps / Significant concerns

Step 7: Present Audit Report

Generate a structured security audit report:

Executive Summary

• Overall Security Posture: Strong / Adequate / Needs Improvement / Critical
• Total Findings: Count by severity (Critical, High, Medium, Low)
• Top 3 Risks: The most important issues to address
• STRIDE Coverage: Summary of threat categories with risk levels

Critical and High Findings (must-fix)

For each finding:

• Severity, category, and OWASP mapping
• File and line number
• Description with evidence
• Step-by-step remediation
• Effort estimate (quick fix / moderate / significant)

Medium Findings (should-fix)

Same format as above, prioritized by risk.

Low and Info Findings (track)

Summary list for tracking as technical debt.

Threat Model Summary

• STRIDE analysis results table
• Top 3 attack trees (visual or structured text)
• Defense gap analysis

Recommended Actions (prioritized)

1. Immediate (Critical): Must fix before next deployment
2. This Sprint (High): Fix within current work cycle
3. Next Sprint (Medium): Schedule for upcoming work
4. Backlog (Low): Track as technical debt

CHECKPOINT: Present the full audit report to the user. Ask which findings they want to remediate now, and which to track for later.

Step 8: Remediation (Optional)

If the user chooses to remediate findings now:

1. Address findings in priority order (Critical first, then High)
2. For each fix:

- Apply the remediation - Re-run the relevant SAST scan to verify the fix - Run tests to ensure no regressions

1. Commit each fix with: fix(security): remediate <finding-description>

After remediation, suggest the next step (see below).

Step 9: Create Issues for Tracked Findings

For any findings that the user chooses to track for later (not remediated immediately in Step 8), create GitHub issues so they are not lost.

CHECKPOINT: Present a summary table of findings to be created as issues before proceeding.

Ask the user to confirm which findings should become issues. Wait for approval before creating.

[!NOTE] Add the build-ready label to findings with a clear, scoped remediation (the fix is known and does not require architectural decisions). Omit build-ready for findings that need further planning via /plan-feature — for example, findings that require schema changes, new infrastructure, or cross-cutting architectural decisions.

1. For each approved finding, create a GitHub issue: ISSUE_URL=$(gh issue create \ --title "security: <finding-description>" \ --body "<body>" \ --label "enhancement" \ --label "security-audit" \ --label "build-ready") ISSUE_NUM=$(echo "$ISSUE_URL" | grep -o '[0-9]*$') Omit --label "build-ready" for findings that need further planning (see NOTE above). The issue body should include:

- Context: Which audit identified this finding and the audit date - Problem: Description with file path and line number - Recommended Fix: Step-by-step remediation from the audit report - Priority: Severity rating and recommended timeline - Files: Affected file paths

1. Add each created issue to the project board and set fields: ITEM_ID=$(gh project item-add {PROJECT_NUMBER} --owner {OWNER} --url "$ISSUE_URL" --format json | python3 -c "import json,sys; print(json.load(sys.stdin)['id'])") gh project item-edit --project-id {PROJECT_ID} --id "$ITEM_ID" --field-id {PHASE_FIELD_ID} --single-select-option-id <phase-option-id> gh project item-edit --project-id {PROJECT_ID} --id "$ITEM_ID" --field-id {SIZE_FIELD_ID} --single-select-option-id <size-option-id> gh project item-edit --project-id {PROJECT_ID} --id "$ITEM_ID" --field-id {START_FIELD_ID} --date <start-date> gh project item-edit --project-id {PROJECT_ID} --id "$ITEM_ID" --field-id {TARGET_FIELD_ID} --date <target-date> Set phase, size, and dates appropriate to the finding's priority and the current milestone: Severity Typical Size Schedule Priority Critical/High M (3 days) or L (5 days) Current active milestone, near the front of the queue Medium S (2 days) Current milestone, after existing scheduled items Low/Info XS (1 day) Current milestone or next, at the end of the queue
2. Report all created issue numbers ($ISSUE_NUM) and URLs to the user.

[!IMPORTANT] Every tracked finding must have both a GitHub issue and a project board entry. Creating an issue without adding it to the project board is what causes orphaned work items that /build-feature cannot discover via auto-pick.

Step 10: Close Tracking Issue (Conditional)

If the user provided a GitHub issue number as an argument (e.g., /security-audit #137):

1. Check off any audit checklist items in the issue body using gh issue edit
2. CHECKPOINT: Ask the user whether to close the issue now that the audit is complete. Present the issue number and title for confirmation.
3. If approved, close the issue: gh issue close <number> --reason completed

If no issue number was provided, skip this step.

Step 11: Hand Off

Present the next step to the user:

• If code was changed (remediation applied): Run /review-code to verify fixes in context of other changes
• If no code was changed (audit-only): No further action needed — track findings as issues or technical debt
• If starting a new feature: Run /plan-feature to begin the next feature with audit findings in mind

Pipeline: /plan-feature → /build-feature or /build-api → /review-code → /submit-pr

Standalone: /security-audit can be run at any point in the pipeline or independently.