Phase 4: Security Design Review

Phase 4: Security Design Review

Type: Evaluative Executor: LLM Knowledge: Control Sets, OWASP References

⚠️ MANDATORY: 4-Phase Gating Protocol (BLOCKING)

CRITICAL: 必须按顺序完成以下四个阶段，并输出每个阶段的结果。跳过任何阶段将导致分析质量下降！

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🧠 THINKING - Phase 4 Entry Gate

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Purpose: 深度理解后再行动，防止基于不完整理解的仓促行动。

⚠️ 你必须输出以下格式的 THINKING 结果：

🧠 THINKING - P4 Entry Gate
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

📌 CORE PROBLEM
评估项目在16个安全域的设计成熟度，识别安全控制缺口

📊 UPSTREAM DATA (从 P1-P3 YAML 读取)
| 指标 | 值 | 来源 |
|------|-----|------|
| P1模块总数 | {实际值} | P1_project_context.yaml → module_inventory.summary.total_modules |
| P2数据流总数 | {实际值} | P2_dfd_elements.yaml → data_flow_traces.summary.total_flows |
| P2数据存储数 | {实际值} | P2_dfd_elements.yaml → data_store_inventory.summary.total_data_stores |
| P3边界数量 | {实际值} | P3_boundary_context.yaml → boundary_context.boundaries 长度 |
| Tech stack | {实际值} | P1_project_context.yaml → project_context.tech_stack |

❓ UNKNOWNS
- 需要通过代码审查确认的安全控制
- 需要KB查询确认的最佳实践

⚠️ RISKS
- 域评估不完整 (16个域未全部覆盖)
- Gap与Module/Flow的追溯链缺失
- 扩展域触发检测遗漏

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
⛔ STOP CHECK
- P1 YAML 已读取? [YES/NO]
- P2 YAML 已读取? [YES/NO]
- P3 YAML 已读取? [YES/NO]
- 上游数据完整? [YES/NO]
- 可以继续PLANNING? [YES/NO]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

⛔ STOP条件: 如果任何 STOP CHECK = NO → 先读取上游数据再继续

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

📋 PLANNING - Sub-task Decomposition

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Step 1: 读取上游数据 (BLOCKING - 必须执行)

# 读取P1-P3 YAML数据 (选择其一)
python scripts/phase_data.py --aggregate --phases 1,2,3 --format summary --root .

# 或直接读取
cat .phase_working/{SESSION_ID}/data/P1_project_context.yaml
cat .phase_working/{SESSION_ID}/data/P2_dfd_elements.yaml
cat .phase_working/{SESSION_ID}/data/P3_boundary_context.yaml

⛔ 如果任何文件不存在或无效 → STOP并返回完成上游Phase

Step 2: 输出子任务表格 (MANDATORY)

⚠️ 你必须输出以下格式的 PLANNING 结果：

📋 PLANNING - P4 Sub-tasks
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

| # | 子任务 | 预期输出 |
|---|--------|----------|
| T1 | 读取P1-P3数据，提取模块/流/边界清单 | 数据结构 |
| T2 | 评估10个核心安全域 (AUTHN→DATA) | 域评估矩阵 |
| T3 | 检测扩展域触发条件 (ext-11→ext-16) | 触发检测 |
| T4 | 评估已触发的扩展域 | 扩展域评估 |
| T5 | 生成Gap清单 (GAP-xxx) | Gap清单 |
| T6 | 写入P4_security_gaps.yaml | PRIMARY输出 |
| T7 | 写入P4-SECURITY-REVIEW.md | SECONDARY输出 |

⛔ PLANNING CHECK
- 子任务已分解? [YES/NO]
- 准备创建 TaskCreate? [YES/NO]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Step 3: TaskCreate for ALL sub-tasks (MANDATORY)

⚠️ 在开始任何实施前，必须执行 TaskCreate 创建所有子任务！

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

⚡ EXECUTION LOOP

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

For each sub-task:

1. TaskUpdate(status: "in_progress")
2. 实施子任务
3. 验证: 输出是否符合预期？
4. If 验证通过: TaskUpdate(status: "completed") → 下一个
5. If 验证失败: 诊断 → 修复 → 重试 (max 3x) → 如仍失败: CHECKPOINT请求用户决策

输出顺序 (CRITICAL):

1. 先写YAML: .phase_working/{SESSION_ID}/data/P4_security_gaps.yaml
2. 后写MD: .phase_working/{SESSION_ID}/reports/P4-SECURITY-REVIEW.md

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🔍 REFLECTION - Completion Verification

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

⚠️ 完成 EXECUTION 后，你必须输出以下格式的 REFLECTION 结果：

🔍 REFLECTION - P4 Completion Check
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

| 检查项 | 状态 |
|--------|------|
| 所有子任务已完成? (TaskList check) | [✅/❌] |
| P4_security_gaps.yaml 存在且有效? | [✅/❌] |
| design_matrix 包含全部16个域? | [✅/❌] |
| 每个Gap有唯一ID (GAP-xxx)? | [✅/❌] |
| coverage_verification 显示100%覆盖? | [✅/❌] |
| input_ref 字段指向 P3_boundary_context.yaml? | [✅/❌] |
| Hook验证通过 (exit 0)? | [✅/❌] |

⛔ COMPLETION GATE
- 所有检查通过? [YES/NO]
- 可以进入P5? [YES/NO]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

⛔ 任何检查失败 → 修复并重新验证，直到全部通过

⚠️ MANDATORY OUTPUT RULES

CRITICAL: Phase 4 requires TWO outputs - a YAML data file AND a Markdown report.

Dual Output Requirement

┌─────────────────────────────────────────────────────────────────────┐
│  PHASE 4 MUST PRODUCE TWO FILES:                                    │
├─────────────────────────────────────────────────────────────────────┤
│                                                                      │
│  1. DATA FILE (PRIMARY - Write First!)                              │
│     Path: .phase_working/{SESSION_ID}/data/P4_security_gaps.yaml    │
│     Purpose: Structured gap data for P5/P6/P7 to read               │
│     Format: Valid YAML with schema_version: "3.0.0 (20260201a)"                   │
│                                                                      │
│  2. REPORT FILE (SECONDARY - Write After Data!)                     │
│     Path: .phase_working/{SESSION_ID}/reports/P4-SECURITY-REVIEW.md │
│     Purpose: Human-readable security design assessment              │
│     Format: Markdown with assessment matrix and gap analysis        │
│                                                                      │
│  ❌ FORBIDDEN: Writing only .md without .yaml                       │
│  ❌ FORBIDDEN: Embedding YAML blocks inside .md as data source      │
│  ✅ REQUIRED: .yaml file is the authoritative data source           │
│                                                                      │
└─────────────────────────────────────────────────────────────────────┘

Required Data Sections in YAML

Validation Gate

Phase 4 CANNOT complete until:

1. .phase_working/{SESSION_ID}/data/P4_security_gaps.yaml exists and is valid YAML
2. design_matrix contains all 16 domains (10 core + 6 extended)
3. Every gap has unique ID (GAP-xxx format) and risk_level assigned
4. Every domain has assessed: true or assessed: false (N/A)
5. .phase_working/{SESSION_ID}/reports/P4-SECURITY-REVIEW.md exists

Coverage Verification (CRITICAL)

P4 MUST verify complete coverage of P1/P2 inputs:

# In P4_security_gaps.yaml - REQUIRED section
coverage_verification:
  # P1 Module Coverage
  p1_modules:
    total_modules: 15          # From P1.module_inventory.summary.total_modules
    modules_assessed: 15       # Modules referenced in gaps or confirmed secure
    coverage_percentage: 100   # MUST be 100%
    unassessed_modules: []     # MUST be empty

  # P2 Data Flow Coverage
  p2_flows:
    total_data_flows: 45       # From P2.dfd_elements.data_flows count
    flows_assessed: 45         # Flows analyzed for security controls
    coverage_percentage: 100   # MUST be 100%
    unassessed_flows: []       # MUST be empty

  # P2 Data Store Coverage
  p2_stores:
    total_data_stores: 8       # From P2.dfd_elements.data_stores count
    stores_assessed: 8         # Stores with DATA domain assessment
    coverage_percentage: 100   # MUST be 100%
    unassessed_stores: []      # MUST be empty

Validation Rules:

• P1 Module Coverage: Every M-xxx from P1 must be referenced in at least one gap's affected_elements OR confirmed as secure
• P2 Flow Coverage: Every DF-xxx from P2 must be assessed for INPUT/OUTPUT/CRYPTO controls
• P2 Store Coverage: Every DS-xxx from P2 must be assessed for DATA domain compliance

BLOCKING: coverage_percentage < 100% for any category

Domain Coverage Verification (GAP-5 FIX)

CRITICAL: P4 MUST verify all 16 security domains are assessed with proper completeness criteria.

Domain Assessment Completeness Matrix

# In P4_security_gaps.yaml - MANDATORY section (GAP-5)
domain_coverage_verification:
  # ============================================================
  # Core Domains (01-10) - MANDATORY Assessment
  # ============================================================
  core_domains:
    assessment_requirement: MANDATORY
    domains_total: 10
    domains_assessed: 10       # MUST equal domains_total
    coverage_percentage: 100   # MUST be 100%

    # Per-domain assessment status
    domain_status:
      AUTHN:
        assessed: true
        assessment_result: Partial    # Yes | Partial | No | N/A
        checks_performed: 6           # From checklist
        checks_passed: 4
        gaps_identified: [GAP-001, GAP-002]
        coverage_detail:
          module_refs: [M-001, M-003]   # Modules assessed for this domain
          flow_refs: [DF-001, DF-005]   # Flows relevant to this domain
        kb_ref: "control-set-01"
      AUTHZ:
        assessed: true
        assessment_result: Yes
        checks_performed: 4
        checks_passed: 4
        gaps_identified: []             # No gaps = fully compliant
        coverage_detail:
          module_refs: [M-001, M-002, M-004]
          flow_refs: [DF-002, DF-003]
        kb_ref: "control-set-02"
      INPUT:
        assessed: true
        assessment_result: Partial
        checks_performed: 5
        checks_passed: 3
        gaps_identified: [GAP-003]
        coverage_detail:
          module_refs: [M-005, M-006]
          flow_refs: [DF-010, DF-011, DF-015]
        kb_ref: "control-set-03"
      OUTPUT:
        assessed: true
        assessment_result: No
        checks_performed: 4
        checks_passed: 1
        gaps_identified: [GAP-004, GAP-005]
        coverage_detail:
          module_refs: [M-007]
          flow_refs: [DF-020, DF-021]
        kb_ref: "control-set-04"
      # ... (CLIENT, CRYPTO, LOG, ERROR, API, DATA)

  # ============================================================
  # Extended Domains (ext-11 to ext-16) - CONDITIONAL Assessment
  # ============================================================
  # AUTHORITY: P1 is the AUTHORITATIVE source for extended domain detection.
  # P4 READS P1.project_context.tech_stack and APPLIES trigger patterns.
  # P4 does NOT redefine detection - it uses P1's detection results.
  # The trigger_patterns below are REFERENCE patterns used to validate
  # P1's detection, not independent detection logic.
  # ============================================================
  extended_domains:
    assessment_requirement: CONDITIONAL
    trigger_detection_source: "P1.project_context.tech_stack"

    # Complete trigger pattern library for all extended domains
    domain_triggers:
      INFRA:  # ext-11: Infrastructure as Code
        trigger_patterns:
          - "Dockerfile"
          - "docker-compose.yml"
          - "docker-compose.yaml"
          - "kubernetes/*.yaml"
          - "k8s/*.yaml"
          - "helm/"
          - ".github/workflows/*.yml"
          - "Vagrantfile"
          - "ansible/"
          - "chef/"
          - "puppet/"
        triggered: true                # Detected in P1
        assessed: true
        assessment_result: Partial
        gaps_identified: [GAP-010]
        trigger_evidence:
          files_found: ["Dockerfile", "kubernetes/deployment.yaml"]
          detection_method: "File pattern match"
      SUPPLY:  # ext-12: Supply Chain
        trigger_patterns:
          - "package.json"
          - "package-lock.json"
          - "yarn.lock"
          - "pnpm-lock.yaml"
          - "requirements.txt"
          - "Pipfile"
          - "Pipfile.lock"
          - "poetry.lock"
          - "go.mod"
          - "go.sum"
          - "Cargo.toml"
          - "Cargo.lock"
          - "pom.xml"
          - "build.gradle"
          - "Gemfile"
          - "composer.json"
        triggered: true
        assessed: true
        assessment_result: No
        gaps_identified: [GAP-011, GAP-012]
        trigger_evidence:
          files_found: ["package.json", "requirements.txt"]
          detection_method: "Dependency manifest detection"
      AI:  # ext-13: AI/LLM Applications
        trigger_patterns:
          - "**/llm/**"
          - "**/model/**"
          - "**/ai/**"
          - "**/ml/**"
          - "langchain"
          - "openai"
          - "anthropic"
          - "huggingface"
          - "transformers"
          - "ollama"
          - "llama"
          - "gpt"
          - "claude"
          - "embedding"
          - "vector"
          - "rag"
        triggered: false               # Not detected
        assessed: false
        assessment_result: N/A
        gaps_identified: []
        trigger_evidence:
          files_found: []
          detection_method: "No LLM/AI patterns detected"
      MOBILE:  # ext-14: Mobile Applications
        trigger_patterns:
          - "*.apk"
          - "*.ipa"
          - "*.aab"
          - "AndroidManifest.xml"
          - "Info.plist"
          - "build.gradle"
          - "Podfile"
          - "*.xcodeproj"
          - "*.xcworkspace"
          - "react-native"
          - "flutter"
          - "ionic"
          - "cordova"
        triggered: false
        assessed: false
        assessment_result: N/A
        gaps_identified: []
        trigger_evidence:
          files_found: []
          detection_method: "No mobile artifacts detected"
      CLOUD:  # ext-15: Cloud Native
        trigger_patterns:
          - "terraform/"
          - "*.tf"
          - "*.tfvars"
          - "pulumi/"
          - "cloudformation/"
          - "*.cdk.ts"
          - "serverless.yml"
          - "sam.yaml"
          - "aws-cdk/"
          - "bicep/"
          - "*.bicep"
          - "gcp/"
          - "azure/"
        triggered: true
        assessed: true
        assessment_result: Partial
        gaps_identified: [GAP-015]
        trigger_evidence:
          files_found: ["terraform/main.tf", "terraform/aws.tf"]
          detection_method: "IaC pattern detection"
      AGENT:  # ext-16: Agentic AI
        trigger_patterns:
          - "**/agent/**"
          - "**/agents/**"
          - "**/tool_use/**"
          - "**/tools/**"
          - "langchain.agents"
          - "crewai"
          - "autogen"
          - "swarm"
          - "autogpt"
          - "babyagi"
          - "function_calling"
          - "tool_calls"
        triggered: false
        assessed: false
        assessment_result: N/A
        gaps_identified: []
        trigger_evidence:
          files_found: []
          detection_method: "No agentic patterns detected"

    # Extended domain summary
    summary:
      domains_triggered: 3           # Domains with detected triggers
      domains_assessed: 3            # MUST equal domains_triggered
      domains_skipped: 3             # N/A domains (not triggered)
      assessment_coverage: 100       # domains_assessed / domains_triggered * 100

  # ============================================================
  # Overall Domain Assessment Summary
  # ============================================================
  overall:
    total_domains: 16
    core_assessed: 10
    extended_triggered: 3
    extended_skipped: 3
    total_assessed: 13               # core + extended_triggered
    total_gaps_by_domain:
      AUTHN: 2
      AUTHZ: 0
      INPUT: 1
      OUTPUT: 2
      # ... etc
    assessment_completeness: 100     # All applicable domains assessed

Domain Assessment Validation Rules

Completeness Criteria Per Assessment Result

BLOCKING Conditions:

• core_domains.coverage_percentage < 100%
• extended_domains.assessment_coverage < 100% (for triggered domains)
• Any gap without domain assignment
• Triggered extended domain with assessed: false

Error Handling

Fallback Strategy: If domain assessment cannot be completed due to insufficient data, mark domain with assessed: partial and document the limitation.

Input Context

← P1/P2/P3: All cumulative findings (from {SESSION_ID}/data/ directory)

Required Input Files:

• P1_project_context.yaml - For tech stack detection (extended domain triggers)
• P2_dfd_elements.yaml - For security control mapping
• P3_boundary_context.yaml - For boundary-aware gap analysis

⚠️ MANDATORY: Query P1-P3 Data Before Analysis

Before starting P4 analysis, LLM MUST execute these queries to obtain upstream data:

# Step 1: Get aggregate summary from P1-P3
python scripts/phase_data.py --aggregate --phases 1,2,3 --format summary --root .

# Step 2: Get P1 modules for coverage tracking
python scripts/phase_data.py --query --phase 1 --type modules --root .

# Step 3: Get P2 DFD elements for security mapping
python scripts/phase_data.py --query --phase 2 --type dfd --root .

# Step 4: Get P3 boundaries for gap context
python scripts/phase_data.py --query --phase 3 --summary --root .

Or read YAML directly:

# PRIMARY sources - ALL REQUIRED
cat .phase_working/{SESSION_ID}/data/P1_project_context.yaml
cat .phase_working/{SESSION_ID}/data/P2_dfd_elements.yaml
cat .phase_working/{SESSION_ID}/data/P3_boundary_context.yaml

CRITICAL: Do NOT assess security domains from memory. MUST read P1-P3 data first!

P3 Boundary Integration:

• Use P3's boundaries[] to identify security control points
• Map gaps to cross_boundary_flows[] for risk context
• Reference data_nodes[] sensitivity when assessing DATA domain
• Consider Model/Agent boundary types for ext-13/ext-16 assessment

Output Context

→ P5: security_gaps {gaps[], design_matrix{}}

Core Analysis Goal

Evaluate project's design maturity across all 16 security domains, identify gaps between current implementation and security best practices.

Knowledge Reference (Progressive Loading)

1. Always load: security-design.yaml (16 domains overview)
2. Per domain: control-set-*.md when assessing that domain
3. For details: reference-set-*.md for specific implementation guidance

Query Commands:

$SKILL_PATH/kb --control authentication       # Domain-specific controls
$SKILL_PATH/kb --stride-controls S            # STRIDE category controls
$SKILL_PATH/kb --control api --full           # Full control details
$SKILL_PATH/kb --control {ext_domain} --full  # Extended domain (may have limited support)

KB Usage Log (MANDATORY per GAP-4 Contract)

CRITICAL: P4 MUST query KB for each assessed security domain per KBQueryContract in assets/contracts/data-model.yaml

Required Queries:

1. --control {domain_code} - For each core domain (AUTHN, AUTHZ, INPUT, etc.)
2. --stride-controls {S|T|R|I|D|E} - For STRIDE-to-domain mapping validation

# In P4_security_gaps.yaml - MANDATORY section (GAP-4 Contract)
kb_usage_log:
  # Query record for auditability
  queries_made:
    - query: "--control authentication"
      timestamp: "2026-01-31T09:15:00Z"
      result_count: 15
      domain_assessed: AUTHN
      findings_informed: [GAP-001, GAP-002]
      cache_hit: false
    - query: "--control authorization"
      timestamp: "2026-01-31T09:15:30Z"
      result_count: 12
      domain_assessed: AUTHZ
      findings_informed: []  # No gaps found - KB confirmed implementation
      cache_hit: false
    - query: "--stride-controls S"
      timestamp: "2026-01-31T09:16:00Z"
      result_count: 8
      domains_assessed: [AUTHN, CLIENT]
      findings_informed: [GAP-003]
      cache_hit: true

  # Domain coverage tracking (MANDATORY)
  coverage:
    core_domains_total: 10          # AUTHN through DATA
    core_domains_queried: 10        # MUST equal core_domains_total
    extended_domains_total: 6       # INFRA through AGENT
    extended_domains_queried: 3     # Best effort - may have limited KB support
    kb_coverage_percentage: 81.25   # (10+3)/(10+6) * 100

  # Error handling
  errors:
    - query: "--control AGENT"
      error_type: "limited_support"
      action_taken: "Used LLM knowledge for ext-16 assessment"
      domain_affected: AGENT

  kb_available: true
  note: "Extended domains (ext-11 to ext-16) may not have full KB support"

Validation Rules (GAP-4 Contract):

• WARNING: kb_coverage_percentage < 50% for core domains (01-10)
• INFO: Extended domains (ext-11 to ext-16) queries are optional
• BLOCKING: None - KB queries are best-effort but strongly recommended

Security Domains (16)

Assessment Process

For each domain:

1. Identify Implementation: What security controls exist in the code?
2. Compare to Standards: How does it compare to OWASP/industry standards?
3. Assess Maturity: Rate as Yes/No/Partial
4. Document Gaps: What's missing or inadequate?
5. Assign Risk Level: High/Medium/Low based on impact

Assessment Matrix Template

Assessment Values:

• Yes: Fully implemented per standards
• Partial: Partially implemented, gaps exist
• No: Not implemented
• N/A: Not applicable to this project

Domain-Specific Checks

AUTHN - Authentication & Session

• Multi-factor authentication available?
• Secure password storage (bcrypt/argon2)?
• Session timeout implemented?
• Session fixation protection?
• Account lockout policy?
• Secure credential recovery?

AUTHZ - Authorization

• Role-based or attribute-based access control?
• Principle of least privilege applied?
• Authorization checks on all endpoints?
• Sensitive operations require re-authentication?

INPUT - Input Validation

• All input validated server-side?
• Allowlist validation preferred?
• File upload restrictions?
• SQL injection prevention?
• Command injection prevention?

OUTPUT - Output Encoding

• Context-aware output encoding?
• XSS prevention measures?
• Content-Type headers set correctly?
• Content Security Policy implemented?

CRYPTO - Cryptography

• TLS 1.2+ for all connections?
• Strong cipher suites only?
• Sensitive data encrypted at rest?
• Proper key management?
• No hardcoded secrets?

LOG - Logging & Monitoring

• Security events logged?
• Log injection prevention?
• Sensitive data masked in logs?
• Log integrity protection?
• Alerting configured?

API - API Security

• API authentication required?
• Rate limiting implemented?
• Input validation on all endpoints?
• CORS properly configured?
• API versioning strategy?

CLIENT - Client-Side Security (05)

• XSS prevention (CSP, output encoding)?
• DOM-based XSS mitigation?
• CSRF protection (tokens, SameSite)?
• Clickjacking prevention (X-Frame-Options)?
• Third-party script isolation?
• Prototype pollution prevention?
• Content Security Policy implemented?
• Subresource Integrity (SRI) for CDN resources?

ERROR - Error Handling (08)

• Generic error messages to users?
• Detailed logging internally?
• Stack traces hidden in production?
• Fail-secure mode implemented?
• Error codes don't leak system info?
• Graceful degradation on failures?

DATA - Data Protection (10)

• PII identified and protected?
• Data classification implemented?
• Data retention policy?
• Secure data deletion?
• Secrets management (no hardcoded)?
• Database credentials secured?

Extended Domain Checks (Conditional - Based on Tech Stack)

Load these checks only when corresponding trigger conditions are detected in P1.

ext-11: INFRA - Infrastructure Security

Trigger: Dockerfile, docker-compose.yml, kubernetes/*.yaml, helm charts

• Container images from trusted registries?
• Base images regularly updated?
• Container runs as non-root?
• Resource limits (CPU/memory) configured?
• Network policies/segmentation?
• Secrets not baked into images?
• Pod security policies/standards?
• Container vulnerability scanning?

ext-12: SUPPLY - Supply Chain Security

Trigger: package.json, requirements.txt, pom.xml, go.mod

• Dependencies vulnerability scanned?
• Lock files (package-lock.json, poetry.lock) committed?
• Private registry security?
• SBOM generated?
• Dependency update policy?
• No known vulnerable dependencies?
• CI/CD pipeline security?
• Artifact signing/verification?

ext-13: AI - AI/LLM Security

Trigger: openai/anthropic API, langchain, LLM model loading

• Prompt injection prevention?
• System prompt protection?
• Output filtering/validation?
• Model access control?
• Training data protection?
• PII not sent to external LLMs?
• Rate limiting on LLM calls?
• Jailbreak detection?

ext-14: MOBILE - Mobile Security

Trigger: iOS/Android project, React Native, Flutter

• Secure local storage (Keychain/Keystore)?
• Certificate pinning implemented?
• Code obfuscation enabled?
• Jailbreak/root detection?
• Biometric authentication?
• Secure IPC (Intent/URL scheme)?
• Binary protections (ASLR, PIE)?
• Debug disabled in release?

ext-15: CLOUD - Cloud Security

Trigger: terraform/*.tf, AWS/Azure/GCP SDK, serverless.yml

• IAM least privilege?
• No wildcards in IAM policies?
• Resources not publicly accessible?
• Security groups properly configured?
• Encryption enabled (at-rest, in-transit)?
• CloudTrail/logging enabled?
• VPC properly configured?
• Secrets in secrets manager (not env vars)?

ext-16: AGENT - Agentic Security

Trigger: langchain/langgraph, crewai/autogen, MCP config, Agent workflows

• Agent goal/intent protection?
• Tool/skill governance (whitelist)?
• Autonomy boundaries defined?
• Human-in-the-loop for critical actions?
• Agent authentication/identity?
• Multi-agent trust boundaries?
• Tool call rate limiting?
• Agent behavior logging/auditing?
• Goal hijack prevention?
• Indirect prompt injection defense?

Gap Documentation Format

File Path: .phase_working/{SESSION_ID}/data/P4_security_gaps.yaml

# P4_security_gaps.yaml - Phase 4 Data Output
schema_version: "3.0.0 (20260201a)"
phase: 4
generated_at: "ISO8601 timestamp"

security_gaps:
  # Summary statistics (REQUIRED)
  summary:
    total_domains: 16
    domains_assessed: 16
    fully_compliant: 8
    partially_compliant: 5
    non_compliant: 2
    not_applicable: 1
    total_gaps: 12
    critical_gaps: 2
    high_gaps: 4
    medium_gaps: 4
    low_gaps: 2

  # Identified security gaps (REQUIRED)
  gaps:
    - id: GAP-001
      domain: AUTHN
      domain_seq: "01"
      description: "Multi-factor authentication not implemented"
      current_state: "Single-factor (password only)"
      expected_state: "TOTP or WebAuthn MFA available"
      impact: "Credential compromise leads to full account takeover"
      risk_level: HIGH  # CRITICAL|HIGH|MEDIUM|LOW
      affected_elements: [P-002, EP-API-001]
      kb_reference: "control-set-01-authentication.md"
      owasp_refs: ["A07:2021-Identification and Authentication Failures"]

    - id: GAP-002
      domain: INPUT
      domain_seq: "03"
      description: "Missing request schema validation"
      current_state: "Basic type checking only"
      expected_state: "JSON Schema validation on all API inputs"
      impact: "Malformed input may bypass security controls"
      risk_level: MEDIUM
      affected_elements: [P-001, DF-001]
      kb_reference: "control-set-03-input-validation.md"
      owasp_refs: ["A03:2021-Injection"]

  # Design matrix - ALL 16 domains (REQUIRED)
  design_matrix:
    # Core Domains (01-10)
    AUTHN:
      seq: "01"
      assessed: true
      rating: Partial  # Yes|Partial|No|N/A
      gaps_count: 2
      risk_level: HIGH
    AUTHZ:
      seq: "02"
      assessed: true
      rating: Yes
      gaps_count: 0
      risk_level: LOW
    INPUT:
      seq: "03"
      assessed: true
      rating: Partial
      gaps_count: 1
      risk_level: MEDIUM
    OUTPUT:
      seq: "04"
      assessed: true
      rating: Yes
      gaps_count: 0
      risk_level: LOW
    CLIENT:
      seq: "05"
      assessed: true
      rating: Partial
      gaps_count: 2
      risk_level: MEDIUM
    CRYPTO:
      seq: "06"
      assessed: true
      rating: Yes
      gaps_count: 0
      risk_level: LOW
    LOG:
      seq: "07"
      assessed: true
      rating: Partial
      gaps_count: 1
      risk_level: MEDIUM
    ERROR:
      seq: "08"
      assessed: true
      rating: Yes
      gaps_count: 0
      risk_level: LOW
    API:
      seq: "09"
      assessed: true
      rating: Partial
      gaps_count: 2
      risk_level: HIGH
    DATA:
      seq: "10"
      assessed: true
      rating: Partial
      gaps_count: 1
      risk_level: MEDIUM

    # Extended Domains (ext-11 to ext-16)
    INFRA:
      seq: "ext-11"
      assessed: true  # or false if not triggered
      triggered: true
      rating: Partial
      gaps_count: 2
      risk_level: MEDIUM
    SUPPLY:
      seq: "ext-12"
      assessed: true
      triggered: true
      rating: Partial
      gaps_count: 1
      risk_level: MEDIUM
    AI:
      seq: "ext-13"
      assessed: true
      triggered: true
      rating: No
      gaps_count: 3
      risk_level: CRITICAL
    MOBILE:
      seq: "ext-14"
      assessed: false
      triggered: false
      rating: N/A
      gaps_count: 0
      risk_level: N/A
    CLOUD:
      seq: "ext-15"
      assessed: true
      triggered: true
      rating: Partial
      gaps_count: 2
      risk_level: HIGH
    AGENT:
      seq: "ext-16"
      assessed: true
      triggered: true
      rating: No
      gaps_count: 4
      risk_level: CRITICAL

  # Extended domains triggered based on P1 tech stack detection
  extended_domains_triggered:
    - { domain: INFRA, trigger: "Dockerfile detected" }
    - { domain: SUPPLY, trigger: "package.json detected" }
    - { domain: AI, trigger: "openai API usage" }
    - { domain: CLOUD, trigger: "terraform/*.tf detected" }
    - { domain: AGENT, trigger: "langchain import" }

Report Template

# P4: Security Design Review

## Executive Summary

- **Domains Assessed**: 16
- **Fully Compliant**: N
- **Partially Compliant**: N
- **Non-Compliant**: N
- **Critical Gaps**: N

## Security Design Assessment Matrix

| Domain | Code | Implementation | Rating | Gaps | Risk | Reference |
|--------|------|----------------|--------|------|------|-----------|
| Authentication | AUTHN | OAuth2, local | Partial | 2 | High | control-set-01 |
| Authorization | AUTHZ | RBAC | Yes | 0 | Low | control-set-02 |
| ... | ... | ... | ... | ... | ... | ... |

## Gap Analysis

### GAP-001: Missing Multi-Factor Authentication

- **Domain**: AUTHN
- **Current State**: Password-only authentication
- **Expected State**: TOTP or WebAuthn MFA
- **Impact**: High - Full account takeover on credential compromise
- **Affected Elements**: P-002, EP-API-001, EP-API-002
- **Recommendation**: Implement TOTP with recovery codes

### GAP-002: ...

## Domain Details

### AUTHN - Authentication & Session

**Implementation Found**:
- OAuth2 with Google/GitHub providers
- Local authentication with bcrypt password hashing
- JWT session tokens

**Gaps Identified**:
1. No MFA support
2. Session timeout not configurable

**Recommendations**:
1. Add TOTP MFA option
2. Implement configurable session timeout

### AUTHZ - Authorization

...

## Summary

[yaml:security_gaps block]

Completion Checklist

Before marking Phase 4 complete:

Data File Requirements:

• .phase_working/{SESSION_ID}/data/P4_security_gaps.yaml exists
• YAML is valid with schema_version: "3.0.0 (20260201a)"
• security_gaps.summary contains all required statistics
• security_gaps.design_matrix has all 16 domains
• Each domain has assessed, rating, gaps_count, risk_level
• Extended domains have triggered field

Gap Documentation:

• All gaps have unique GAP-xxx IDs
• Each gap has domain, domain_seq, description
• Each gap has current_state, expected_state, impact
• Risk levels assigned (CRITICAL/HIGH/MEDIUM/LOW)
• KB references provided for each gap
• OWASP refs mapped where applicable

Report File Requirements:

• .phase_working/{SESSION_ID}/reports/P4-SECURITY-REVIEW.md exists
• Executive summary with statistics
• Complete assessment matrix (16 domains)
• Gap analysis with recommendations
• Domain details section

Validation:

• Hook validation passed (exit 0)
• Count conservation: summary.total_gaps == len(gaps[])

End of Phase 4 Instructions (~200 lines, ~2K tokens)