pentest（pentest 测试）

Coordinate penetration testing. Deploy executors, aggregate results, generate reports. Use it when user requests pentesting, security assessment, vulnerability testing, bug bounty hunting.

Workflow

Phase 1: Initialization

1. Gather scope: Target URL, restrictions, testing window
2. Create engagement folder: outputs/{engagement-name}/

Phase 2: Reconnaissance (Read reference/RECONNAISSANCE_OUTPUT.md to get outputs format)

1. Select reconnaissance tools based on asset type (domains, web apps, APIs, network, cloud)
2. Run tools in parallel using pentest-executor agents
3. Generate asset-specific inventory files (JSON format per asset type) using the defined output format
4. Generate the final reconnaissance_report.md file using the defined output format

Phase 3: Planning & Approval (MANDATORY)

1. Analyze reconnaissance findings from the outputs/{engagement}/reconnaissance/ and the reports.md files
2. Create test plan: Executors to deploy, attack surface justification, testing approach
3. Present plan to user via AskUserQuestion
4. Get explicit approval if not already approved in the first phase: "Approve plan?", "Modify executors?", "Cancel?"
5. CRITICAL: Do NOT proceed to Phase 4 without user approval

Phase 4: Vulnerability Testing

1. Deploy approved executors in parallel (single Task call with run_in_background=True)
2. Monitor progress: Periodic TaskOutput(block=False)
3. Recursive spawning: New discoveries trigger new executors (ask approval if major change)

Phase 5: Aggregation

1. Collect findings from all executors
2. Deduplicate (same vuln + location = duplicate)
3. Identify exploit chains
4. Calculate severity metrics

Phase 6: Reporting (Read reference/FINAL_REPORT.md - includes DOCX conversion)

1. CRITICAL: Create folder structure FIRST: report/ and processed/ with subdirectories
2. Move ALL working files to processed/: reconnaissance/, findings/, activity/ → processed/
3. Move ALL intermediate files: ANY.md files, drafts, analysis → processed/intermediate-reports/
4. Generate markdown report: Use reference/FINAL_REPORT.md template → processed/intermediate-reports/pentest-final-report.md
5. REQUIRED: Generate.docx: Run pandoc command → report/Penetration-Test-Report.docx (cover page, TOC, body, appendix section)
6. Optional: Generate PDF: If LaTeX available → report/Penetration-Test-Report.pdf, else skip (DOCX is primary deliverable)
7. Copy referenced evidence: Organize by finding → report/appendix/finding-{id}/
8. Create report README: Document deliverables in report/README.md
9. VERIFY CLEAN STRUCTURE: ls -la outputs/{engagement}/ shows ONLY report/ and processed/
10. CRITICAL: NO intermediate files in root or report/ - Everything goes to processed/ except final deliverables (.docx,.json, README, appendix/)

What This Skill Does

1. Attack Index - References 50+ attack types with documentation paths
2. Methodology Frameworks - PTES, OWASP WSTG, MITRE ATT&CK, Flaw Hypothesis
3. Coordination - Guides pentester agent to deploy specialized attack agents
4. Documentation - PortSwigger labs, cheat sheets, quickstarts per attack

Execution: Delegated to specialized agents (SQL Injection Agent, XSS Agent, SSRF Agent, etc.)

Attack Categories

9 categories, 50+ attack types:

• Injection (6) | Client-Side (6) | Server-Side (6)
• Authentication (4) | API Security (4) | Web Applications (6)
• Cloud & Containers (5) | System (3) | IP Infrastructure (8) | Physical & Social (1)

See reference/ATTACK_INDEX.md for complete list with agent mappings.

Reconnaissance Asset Types

Five asset-specific output formats:

• Domains - Subdomains, DNS records, tech stack per subdomain
• Web Applications - Endpoints, forms, tech stack, cookies, JS analysis
• APIs - REST/GraphQL/WebSocket, auth methods, Swagger docs
• Network Services - Port scans, service versions, CVE candidates
• Cloud Infrastructure - S3 buckets, EC2 instances, security groups

See reference/RECONNAISSANCE_OUTPUT.md for complete format specifications and JSON schemas.

Final report

See reference/FINAL_REPORT.md for complete format specifications of the final report.

Output Structure

Complete folder organization (See reference/OUTPUT_STRUCTURE.md for details):

outputs/{engagement-name}/
├── report/                         # Complete deliverable package (3 files + appendix)
│   ├── Penetration-Test-Report.docx     # Main report (includes Referenced Files section)
│   ├── Penetration-Test-Report.pdf      # Optional PDF export
│   ├── pentest-report.json              # Machine-readable export
│   └── appendix/                        # Referenced evidence only
│       ├── finding-001/
│       ├── finding-002/
│       └── reconnaissance-summary.json
└── processed/                      # All working/testing artifacts
    ├── reconnaissance/             # Phase 2 outputs
    │   ├── inventory/
    │   ├── analysis/
    │   └── reconnaissance_report.md
    ├── findings/                   # Phase 4 raw findings
    │   └── {finding-id}/
    ├── activity/                   # NDJSON logs
    │   └── {executor-name}.log
    ├── helpers/                    # Testing utilities
    ├── test-frameworks/            # Testing scripts
    └── intermediate-reports/       # Drafts, markdown source, etc.

Critical: report/ = 3 files max + appendix/ subfolder. ALL intermediate files go to processed/.

Methodologies

PTES - 7-phase engagement lifecycle OWASP WSTG - 11 testing categories MITRE ATT&CK - TTP mapping across 14 phases Flaw Hypothesis - Stack analysis → Predict → Test → Generalize → Correlate

Integration

• /authenticating - Authentication testing workflows
• /ai-threat-testing - LLM vulnerability testing
• /domain-assessment - Domain reconnaissance
• /web-application-mapping - Web app reconnaissance
• /cve-testing - CVE vulnerability testing

Critical Rules

Testing Rules

• Orchestration only - Never execute attacks directly
• Delegate execution - Deploy specialized agents for testing
• Documentation index - Reference attack folders for techniques
• Working PoCs required - Specialized agents must provide evidence
• Activity logging - All agents log actions to NDJSON activity logs

Output Organization Rules (PHASE 6 - CRITICAL)

• Two-folder structure ONLY: report/ (final deliverables) and processed/ (working files)
• NO files in engagement root: Everything must be in report/ or processed/
• Report folder contents: ONLY pentest-report.json, README.md, appendix/ folder (max 2-3 files + 1 folder)
• ALL intermediate files → processed/:.md files, drafts, analysis, summaries, checklists
• Reconnaissance → processed/reconnaissance/: ALL recon outputs
• Findings → processed/findings/: ALL raw finding details
• Activity logs → processed/activity/: ALL NDJSON logs
• Test frameworks → processed/test-frameworks/: SQL injection, command injection scripts
• Markdown reports → processed/intermediate-reports/: pentest-final-report.md, executive-summary.md, etc.
• VERIFY CLEAN: Before completing Phase 6, run ls -la outputs/{engagement}/ - must show ONLY report/ and processed/

Payload Reference

Each attack category contains a payloads/ subdirectory with curated PATT payloads (<200 lines/file).

• Browse: attacks/<group>/<category>/payloads/
• On-demand fetch: patt-fetcher agent → "<category name>"
• Standard: PATT_STANDARD.md — follow this for future curation sessions
• P1/P2 stubs: stub files with priority: high/medium — ready to fill next session