Shannon Pentest Methodology

Principle: No Exploit, No Report. Only report vulnerabilities that have been practically exploited or validated.

Methodology derived from Shannon (AGPL-3.0, Keygraph), adapted for Kimi CLI.

5-Phase Pipeline

Execute phases sequentially. Each phase saves a deliverable to .shannon/deliverables/. Do not skip phases unless the user explicitly requests a subset.

Pre-Recon → Recon → Vulnerability Analysis → Exploitation → Reporting

Phase 1: Pre-Reconnaissance

Goal: External footprinting + initial source-code mapping.

1. Read references/pre-recon.md.
2. If target is a live web app/API:
• Run nmap, subfinder, whatweb via Bash.
• Capture open ports, subdomains, tech stack fingerprints.
3. If source code is available:
• Grep for hardcoded secrets (API_KEY, SECRET, password, token).
• Identify entry points (controllers, routes, handlers).
4. Save deliverable: python .claw/skills/shannon-pentest/scripts/save-deliverable.py --type PRE_RECON --content "..."

Phase 2: Reconnaissance

Goal: Browser automation + attack-surface mapping.

1. Read references/recon.md.
2. Use Playwright (via Bash or configured MCP) to:
• Navigate the application, authenticate if credentials/TOTP are provided.
• Map all forms, links, API endpoints, and URL parameters.
• Capture screenshots of critical flows.
3. Correlate dynamic findings with static code analysis.
4. Save deliverable: --type RECON

Phase 3: Vulnerability Analysis

Goal: Identify concrete vulnerability hypotheses.

1. Read references/vuln-analysis.md and references/schemas.md.
2. Launch up to 5 parallel Agent subtasks (one per vulnerability class):
• injection-vuln (SQLi, Command Injection, SSTI, etc.)
• xss-vuln (Stored/Reflected/DOM XSS)
• auth-vuln (Broken Authentication, Session Management)
• ssrf-vuln (Server-Side Request Forgery)
• authz-vuln (Broken Access Control, IDOR)
3. Each agent must output a structured JSON queue matching the schema in references/schemas.md.
4. Save deliverables: injection_analysis_deliverable.md, xss_analysis_deliverable.md, etc. (use --type INJECTION_ANALYSIS, XSS_ANALYSIS, etc.)

Phase 4: Exploitation

Goal: Validate hypotheses from Phase 3.

1. Read references/exploitation.md.
2. For each non-empty vulnerability queue from Phase 3, launch a focused Agent subtask to attempt practical exploitation.
• Use Bash to run exploit payloads (curl, playwright scripts, custom Python).
• Capture evidence: HTTP responses, screenshots, source-code diffs.
3. If a hypothesis cannot be exploited after reasonable effort, mark it unconfirmed and drop it from the report.
4. Save deliverables: *_exploitation_queue.json and exploitation_evidence/ folder.

Phase 5: Reporting

Goal: Compile only validated findings into a professional report.

1. Read references/reporting.md.
2. Aggregate all exploitation evidence from .shannon/deliverables/.
3. Generate a markdown report with:
• Executive Summary
• Risk ratings (CVSS or Critical/High/Medium/Low)
• Step-by-step reproduction instructions
• Impact analysis
• Remediation guidance
4. Save deliverable: --type REPORT (saved as report.md).

Helper Scripts

TOTP Generation

For 2FA/MFA login flows:

python .claw/skills/shannon-pentest/scripts/generate-totp.py --secret BASE32SECRET

Save Deliverable

Always use the provided script to persist phase outputs:

python .claw/skills/shannon-pentest/scripts/save-deliverable.py --type <TYPE> --content "..."
# Or from file:
python .claw/skills/shannon-pentest/scripts/save-deliverable.py --type <TYPE> --file-path /path/to/content.md

Valid types: PRE_RECON, RECON, INJECTION_ANALYSIS, XSS_ANALYSIS, AUTH_ANALYSIS, SSRF_ANALYSIS, AUTHZ_ANALYSIS, EXPLOITATION_EVIDENCE, REPORT

Workspace Conventions

• All deliverables live in ./.shannon/deliverables/ (created automatically).
• Use git init inside .shannon/deliverables/ if you want checkpoint/rollback support.
• Maintain a scratchpad directory .shannon/scratchpad/ for temporary files, screenshots, and raw tool output.

Error Handling

• If a tool (nmap, playwright, etc.) is not installed, attempt to install it or ask the user.
• If an Agent subtask fails, retry once with a more focused prompt. If it still fails, note the failure in the deliverable and continue.
• Never halt the entire pipeline because one vulnerability-class agent failed.