anjian-audit

AnJian authorized website assessment

Use this skill when the user asks to assess, audit, scan, baseline, or retest a website with AnJian.

One-URL default workflow

When the user gives a URL and asks to assess, audit, scan, test, or generate a security report:

• Call anjian_quick_assess immediately with only that URL.
• Do not call doctor, capability planning, assessment creation, or ask the old multi-question intake first.
• The Python tool enforces a reusable exact-target authorization that the user must have registered interactively in a local terminal.
• If it returns authorization_required or authorization_expired, no network action occurred. Show only the returned one-time setup command.
• If it completes, summarize the stages and give the Markdown report_path.

MCP and Claude must never create or silently broaden a reusable authorization profile.

Workflow

1. Prefer anjian_quick_assess for the ordinary one-URL request.
2. Use the manual tools below only when the user explicitly asks for a custom workflow or supplies extra evidence/source paths.
3. Call anjian_create_assessment only for a custom assessment not covered by a registered quick profile.
4. Use passive only when the user asked for public context without target connections.
5. Use Burp XML/HAR and PCAP only when the user provides a local file path; use source scan only for an explicitly authorized local source directory.
6. Never unlock active mode or register a target from MCP. That is a terminal-only user action.
7. Read the consolidated observations. Treat suspected as unverified; multiple sources increase confidence but do not confirm a vulnerability.
8. Only call anjian_validate_observation after documenting a reproducible validation rationale. If evidence is insufficient, leave it suspected.
9. State skipped/blocked tools, untested areas, scope, and limitations.

Hard safety rules

• Never expand from the authorized host to discovered subdomains unless allow_subdomains was explicitly recorded; never expand to parent domains, redirects, IP neighbors, or third-party services.
• Never use brute force, denial of service, exploit, persistence, credential testing, data extraction, SQLMap database enumeration, FFUF recursion, or arbitrary tool arguments.
• Never start live packet capture through MCP. Analyze only a user-selected capture file and never persist packet payloads.
• Never describe an automated match as a confirmed vulnerability.
• Never upload client evidence to an online service unless the client explicitly authorized it.