web

name: web description: "Web application exploitation — the primary category skill for all web-based attacks. This is a routing skill: read this first to identify the attack type, then load the appropriate specialized sub-skill for detailed procedures. Covers 11 technique areas across injection, file access, authentication, and API exploitation." allowed-tools: Bash Read Write metadata: subdomain: execution when_to_use: "web exploit, web application, http, https, web vulnerability, injection, web attack, web service, api, cookie, session, authentication bypass, web shell, form, parameter, query string, POST, GET, request, response, web server, apache, nginx, flask, django, express, php, java web, asp.net, ruby on rails, spring, node.js, deserialization, SQL injection, sqlmap, SSTI, template injection, SSRF, IDOR, command injection, RCE, xss, cross-site scripting, xxe, xml, lfi, path traversal, file upload, graphql" tags: web-application, exploitation, injection mitre_attack: T1190, T1059, T1203

Web Application Exploitation — Category Overview

This is a routing skill. It helps you identify the correct attack technique, then directs you to the specialized sub-skill with full exploitation procedures.

Attack Technique Routing

Match the target's characteristics to the right sub-skill:

Quick Detection — Which Attack Type?

Run these probes to identify which sub-skill to load:

# 1. SQL Injection — single quote error
curl -s 'https://<TARGET>/page?id=1%27' -o /dev/null -w '%{http_code}'

# 2. SSTI — math evaluation
curl -s 'https://<TARGET>/page?input={{7*7}}' | grep -o '49'

# 3. XSS — reflection check
curl -s 'https://<TARGET>/search?q=<script>test</script>' | grep '<script>test'

# 4. LFI — path traversal
curl -s 'https://<TARGET>/file?name=../../../etc/passwd' | grep 'root:'

# 5. Command Injection — command chaining
curl -s 'https://<TARGET>/ping?host=127.0.0.1;id' | grep 'uid='

# 6. SSRF — localhost access
curl -s 'https://<TARGET>/fetch?url=http://127.0.0.1/' -o ssrf_test.txt

# 7. XXE — XML acceptance
curl -s 'https://<TARGET>/api' -H 'Content-Type: application/xml' -d '<?xml version="1.0"?><test>hello</test>'

# 8. GraphQL — endpoint discovery
curl -s 'https://<TARGET>/graphql' -H 'Content-Type: application/json' -d '{"query":"{ __typename }"}'

Decision Flow

Web target identified?
├── Has query parameters → Try sqli.md, then ssti.md
├── Has file/path parameters → Try lfi.md
├── Has URL/fetch parameters → Try ssrf.md
├── Has search/input reflection → Try xss.md
├── Has file upload form → Try file-upload.md
├── Accepts XML input → Try xxe.md
├── Has /graphql endpoint → Try graphql.md
├── Has ping/exec functionality → Try command-injection.md
├── Has serialized data (cookies, POST) → Try deserialization.md
├── Has object IDs in URLs → Try idor.md
├── Tag/hint includes race/concurrent/toctou → Try race-condition.md
├── Tag/hint includes smuggling/desync/hrs OR multi-proxy stack detected → Try smuggling.md
├── Tag/hint includes crypto/cipher/oracle/captcha OR base64 cookie %16==0 with distinct pad-vs-auth errors → Try crypto.md
├── Tag includes business_logic / privilege_escalation / 2fa_bypass / auth_bypass → Try business-logic.md
├── Tag includes cve OR challenge description mentions a specific CVE/plugin vulnerability → Load cve.md
└── Unknown → Run quick detection probes above

CVE / Known-Exploit Methodology

For CVE-class exploitation, load cve.md (full procedure includes fingerprinting → cve_lookup tool call → cve_poc_lookup → PoC adaptation → flag sweep). Once the CVE class is identified (SQLi/LFI/RCE/auth bypass), also load the matching attack-class sub-skill (sqli.md, lfi.md, command-injection.md, etc.) for the continuation primitive.

Payload Pacing — Cross-Technique Rule

After 3 payloads of the same bypass class against the same endpoint all produce the same-class negative result (all blocked / all 404 / all no-reflection), pivot to the next bypass class from the sub-skill's cookbook. Do not iterate further within the exhausted class.

A "class" is the bypass mechanism, not the payload string. Three payloads that differ only in the injected JS expression (e.g. alert(1) → alert(2) → confirm(1)) are ONE class.

After exhausting 3 bypass classes on the same endpoint: the endpoint is hardened. Write exploit/PIVOT.md and switch to a different endpoint or vulnerability class from recon SUMMARY.md.

Sub-Skill Loading Rule

ALWAYS load the specialized sub-skill before executing any attack. This overview provides routing only — the sub-skills contain the actual exploitation procedures, tool commands, bypass techniques, and decision trees needed for successful exploitation.

The agent loop (intake handoff verification, confirm-gate phase, verification, silent-stall recovery, three-strike rule) is already loaded into your system prompt at agent boot — no load_skill call needed for it.