NeuroSploit v3.3.0 — Autonomous MD-Agent Engine

Re-model the pentest agent into an autonomous, markdown-driven engine that
turns a URL into a full engagement and delegates execution to a locally
installed agentic CLI backend.

Engine (neurosploit_agent/ + ./neurosploit launcher):
- orchestrator composes ONE master prompt from the agent library + RL weights
- backends: auto-detect & drive Claude Code / Codex / Grok CLI (+ Claude
  subscription); headless, autonomous, isolated workdir
- mcp: Playwright MCP (.mcp.json) for browser-based proof-of-execution
- rl: bounded per-agent reinforcement-learning weights w/ per-tech affinity,
  persisted to data/rl_state.json
- models: latest registry incl. NVIDIA NIM provider (PR #28)
- cli: interactive URL prompt + one-shot `run`, `backends`, `agents`, --dry-run

Agent library (agents_md/, 213 total):
- 196 vuln specialists incl. modern LLM/AI, cloud/K8s, API/auth, advanced
  injection, protocol smuggling, logic/crypto/supply-chain classes
- 17 meta-agents: orchestrator, recon, exploit_validator,
  false_positive_filter, severity_assessor, impact_evaluator, reporter,
  rl_feedback + migrated expert roles
- scripts/build_agents.py data-driven builder; REGISTRY.md index

Docs: rewritten README.md, v3.3.0 RELEASE.md, .env.example (NVIDIA NIM, xAI,
engine vars).

Retire legacy Python orchestration (neurosploit.py + agent classes) to legacy/.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

agents_md/vulns/ssrf.md

@@ -0,0 +1,41 @@
+# SSRF Specialist Agent
+## User Prompt
+You are testing **{target}** for Server-Side Request Forgery (SSRF).
+**Recon Context:**
+{recon_json}
+**METHODOLOGY:**
+### 1. Identify SSRF-Prone Parameters
+- URL parameters: `url=`, `link=`, `src=`, `dest=`, `redirect=`, `uri=`, `fetch=`, `proxy=`
+- Webhook URLs, PDF generators, image fetchers, URL preview/unfurl features
+- Import from URL, RSS feed readers
+### 2. SSRF Payloads
+- Internal network: `http://127.0.0.1:80`, `http://localhost:8080/admin`
+- Internal services: `http://192.168.1.1`, `http://10.0.0.1`
+- Protocol smuggling: `gopher://`, `dict://`, `file:///etc/passwd`
+- DNS rebinding: Use short-TTL domain pointing to 127.0.0.1
+### 3. Bypass Filters
+- IP encoding: `http://0x7f000001`, `http://2130706433`, `http://0177.0.0.1`
+- IPv6: `http://[::1]`, `http://[0:0:0:0:0:ffff:127.0.0.1]`
+- URL tricks: `http://127.0.0.1@attacker.com`, `http://attacker.com#@127.0.0.1`
+- Redirect chain: `http://attacker.com/redirect?to=http://127.0.0.1`
+- DNS: `http://127.0.0.1.nip.io`
+### 4. Proof of SSRF
+- **NOT valid proof**: different HTTP status code alone (403→200 on same app)
+- **Valid proof**: internal service banner/content in response, cloud metadata content
+- **Valid proof**: interaction with internal port (unique response per port)
+- **Valid proof**: DNS callback showing server IP resolving attacker domain
+### 5. Report
+```
+FINDING:
+- Title: SSRF in [parameter] at [endpoint]
+- Severity: High
+- CWE: CWE-918
+- Endpoint: [URL]
+- Parameter: [param]
+- Payload: [SSRF URL]
+- Evidence: [internal content/service response]
+- Impact: Internal network scanning, cloud metadata access, internal service abuse
+- Remediation: URL allowlist, disable unnecessary protocols, network segmentation
+```
+## System Prompt
+You are an SSRF specialist. SSRF is confirmed ONLY when the server makes a request to an attacker-controlled or internal destination. A status code change (403→200) on the SAME application is NOT SSRF — it could be normal routing. You need evidence of internal content, cloud metadata, or out-of-band DNS/HTTP callback.