Tool Family: attack-modules

• Registration source: tools/mcp_tools/attack_modules.py:131 register_attack_module_tools(mcp, *, ctx) — auto-discovered. Holds module dispatch, CVE synthesis, campaign orchestration, and web-attack probes.
• Gate: target-touching tools @require_allowlist(); no-target tools plain (no decorator needed; audit-only is implicit when no target). run_campaign_step re-validates target from state.json via check_targets_allowlist.
• Shared helper: _identify_hash_modes(h) (tools/mcp_tools/attack_modules.py:22-128) — single source for hash_crack_identify + run_hash_crack (cracking family) mapping hash→hashcat mode.

Tools Exported (20)

Tool	Params	Result Shape	Notes
jwt_tamper	target_ip, jwt_token=""	JWT_TAMPER_RESULTS: ...\nHeader: ...\n--- alg:none attack ---\nNone-alg token: ...\n--- Weak HMAC secret brute-force ---\nWEAK SECRET FOUND? ...\n--- HMAC-to-RSA confusion	Auto-discovers JWT via 16 HTTP paths when jwt_token empty; tests alg:none, weak secrets (32-word rockyou/jwt-secrets list), RSA confusion.
ssti_probe	target_ip, port=80	SSTI_PROBE_RESULTS: ... + [DETECTED] engine at path?param or No SSTI	Tests 10 math payloads ({{7*7}}, ${7*7}, etc.) across 18 endpoints × 19 params via raw socket.
graphql_introspect	target_ip, port=80	GRAPHQL_INTROSPECT_RESULTS + [+] endpoint found, types, batching test	14 endpoints, introspection query; batch attack test sends 5 queries.
race_request	target_ip, port=80, endpoint="/api/redeem", concurrent=20 (2..200)	RACE_REQUEST_RESULTS: ... Concurrent: ... Completed in Xs Success: N Failure: M Mixed? [!]	ThreadPoolExecutor(max 50) fires POST {"code":"TEST100"} concurrently; detects TOCTOU via mixed status / multiple successes.
timing_oracle	target_ip, port=80	TIMING_ORACLE_RESULTS + Login/reset timing means + [+] TIMING ORACLE DETECTED	Measures valid vs invalid user on /api/login and existing vs non-existing on /api/reset-password (8 samples, 150ms sleep, 50ms diff threshold).
request_smuggling_probe	target_ip, port=80	REQUEST_SMUGGLING_RESULTS + Baseline / CL.TE / TE.CL / TE.TE sections	Sends raw CL.TE, TE.CL, TE.TE probes, flags smuggling when GPOST leaks or baseline differs >200B.
password_spray	target_ip, port=80, password="Password1"	PASSWORD_SPRAY_RESULTS: ... + per-user SUCCESS	43 usernames (service/cloud defaults), POST /api/login per user, 1.5s delay.
cve_to_exploit_synth	target_ip, cve_id, service_name="", version=""	Writes exploits/<cve>_<svc>.py under attempt dir, returns CVE_TO_EXPLOIT_SYNTH_RESULT: ... PATH: ... SHA256: ... + rendered CVE-family template	Validates IP + CVE format, service/version strict regex (no newline/quote injection); dispatches to 14 _render_*_template (log4j, eternalblue, smbghost, bluekeep, regresshion, xz, activemq, confluence, ivanti, panos, citrix, connectwise, jenkins, joomla, text4shell, php_cgi, http2-rapid-reset, generic). Uses verified cve_to_poc (never invents URL).
hash_crack_identify	hash_value	HASH_IDENTIFY_RESULT: ...\nNTLM: mode 1000\nMD5: ... or Unknown	Delegates to _identify_hash_modes; reports all matching modes (32-hex → NTLM + MD5).
create_attack_plan	target_ip, target_os="", known_cves=""	ATTACK_PLAN_CREATED: ... STEPS: N + serialized plan preview	AttackPlanner.create_plan(target_ip, target_os, known_cves.split(",")) → persists plans/<ip>_plan.json.
get_current_plan	target_ip	CURRENT_PLAN: ... or NO_PLAN_FOUND	AttackPlanner.load_plan(target_ip) → plan.to_dict() preview.
replan	target_ip, failure_reason	REPLAN_RESULT: ... + new/mutated plan	AttackPlanner.replan(target_ip, failure_reason) (via build_replanning_prompt + parse_replan_json).
list_attack_modules	—	`ATTACK_MODULES: N ... - name	services
run_attack_module	module_name, target_ip, options=""	MODULE_RESULT: ... OUTPUT: ...	get_module(name).run(ModuleContext(target_ip, options)) dispatched per module.
craft_exploit	target_ip, service_name, version="", os_hint="", module_name=""	CRAFT_RESULT: ... PATH: ...	Selects module by service/CVE + renders exploit file; similar template routing to synth.
mutate_exploit	script_id, failure_output	MUTATE_RESULT: ... NEW_PATH: ...	ExploitMutator.mutate(script_id, failure_output) — code-aware rewrite loop.
start_autonomous_campaign	target_ip, goal="initial_access", aggression_level="normal"	CAMPAIGN_STARTED: id=... + running background task	Creates AutonomousOrchestrator task, holds strong ref in _running_campaign_tasks + _campaign_orchestrators[campaign_id]; done callback drops refs (tools/mcp_tools/attack_modules.py:9-19, 2239-2410).
get_campaign_status	campaign_id	`CAMPAIGN_STATUS: running	completed
run_campaign_step	campaign_id (async)	CAMPAIGN_STEP_RESULT: ... or BLOCKED: Host ... not in allowlist	Loads campaigns/<id>/state.json, re-validates target_ip from state via check_targets_allowlist before executing one orchestrator step; audit-gated.
stop_campaign	campaign_id	CAMPAIGN_STOPPED: id=... SUCCESS: ...	Signals AutonomousOrchestrator.stop() via _campaign_orchestrators[campaign_id].

Parameters

• Target IPs via validate_target_or_ip; CVE IDs strict; service/version quoted-regex gated for synthesis (newline/quote injection blocked — tests/test_mcp_injection_hardening.py:672-707).
• concurrent 2..200; port 1..65535.

Dependencies

• tools/attack_modules (module registry), tools/attack_planner, tools/exploit_mutator, tools/autonomous_orchestrator, tools/payload_crafter
• tools/validation_utils, tools/kernel/allowlist

Config

• exploit.require_explicit_allowlist, exploit.allowed_targets
• attack_modules.* / autonomous_orchestrator aggression settings

Auditing

• Target-touching: @require_allowlist() + started/completed|blocked records.
• run_campaign_step uses @audit_tool + manual check_targets_allowlist on target from state.json (LLM-influenced path).
• list_attack_modules/mutate_exploit no target, no audit gate needed.
• Secrets in options masked via _mask_secret_content before audit.

Tests

• tests/test_mcp_injection_hardening.py:672 — cve_to_exploit_synth rejects invalid IP/newline/quote
• tests/test_mcp_tool_registration.py — expects list_attack_modules, run_attack_module, create_attack_plan, start_autonomous_campaign, get_campaign_status, run_campaign_step, stop_campaign

Related Docs

• docs/attack-modules.md
• docs/mcp/tool-families/credentials.md (post-exploit chain)