NeuroSploit v4.4.0

No PR reviews yet. Run neurosploit pr owner/repo 123 or use the MCP server.

Target Authentication

Credentials used so agents test as a logged-in user. Kept only for this session.

Auth header

Sent to the CLI as an ephemeral creds file.

Named roles (multi-identity access-control testing)

Two or more roles enable IDOR/BOLA/BFLA cross-role testing.

Credentials File

creds.yaml path (overrides target auth above)

An explicit file on disk — see neurosploit-rs/creds.example.yaml for the schema.

API collection (Postman / OpenAPI / Swagger / HAR)

Imports endpoints as known surface — file or directory of .json files.

Capability Token

Capability token

Signed grant for this engagement: who authorized it, which hosts, which environment, until when. Verified by the harness. Mint with neurosploit capability issue.

Additional authorized hosts

Without this the engagement is authorized against the target and nothing else.

Scoping & Guardrails enforced in code

The hard scope is the boundary: a request whose host is not listed is refused before it is sent. Exclusions always win. A capability token still caps all of this.

Hard scope — the allowlist

Exact host · *.wildcard · CIDR · https://host/path prefix. Empty = nothing extra enforced.

Exclusions — always win

Refused even if the allowlist would cover them.

Observe-only

May be looked at (recon) but never attacked.

State-changing methods

Allow DELETE / PUT / PATCH

Off by default — a scan should not change the target's state.

Test accounts

Allow account creation

Max accounts

0 = unlimited.

Requests / min

Whole engagement. 0 = unlimited.

Forbidden payloads

Substrings NEVER acceptable. Extends built-in defaults.

Notes — guidance, NOT enforced

Context passed to agents. Prose is not a control.

Environment & Policy

Environment

Scales every risk score.

Policy profile

OT blocks writes, disruptive actions, fuzzing and exploit payloads over industrial protocols.

Egress & Out-of-band

Transport (how traffic reaches the target)

An internal target with no transport is refused, not tested.

Out-of-band domain

Without it blind SSRF/XXE/RCE can only be reported as leads.

OOB HTTP listener

Where callbacks land.

OOB DNS listener

Only if the zone is delegated here.

Inbound SMS (OTP & rate-limit work)

A throttling claim counts delivered messages carrying distinct codes, not HTTP 200s.

Engagement name

Identifies this engagement in the sidebar and run history — required.

What are you testing?

Pick the engagement type — this decides which CLI subcommand runs underneath.

Target URL

The application, host, or endpoint to test.

Model

Pick a provider and model from the harness's live catalog.

Provider

Model

Auth mode

Uses the API key set in Auth & Keys for this provider.

Playwright MCP (browser tool access, subscription backends only)

Run settings

Votes

Models that must agree before a finding is kept.

Chain depth

How far findings get chained into attack paths.

Max agents

Concurrent agents. Empty = harness default.

Recon intensity

Deeper recon finds more surface but costs more tokens.

Reasoning budget

Optional. Left on unlimited, the run behaves exactly as it always has — full depth, no cap.

🔬 Research mode — hunt a novel, CVE-reportable bug (whitebox/greybox)

Known-CVE de-dup + patch-diff variant analysis. Best with a source repo; reports only genuinely new issues (or a concrete patch bypass).

⚡ Quick mode — short, low-cost test

Economy preset: 1 voter, 1 chain round, light recon, ≤6 agents, eco budget. The big token saver. Wins over the settings below.

Budget mode

How compute is spent, not how much there is.

Token limit

Hard ceiling for the run. Empty or 0 = none.

Order

Map everything first, or chase a lead as it appears.

Sample / route

Requests per endpoint family (/api/users/{id} is sampled, not enumerated).

Tooling & assurance

Optional. Route through a proxy, run in a container, re-check every PoC, and frame findings against a compliance framework.

Intercepting proxy

The harness and agent commands route through it. Full HTTPS interception needs one of the tools (own tunnels TLS).

Sandbox

Runs attack commands off the host, with the Kali toolbox. Needs docker or podman.

PoC re-validation

Re-run every PoC; demote what no longer reproduces

TypeSafe (System One)

Additional confirmation strategy (Choice/Noul over real replays). Off runs the identical pipeline so you can compare with/without.

Compliance mapping

PCI-DSS v4.0

HIPAA Security Rule

SOC 2 (Trust Services Criteria)

GDPR (data-protection articles)

Maps confirmed findings onto control requirements in the report. Indicates gaps for an assessor — never a compliance verdict.