Fast. Precise. Intelligent.a reconnaissance engine for authorized engagements

The attack engine that stops where the proof stops.

nice_scan plans reconnaissance and exploitation actions, chains confirmed findings into deeper probes, and reports what actually happened. Every result is classified on execution: verified probes earned it on the wire, observed labels a signal the target itself showed, hypothesis is a chain deliberately not counted.

165
fingerprints
14
attack actions
2
held behind --destructive
10
chain patterns
0
claims without evidence

Action inventory

What the engine runs when it reaches a target. Two rows are consent-gated: command injection and web-shell upload never fire without --destructive.

Recon

  • Passive reconcrt.sh and Wayback feeds mapped before the first packet touches the target
  • Crawlbreadth-first discovery of pages, forms and script files, same-origin only — plus route literals harvested from first-party JS bundles
  • Fuzzknown paths and parameter sets probed; 404-baseline aware so soft-404 noise stays out — OpenAPI specs and JS-declared routes widen the surface
  • Port scanconcurrent sweep across the host's interface addresses

Probes — detection

  • XSScanary-tagged reflection; a claim needs an executable context the baseline cannot echo
  • SQL injectionboolean-differential and time oracle gated 2-of-2; errors confirmed against a neutral round
  • Local file inclusionstructural file records diffed against a no-payload baseline, then read for content
  • GraphQLintrospection and field exposure checked against the schema the endpoint reveals
  • S3 bucketscandidate enumeration and readable-bucket confirmation, evidence above name collisions

Exploitation & signals

  • Login brutedifferential verdict against a junk-credential baseline — the page lying about success is filtered out
  • JWT forgealg=none and common-secret signing; access requires more than an accepted token shape
  • OOB callbacksblind SSRF/SSTI/XSS sink with a reachability verdict — an unreachable sink says so instead of promising

Consent-gated do not fire without --destructive

  • Command injectionquoted-marker probe; uid= output from a real execution — nothing else — releases the verdict
  • Web shell uploadupload accepted is reported as medium; a deployed shell is claimed only when its marked probe executes

The verdict system

The report's structure forces honesty: findings can only be filed by execution. Signals and untested chains are recorded, but they never inflate the risk score.

Findings example posture
VERIFIED
SQL injection — login.phpboolean oracle ×2 · 142 records extracted to sql_dump/
VERIFIED
Local file inclusion — page=passwd record outside baseline · 12 files to lfi_files/
OBSERVED
Admin panelkeyword match on a dashboard — no access attempted, nothing forged
HYPOTHESIS
CORS → XSS chainreflected-origin echo unproven as executable — recorded, weighted 0

Install

Fail-closed: the installer resolves one tag, verifies SHA256 and the pinned GPG key, and aborts with nothing installed when any check refuses.

# Linux / macOS — fail-closed installer
curl -sfL https://raw.githubusercontent.com/NICE-DEV226/nice-Scan/main/scripts/install.sh | sh

Air-gapped? Ship the release archive + checksums.txt (+ .sig) on any media, then install.sh --archive nice_scan_<ver>_linux_amd64.tar.gz — same fail-closed verification, no network needed.

Usage

Four frames for the same engine — pick the one the engagement asks for.

  1. nice_scan hack example.com -R report.html

    Full engagement. Recon, probes, chaining and extraction in one run, with the verdict-ordered report at the end.

  2. nice_scan scan example.com -i

    Live dashboard. Streaming findings and progress while the scan talks, for monitoring a longer sweep.

  3. nice_scan exploit https://example.com --login --idor 1-100

    Targeted exploitation. Run selected modules against endpoints you already know. --register is refused without consent.

  4. nice_scan hack target.com --timeout 30s --destructive

    Consented run. Time-boxed, and with the flag only the operator types when written permission covers state changes — command and shell modules included.

  5. nice_scan scan staging.example.com --sarif results.sarif --fail-on-severity high

    Pipeline gate. SARIF 2.1.0 feeds GitHub Code Scanning, GitLab and Azure DevOps; the severity flag fails the build. A composite GitHub Action ships at the repo root.

  6. nice_scan scan app.example.com --openapi spec.yaml \
      --identity-a alice@corp --identity-b bob@corp \
      --session-a "Cookie: sid=A" --session-b "Cookie: sid=B"

    Grey-box differential. An OpenAPI spec guides probes past the SPA shell, and two sessions replay object URLs — a BOLA finding fires only on a proven identity marker, never on two 200s.

Trust the binary before you run it

Nothing here is marketing posture. Items that are not shipped are marked as such.

Checksums + signature

  1. Every release publishes checksums.txt; GPG-signed when signing keys are configured.
  2. The installer pins fingerprint 4CE4…A4DB from keyserver.ubuntu.com.
  3. Mismatched key or unmatched checksum aborts, nothing executes.
sha256sum --check checksums.txt --ignore-missing
gpg --verify checksums.txt.sig checksums.txt

Provenance & SBOM

  1. GitHub build-provenance attestations are emitted per release — releases cut after 2026-10-24, nothing retroactive.
  2. Cosign keyless bundles arrive alongside the manifest.
  3. An SPDX SBOM ships in the release assets.
gh attestation verify nice_scan -R NICE-DEV226/nice-Scan

Audit the code instead

  1. go install anchors strictly to a resolved tag.
  2. Cloning a pinned tag removes every trust intermediary at once.
git clone --branch v0.3.7 https://github.com/NICE-DEV226/nice-Scan.git