使用方法
- 复制安装命令,在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/security-talent/bugskill-ai --skill otp-bruteforce-testing开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。
OTP Brute-Force Testing
Validate whether an OTP verification endpoint can be brute-forced: unlimited attempts, no lockout, long TTL, weak entropy, or a response oracle that reveals the correct code. Targets must be within the scope of the engagement you are authorized to test.
When to use
- Phone/email verification flows (e.g., "add phone number to profile")
- MFA / 2FA challenge steps and login-by-OTP
- Password reset and account recovery
- Transaction confirmation (payments, withdrawals, admin actions)
- Any endpoint that accepts a 4-8 digit numeric code and returns success/failure
Core attack pattern (reference: HackerOne #3265780, CoinMate.io)
- Trigger an OTP challenge, submit any code, and **intercept the verification
request** (Burp proxy).
- Brute-force the OTP space (100000–999999) with Burp Intruder or the bundled
script.
- Detect the valid OTP via a response oracle — in the CoinMate case, the
response containing a valid OTP had a Content-Length of 961 bytes vs. ~150 for failures.
- Replay the original intercepted request with the recovered OTP to
complete the operation (the phone number is added to the profile).
- Chain the impact: attacker-controlled phone number on the account → the
number receives login/reset OTPs → account takeover.
Phase 0 — Recon the OTP flow
- Map all OTP-issuing and OTP-verifying endpoints; note parameter names
(otp, code, token, verificationCode), length (4/6/8 digits), format.
- Check whether the code is ever returned in responses, logs, or debug headers.
- Check OTP entropy: sequential, timestamp-derived, or weak-PRNG codes have
a much smaller candidate space than the nominal digit range.
- Note whether the same OTP is reusable across accounts or across flows
(verify vs. reset vs. login).
Phase 1 — Baseline and response-oracle discovery
Send 10–20 wrong OTPs plus 1 valid OTP and diff the responses. Use the script's probe mode to inspect a single attempt:
python3 scripts/otp_bruteforce.py --url https://target/api/verify-phone \
--method POST \
--headers '{"Content-Type":"application/json","Cookie":"session=abc"}' \
--body '{"phone":"+15551234567","otp":"{{OTP}}"}' \
--probe 123456Signals to look for:
- Content-Length (e.g., valid = 961 bytes — the CoinMate oracle)
- HTTP status (200 vs 400/403)
- Body markers (
"verified": true, new session token, redirect) - Timing (valid code may take a different code path)
- Set-Cookie / session mutation
If no oracle is obvious, run the full sweep in auto mode: the valid OTP is statistically the outlier bucket among otherwise uniform responses.
Phase 2 — Rate-limit and lockout probing
- Fire 10–30 rapid wrong attempts. If you never see 429/403, CAPTCHA, lockout,
or escalating delay, the endpoint is a brute-force candidate.
- Test throttling scope: limits keyed to session/IP only are bypassable via
new sessions, IP rotation, or X-Forwarded-For spoofing — retest per account.
- Check lockout semantics: per-account, per-session, or global? Does a lockout
also kill the legitimate user (DoS angle)?
- Test OTP TTL and reuse: does a code survive past expiry? Is it
single-use? Does resending invalidate the previous code?
Phase 3 — Brute-force automation
Use scripts/otp_bruteforce.py:
# Known oracle: valid responses are 961 bytes (the CoinMate signal)
python3 scripts/otp_bruteforce.py --url https://target/api/verify-phone \
--method POST \
--headers '{"Content-Type":"application/json","Cookie":"session=abc"}' \
--body '{"phone":"+15551234567","otp":"{{OTP}}"}' \
--oracle-length 961 --threads 10
# No oracle known: auto-detect the outlier response
python3 scripts/otp_bruteforce.py --url https://target/api/verify-phone \
--method POST \
--headers '{"Content-Type":"application/json","Cookie":"session=abc"}' \
--body '{"phone":"+15551234567","otp":"{{OTP}}"}' \
--range 100000-999999 --threads 20 --delay 0.05
# Route everything through Burp for visibility
python3 scripts/otp_bruteforce.py ... --proxy http://127.0.0.1:8080Burp Intruder alternative (the original researcher's method):
- Positions: replace the OTP value with
§100000§. - Payload type: Numbers, 100000 to 999999, step 1, **min/max fraction digits =
6**.
- Resource pool: 10–20 concurrent requests; watch for 429 and slow down.
- Grep-Match on the oracle signal (length column, marker, or status).
- On hit: copy the valid response, replay the original intercepted request
with that OTP (Burp Repeater) to complete the action.
Phase 4 — Confirmation and impact chaining
- Confirm the OTP on a fresh replay of the original request — this proves
the finding end-to-end (the CoinMate report's Step 4). For single-use code flows, use --confirm-trials 1 since the sweep may have consumed the code.
- Chain impacts and re-score:
- Phone added to profile → OTP-based password reset → ATO
- MFA bypass → session hijack on victim accounts
- Payment confirmation bypass → financial fraud
- Reset-flow brute force → mass account takeover
- If the flow targets victim accounts (attacker triggers reset on someone
else's account), interaction/privilege requirements change the score.
Detection heuristics summary
| Signal | Meaning |
|---|---|
| Unique content-length bucket (e.g., 961) | Valid-code oracle |
| Rare HTTP status (200 vs 400) | Valid-code oracle |
| Success marker / token in body | Valid-code oracle |
| No 429/403/lockout after 30 rapid tries | No rate limiting |
| Lockout only per-IP/session | Throttling bypassable |
| Old code still valid after resend/expiry | Weak TTL/reuse |
| Constant response size for all inputs | No oracle — needs timing or other side channel |
Reporting
- CWE-307 (Improper Restriction of Excessive Authentication Attempts) is the
primary mapping; also consider CWE-640 (weak password recovery) and CWE-287 (improper authentication). Note: platform auto-tags can be off — the CoinMate case was tagged "Insecure Storage of Sensitive Information".
- CVSS 3.1: base ~4.3–6.5 medium
(AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N). Escalate to 7.5+ high when it directly yields account takeover or bypasses MFA on victim accounts.
- PoC write-up: endpoint, request template, oracle used, candidate space,
request rate, time-to-recovery, and the replay confirmation.
Remediation checklist
- Rate-limit per account + IP (e.g., 5 attempts / 15 min) with CAPTCHA
escalation; key limits on the user identifier, not just session/IP.
- Short OTP TTL (2–5 min), single-use, invalidated on resend.
- Cryptographically secure RNG for code generation; never sequential or
time-derived codes.
- Server-side attempt counter with lockout; uniform responses so no
length/status oracle leaks validity.
- Monitor and alert on verification-failure bursts; add jitter to error paths.
- Prefer TOTP/HOTP (RFC 6238/4226) with rate-limited verification windows.
Files
scripts/otp_bruteforce.py— threaded Python3 brute-forcer with explicit
and auto-detect oracles, throttling backoff, confirmation replays, and Burp proxy support.