| name | misc-deep-cuts |
| description | Final gap-fill checklist of miscellaneous techniques that don't fit a single vuln-class skill -- OTP exposure patterns, feature-flag/beta-access toggles, multi-step flow step-skipping, "remember me" cookie forgery, image-proxy LFI, CSV/formula injection, PDF-generator SSRF, header-based admin spoofing, filename-based attacks, and referral/coupon abuse. Converted from master-pentest-prompt.md Phase 28. Use as a final pass after the mainstream vuln-class skills, catching real-world bugs that don't slot neatly into one category. |
Misc deep cuts -- final gap fill
When to use
A final pass after the mainstream vuln-class skills -- these are real,
recurring bug patterns that just don't map cleanly onto a single
category, so they're easy to skip if testing is purely checklist-driven
by vuln class.
Checklist
- OTP exposure: the OTP returned directly in the response body, an
OTP embedded in a URL, or an OTP predictable from a timestamp
(including backup codes shown in a response where they shouldn't be).
- Feature flag / beta access: toggling
X-Feature-Flags,
X-Enable-Beta, ?beta=1, ?internal=1 -- these can reveal hidden
admin routes that were only ever "hidden," never actually
access-controlled.
- Step-skip logic: in a multi-step flow (signup -> verify ->
payment), POST directly to a later step, reorder the steps, replay a
step, or skip whatever CSRF protection only the last step has.
- "Remember me" persistent cookie: if the token is a base64/JSON
blob rather than an opaque session ID, decrypting or just decoding it
can lead straight to session takeover, especially with weak entropy.
- Avatar/image URL endpoints: an image proxy that reads a
user-supplied URL/path can become an LFI vector, not just SSRF (see the
ssrf skill for the URL-fetching side of this).
- CSV/Excel export injection: formula injection (
=, +, -, @
prefixes triggering DDE/command execution) in exported spreadsheet
data an admin later opens.
- Print/PDF generators: SSRF via externally-loaded images in the
rendered document, HTML injection into a server-side-rendered PDF,
RCE through the rendering engine/font-handling layer itself.
- Header-based admin detection:
X-Admin, X-User, X-Role
headers that the backend trusts if simply forged by the client.
- Filename-based attacks: CRLF or
../ inside a filename parameter
-- can produce stored XSS on download or path traversal via the
filename itself, not just the file content.
- Large file / decompression bombs: zip bombs, PDF bombs -- see the
dos-and-resilience skill for the broader DoS technique list.
- GraphQL loose ends: field suggestion left enabled,
__typename
leaking schema info, unbounded mutation complexity, WebSocket
subscriptions with no auth check (see deep-cut-surfaces for the
broader GraphQL/WebSocket list).
- Email verification bypass: changing the verification target to an
attacker-owned email mid-flow, using a verification link with the
attacker's own email against a victim's account, or an IDOR in the
email-template's own ID parameter.
- Referral/program abuse: self-referrals, coupon stacking, invite-
bonus farming through repeated fake signups.
- API documentation exposure:
/swagger-ui, /redoc, /api-docs,
a live /graphql playground, /v3/api-docs -- these map the entire
API surface for free once found.