agentleFS
KaanBicaklarKaanBicaklar/mona-sast-skills/skills/unrestricted-file-upload/SKILL.md

unrestricted-file-upload

Skill2 starsChanged 4 months agoMITView on GitHub

SAST detection methodology for unrestricted / dangerous file upload (CWE-434), including extension and Content-Type trust bypass, double extension, polyglot files, filename-driven path traversal into the webroot, SVG/HTML stored XSS, and parser-exploit RCE (ImageMagick/Ghostscript/ExifTool/ffmpeg). Use when reviewing code that accepts, stores, serves, or parses uploaded files. Writes confirmed findings to findings/26-unrestricted-file-upload.md.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Decide type by content, not metadata: verify magic bytes / sniff the real MIME (file/libmagic, image/* header checks), and require it to …

    line 82
  2. Distinctive:

    Missing magic-byte / MIME sniffing: no content verification at all.

    line 158
  3. Very specific:

    Verify file type by content (magic bytes / sniffing) against a strict allowlist, never by the client extension or Content-Type.

    line 187

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Unrestricted File Upload — SAST Methodology

Class: Unrestricted File Upload · CWE-434 · OWASP: A04 Insecure Design Findings file: findings/26-unrestricted-file-upload.md

1. Overview

An upload flaw exists when the server accepts a file whose type, content, or name is attacker-controlled and then does something dangerous with it: serves it from a location where it executes, stores it under an attacker-chosen path, or parses it with a fragile/unsafe processor. The core test: can an attacker upload a file that (a) executes as server-side code when later requested, (b) is written to an executable or attacker-chosen location, (c) triggers a parser/deserialization exploit on ingest, or (d) is served back and rendered as active content (SVG/HTML → XSS)? Validation based on the client's extension or Content-Type is not a control — both are attacker-supplied.

2. Where it lives

  • Profile/avatar and cover-image uploads, document/attachment handlers, resume/CV and "import" features, media galleries, message attachments.
  • CSV/XLSX/JSON import, plugin/theme/template upload, firmware/backup restore.
  • Media pipelines that shell out to image/video processors on ingest (thumbnailing, transcoding, EXIF stripping, virus scan, OCR).
  • "Fetch from URL" avatar/import features (the fetched bytes are the upload, and the fetch overlaps SSRF).

3. Sources (tainted input)

The uploaded file bytes, the client filename, and the Content-Type header — all attacker-controlled. For archives, every entry name and entry content inside the uploaded zip/tar. Second-order: a stored file re-served later, or a stored filename joined into a later path (overlaps path traversal).

4. Sinks (dangerous operations)

// PHP — dangerous
move_uploaded_file($_FILES['f']['tmp_name'],
    "/var/www/html/uploads/" . $_FILES['f']['name']);   // client name → shell.php in webroot
// Node — dangerous
multer({ storage: multer.diskStorage({
  destination: 'public/uploads',                         // served & guessable dir
  filename: (r, file, cb) => cb(null, file.originalname) // attacker name: ../ or .php.jpg
})});
if (file.mimetype === 'image/png') accept();             // client-set MIME, trusted
# Python — dangerous
f = request.files["upload"]
f.save(os.path.join(UPLOAD_DIR, f.filename))             # filename may contain ../
subprocess.run(["convert", tmp_path, out])               # ImageMagick on attacker bytes
Image.open(f).thumbnail((100,100))                        # parser exploits on ingest
// Java — dangerous
String name = part.getSubmittedFileName();               // client-controlled
Files.copy(part.getInputStream(), Paths.get(webroot, name));  // → .jsp in served dir

Dangerous whenever: the storage path/filename comes from the client, the storage directory is served and allows script execution, acceptance is decided by extension/Content-Type, or the bytes are handed to a parser (image/video/office) that has known RCE surface.

5. Sanitizers / safe patterns

Safe:

  • Decide type by content, not metadata: verify magic bytes / sniff the real MIME (file/libmagic, image/* header checks), and require it to match an allowlist — while still not trusting the client extension or Content-Type.
  • Generate the stored name server-side (random UUID) with a fixed, safe extension; never reuse the client filename for the path.
  • Store outside the webroot or on object storage; serve through a handler that sets Content-Type from the verified type, Content-Disposition: attachment, and X-Content-Type-Options: nosniff.
  • Disable execution in the upload directory (no PHP/CGI handler, php_admin_flag engine off, no Options ExecCGI).
  • Re-encode images through a trusted library (strip to a clean canvas) so polyglots and embedded payloads are destroyed; sandbox/patch/limit media processors (ImageMagick policy.xml, Ghostscript -dSAFER, disable coders).
  • For archives, validate each entry destination (zip slip) and re-check every extracted file's type.

Fails / not a real sanitizer:

  • Trusting the extension or Content-Type — both are set by the attacker.
  • Blocklisting extensions — bypassed by .php5, .phtml, .pht, .phar, .jsp/.jspx, .asp/.aspx/.cshtml, case (.PHp), trailing dot/space (shell.php., shell.php on Windows), and double extension shell.php.jpg (served as PHP where the web server maps by an interior known extension).
  • Magic-byte check only on the first bytes — a polyglot has a valid image header and executable content after it.
  • Allowlisting the extension but keeping the client filename for the path — ../../shell.php still traverses into the webroot.
  • Checking type but storing in an executable, served directory — a valid image that also contains PHP still runs if the URL ends in .php.
  • Re-encoding declared but skipped for some formats (e.g. SVG passed through as "image").

6. Detection methodology

  1. Find sinks: grep for upload receivers, storage writes, and media shell-outs.
    rg -n 'move_uploaded_file|\$_FILES|multipart|MultipartFile|getSubmittedFileName'
    rg -n 'multer|formidable|busboy|request\.files|part\.getInputStream|save\(|copyToLocal'
    rg -n 'originalname|filename|getName\(\)|submittedFileName'   # client-name into path
    rg -n 'convert |mogrify|gm |gs |ghostscript|exiftool|ffmpeg|Image\.open|ImageIO\.read'
    rg -n 'content[_-]?type|mimetype|getContentType'             # metadata-based checks
  2. Confirm trust in client metadata: is acceptance gated on extension/ Content-Type rather than content sniffing?
  3. Trace the stored path and name: is the client filename used for the path? Is the destination inside the webroot / an executable directory?
  4. Check post-storage handling: is it served with a sniffable type / inline disposition? Is it parsed by a fragile processor?
  5. Confirm reachability: exposed upload route; pre-auth? Is the stored file retrievable at a predictable URL?
  6. Classify: exec-in-webroot RCE, filename traversal, stored XSS (SVG/HTML), parser RCE, or archive-based traversal.

7. Modern & niche variants

  • Extension / Content-Type trust bypass: the root cause — both are attacker- set. Any control built on them alone is defeated.
  • Double extension (shell.php.jpg): exploits web-server handler mapping (e.g. legacy Apache AddHandler/mod_mime matching an interior .php) so a file that "looks" like a JPEG executes as PHP.
  • Polyglot files: a single file valid as two formats — GIFAR (GIF + JAR), phar-JPEG, PDF-JS, and image+HTML polyglots — passes an image check yet is interpreted as code/archive by another consumer.
  • Attacker-controlled filename → path traversal into the webroot: a filename like ../../var/www/html/shell.php writes an executable file into a served directory (direct overlap with path-traversal write-side).
  • SVG / HTML upload → stored XSS: SVG is an image but rendered as markup — <script>/<foreignObject>/event handlers execute in the app origin when the file is opened inline. HTML/.xht uploads served inline do the same. SVG can also trigger SSRF/XXE via external references.
  • Image-parser exploits: ImageTragick (CVE-2016-3714) — ImageMagick delegate/ MSL/coder abuse turns a crafted image into command execution during thumbnailing.
  • Deserialize / command execution on parse: Ghostscript -dSAFER bypasses (PostScript/PDF → RCE), ExifTool DjVu injection (CVE-2021-22204 → RCE), ImageMagick MVG/MSL/ephemeral: coders, and ffmpeg HLS/concat/SSRF reads — any processor invoked on attacker bytes is an ingest RCE surface.
  • Missing magic-byte / MIME sniffing: no content verification at all; the file is accepted on name alone.
  • Upload into an executable / served directory: even a correctly typed file is dangerous if the storage dir runs scripts or the type can be sniffed by the browser (missing nosniff).
  • Archive upload: uploaded zip/tar extracted server-side → zip slip writes a web-shell outside the target, or bombs the disk (DoS).

8. Common false positives

  • Content verified by magic bytes against an allowlist, stored with a server- generated random name and safe extension, outside the webroot / on object storage.
  • Files served only as attachment with nosniff and a fixed non-executable Content-Type, from a directory with script execution disabled.
  • Images re-encoded through a trusted library that strips embedded payloads.
  • Uploads never served back and never parsed by a media processor.

9. Severity & exploitability

Base High. Critical when it yields RCE — a web-shell executing in the webroot, an ingest parser exploit (ImageMagick/Ghostscript/ExifTool/ffmpeg), or an archive drop into an executable path — or when it is pre-auth. Stored XSS via SVG/HTML is High in an authenticated context. Lower to Medium when the file is stored inert (random name, outside webroot, attachment-only, never parsed) and only metadata is questionable. See references/severity-model.md.

10. Remediation

Verify file type by content (magic bytes / sniffing) against a strict allowlist, never by the client extension or Content-Type. Store with a server-generated random name and a fixed safe extension, outside the webroot or on object storage, in a directory where script execution is disabled; serve as attachment with nosniff. Re-encode images through a trusted library and never render SVG/HTML inline in the app origin. Sandbox, patch, and policy-restrict any media processor (ImageMagick policy.xml, Ghostscript -dSAFER); validate archive entries against zip slip.

11. Output

Append each confirmed finding to findings/26-unrestricted-file-upload.md using references/finding-template.md. Set Class: Unrestricted File Upload, CWE: CWE-434, and in Chain potential name primitives such as RCE (web-shell or parser exploit — the terminal primitive of many chains), arbitrary file write (consumes/overlaps a path-traversal primitive via the filename), stored XSS (SVG/HTML → session/credential theft), and SSRF (SVG/ffmpeg external refs). Note when the finding consumes a path-traversal primitive to control the stored path.

Primitives (controlled): provides FILE_WRITE,HTML_OUTPUT (SVG/HTML),CODE_EXEC; consumes none

References

  • CWE-434 Unrestricted Upload of File with Dangerous Type; OWASP A04:2021 Insecure Design and A08:2021 Software and Data Integrity Failures; OWASP File Upload Cheat Sheet; ImageTragick (CVE-2016-3714) and ExifTool (CVE-2021-22204) advisories.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.