LVL 01
Easy
Credential Gate Bypass
A staff console authenticates against the users table. Walk in as the administrator without ever knowing the password.
Not solved
Open →
CSPSHIVAM // security playground
Learn web & AI security by breaking things — safely.
A hands-on lab environment for students and enthusiasts. Practice real attacks — SQL injection, SSRF, modern web bugs, CSRF, open redirects, XSS, and AI/ML exploits — in isolated, intentionally-vulnerable sandboxes. No setup. Each lab nudges you toward the technique — and, for server-side bugs, where to point your proxy — but never the payload. Capture the flag, then read the walkthrough to understand exactly why it worked.
58
labs
7
attack tracks
0
solved by you
SQLi
SQL Injection
Bend query logic through untrusted input — auth bypass, extraction and blind inference. The injectable channel is off the visible form: bring a proxy.
0 / 6
✓ Injection Operator earned
LVL 02
Medium
Column Harvesting (UNION)
A catalogue endpoint returns product rows. Splice a second query onto it and lift data from a table you were never meant to read.
Not solved
Open →
LVL 03
Medium
Verbose-Error Extraction
A lookup surfaces raw database errors to the page. Turn those errors into a data-exfiltration channel.
Not solved
Open →
LVL 04
Hard
Boolean Inference
The endpoint answers with nothing but "match" or "no match". Interrogate the database one true/false question at a time.
Not solved
Open →
LVL 05
Hard
Timing Inference
No body change, no error — only latency. Read the database purely by how long each response takes.
Not solved
Open →
LVL 06
Hard
Keyword-Filter Evasion
A blocklist scrubs the obvious SQL keywords before the query runs. Reconstruct your injection past it.
Not solved
Open →
SSRF
Server-Side Request Forgery
Coerce the server into fetching internal-only resources. The fetch target lives in a request element the UI pins — you override it in the proxy.
0 / 6
✓ SSRF Operator earned
LVL 01
Easy
Reach the Intranet
A link-preview service fetches whatever address it is handed. Steer it at an internal-only panel the box can see but you cannot.
Not solved
Open →
LVL 02
Medium
Harvest Instance Creds
The fetcher runs on a cloud instance. Make it read the instance metadata service and hand you its credentials.
Not solved
Open →
LVL 03
Hard
Loopback Blocklist Evasion
Literal localhost and 127.0.0.1 are refused. Reach the loopback service through a spelling the filter never anticipated.
Not solved
Open →
LVL 04
Hard
Redirect-Chained Fetch
Internal hosts are rejected up front, but redirects are followed blindly. Bounce the fetch inward through an allowed origin.
Not solved
Open →
LVL 05
Hard
Blind Callback Proof
The response body is swallowed. Prove the server reached your target by driving it to a unique out-of-band host.
Not solved
Open →
LVL 06
Medium
Internal Port Sweep
The fetcher reports reachability. Use it to map which services listen on the internal host and read what one of them leaks.
Not solved
Open →
Modern
Modern Web Attacks
Access control, SSTI, JWT, uploads, host-header and reset-flow abuse, response tampering. The advanced track — nearly all of it is proxy-driven.
0 / 14
✓ Web Exploitation Operator earned
LVL 01
Medium
Object-Reference Abuse (IDOR)
A billing portal keys invoices by opaque reference and never checks who owns them, while a sibling API quietly lists them. Open an invoice that is not yours.
Not solved
Open →
LVL 02
Medium
OS Command Injection
A diagnostics tool shells out to probe a host. Ride the command it builds and run your own.
Not solved
Open →
LVL 03
Medium
Directory Traversal (LFI)
A document viewer resolves pages by name inside a base folder with no containment. Escape it and read a file outside.
Not solved
Open →
LVL 04
Hard
Template Injection (SSTI)
A personalised banner is rendered from a template that splices in a request value as source, not data. Make the engine evaluate your expression.
Not solved
Open →
LVL 05
Hard
JWT Forgery (alg abuse)
The session is a JSON Web Token the server trusts to describe your role. Mint one that promotes you.
Not solved
Open →
LVL 06
Hard
Executable Upload
An avatar uploader screens for the obvious extension only. Slip through a file the server still runs as code.
Not solved
Open →
LVL 07
Hard
XML External Entity (XXE)
A feedback endpoint parses XML you submit with entity resolution left on. Coerce the parser into reading a local file.
Not solved
Open →
LVL 08
Medium
Over-Trusting Mass Assignment
A profile form claims to edit only your name and bio, but the save endpoint trusts every field the read API over-shares. Assign yourself a role you should not have.
Not solved
Open →
LVL 09
Hard
Reset-Link Poisoning
A "forgot password" flow builds the emailed reset link from the incoming request. Make the link point at infrastructure you control so the victim's token lands in your lap.
Not solved
Open →
LVL 10
Hard
Reset-Token Takeover
The password-reset confirmation trusts a token and a user reference carried in the request — and the token is not unguessable. Complete a reset for an account that is not yours.
Not solved
Open →
LVL 11
Hard
API Login Bypass (SQLi)
The web login is hardened, but a JSON authentication API behind it still stitches your input into the query. Authenticate as an operator through the API.
Not solved
Open →
LVL 12
Medium
Host-Header Injection
A page caches and links itself using the Host the request arrives with. Poison what it emits by controlling that header.
Not solved
Open →
LVL 13
Medium
Response Tampering
An entitlement check runs client-side against a verdict the server already sent. Rewrite the verdict on its way back and unlock the gated resource.
Not solved
Open →
LVL 14
Medium
Undocumented POST Field
A form submits a tidy set of fields, but the handler honours one more that never appears on the page. Discover it and send it.
Not solved
Open →
CSRF
Cross-Site Request Forgery
Ride a victim session to force state-changing actions. Read the genuine request in a proxy, then build the cross-origin forgery that reproduces it.
0 / 6
✓ Session Forgery Operator earned
LVL 01
Easy
Unguarded Email Change
A signed-in account can swap its email with no anti-forgery token. Drive that change from a request the user never intended.
Not solved
Open →
LVL 02
Easy
State Change over GET
A sensitive action fires on a bare GET. Trigger it without the victim lifting a finger.
Not solved
Open →
LVL 03
Medium
Guessable Token
A token exists, but it is derived from public data. Reproduce it and push the request through.
Not solved
Open →
LVL 04
Medium
Sloppy Referer Gate
The server sniffs the Referer header — and does it badly. Satisfy the check from a host you own.
Not solved
Open →
LVL 05
Medium
Login CSRF
Instead of hijacking the victim, plant them inside an account you control. Forge the sign-in.
Not solved
Open →
LVL 06
Hard
Session-Unbound Token
Tokens are checked for shape but never tied to a session. Reuse one across users.
Not solved
Open →
Redirect
Open Redirect
Abuse trust in redirect targets to bounce a victim to an origin you control. URL-driven — solvable straight from the browser.
0 / 6
✓ Redirect Operator earned
LVL 01
Easy
Raw url Parameter
A "continue" link forwards to whatever a parameter says. Bounce a visitor to an origin you control.
Not solved
Open →
LVL 02
Medium
Prefix-Check Escape
The guard insists the target starts with the trusted host. Grow a URL that begins with it and ends somewhere else.
Not solved
Open →
LVL 03
Medium
Client-Side Redirect
The forward is performed in the browser. Own the destination anyway.
Not solved
Open →
LVL 04
Hard
Parser-Confusion Bypass
An allowlist inspects the wrong slice of the URL. Exploit the gap between what it reads and where the browser goes.
Not solved
Open →
LVL 05
Hard
Fragment Token Leak
A freshly issued token rides along in the URL fragment. Redirect the flow so it spills onto your page.
Not solved
Open →
LVL 06
Medium
Suffix-Match Loophole
Only cspshivam.com and its subdomains are meant to pass. Register something that ends in those characters yet is entirely yours.
Not solved
Open →
XSS
Cross-Site Scripting
Land script execution in another user's browser: reflected, stored, DOM and sink-specific contexts. In-browser — no proxy needed.
0 / 10
✓ Client-Side Operator earned
LVL 01
Easy
Reflected — Body Context
A search box drops your query straight into the page body. Turn that echo into script execution.
Not solved
Open →
LVL 02
Easy
Reflected — Attribute Break
Your value lands inside an HTML attribute. Break out of it and run code.
Not solved
Open →
LVL 03
Medium
Reflected — Script String
Your input is embedded inside an inline script string. Terminate the string and execute.
Not solved
Open →
LVL 04
Medium
Stored — Public Wall
Notes you leave are persisted and replayed to every visitor. Plant one that runs on view.
Not solved
Open →
LVL 05
Medium
DOM — Fragment Sink
Browser-side code copies the URL fragment into the DOM. No server ever sees the payload.
Not solved
Open →
LVL 06
Medium
Filter — Tag Stripping
The app deletes the script tag. Execute without ever using it.
Not solved
Open →
LVL 07
Hard
Filter — Case & Encoding
A shallow keyword filter blocks a few words. Reshape your payload so it slips by.
Not solved
Open →
LVL 08
Medium
Sink — javascript: URI
Your input becomes a link target. Convert the navigation into execution.
Not solved
Open →
LVL 09
Medium
Sink — Event Handlers
Angle brackets survive but "script" is banned. Fire a handler that runs on its own.
Not solved
Open →
LVL 10
Hard
Context — JSON under CSP
Your input is reflected into a JSON blob served under a lax policy. Achieve execution regardless.
Not solved
Open →
AI Sec
AI / ML Security
Prompt injection, jailbreaks, system-prompt leakage, unsafe output handling and over-broad agency in LLM apps. Chat-driven — no proxy needed.
0 / 10
✓ AI Security Operator earned
LVL 01
Easy
Direct Prompt Injection
A support bot is told to guard a secret. Talk it out of it.
Not solved
Open →
LVL 02
Medium
Indirect Prompt Injection
The assistant summarises a document you hand it. Smuggle instructions inside the text.
Not solved
Open →
LVL 03
Medium
System-Prompt Disclosure
Recover the hidden system prompt the assistant runs under.
Not solved
Open →
LVL 04
Medium
Guardrail Jailbreak
The model declines restricted asks. Reframe until it complies.
Not solved
Open →
LVL 05
Hard
Unsafe Output Rendering
Replies are painted into the page as HTML. Turn one into cross-site script.
Not solved
Open →
LVL 06
Hard
Tool Abuse via Injection
The agent wields tools. Coax it into calling one it must not.
Not solved
Open →
LVL 07
Medium
Memorised-Data Extraction
A sensitive record sits in the model's context. Pull it out verbatim.
Not solved
Open →
LVL 08
Hard
Excessive Agency — SSRF
The agent can fetch URLs. Aim it at the internal metadata service.
Not solved
Open →
LVL 09
Medium
Classifier Evasion
A toxicity gate blocks certain messages. Get a blocked one through intact.
Not solved
Open →
LVL 10
Hard
Feedback-Loop Poisoning
A spam filter learns from user labels. Teach it the wrong lesson.
Not solved
Open →
About the playground
Built to be attacked — in the right places only.
CSPSHIVAM Playground is a self-contained training ground. Each lab is an isolated, deliberately vulnerable mini-app. The platform around it — this site, your progress, the walkthroughs — is hardened and holds no real accounts or data, so there is nothing here worth stealing. Attack the labs; the playground itself stays out of scope.
01 — Capture flags
Exploit, don't guess
Every lab hides a flag like CSPSHIVAM{…} that only appears once you actually pull off the
attack. Submit it on the lab page to mark it solved.
02 — Earn badges
Clear a whole track
Finish every level in a category and you'll be prompted for a name and a downloadable badge — Injection Operator, AI Security Operator, and more. Share it and tag #CSPSHIVAMPlayground.
03 — Understand it
Hints point, they don't solve
Each lab nudges you toward the technique — and, for proxy labs, which header or parameter to intercept — but never the payload. When you're ready, the Walkthroughs tab explains each bug end-to-end, with the working exploit and how to fix it.
04 — Stay safe
Sandboxed by design
Dangerous primitives (SSRF, command execution, file reads) run against bundled mocks — never the real host or network — so nothing you do here can escape the sandbox.
Progress is stored only in this browser. Reset my progress.