HACKAIZ / DOCS
Documentation
From your first scan to verifying fixes.
50 articles
Getting started
Choose the right first scan for your platform.
Connecting GitHub
Connect a public or private repository without uploading files.
Understanding results
Read severity, evidence and unfinished coverage together.
Fixes and verification
Review a change before approval and verify afterward.
Credits and interrupted work
Understand a pause and where to purchase credits.
Security coverage and access
Know what an address, repository or database access can establish.
LLM surface mapping
Discovers AI endpoints, declared tools, plugins and data sources the assistant can reach.
Prompt injection (direct and indirect)
Sends bounded probes and reports only when the answer really deviates.
System prompt and context leakage
Asks for hidden instructions and flags the answer only when instructions actually come back.
Unsafe handling of model output
Checks whether model output is rendered as active HTML, links or commands.
Unauthenticated or billable AI endpoint
Confirms a stranger can call the AI endpoint, and whether the call costs money.
Vector store and agent memory exposure
Checks whether embeddings or agent memory are readable or writable without credentials.
Model and plugin supply chain
Lists declared third-party models, plugins and MCP servers used by the project.
Training and retrieval data poisoning
Checks who can write into the sources the model learns from or retrieves.
Excessive agency and missing human confirmation
Finds sensitive tools an assistant can trigger with no human approval step.
Tool misuse and reachable agent APIs
Any API an assistant can call is treated as public, so we check it as a stranger would.
Agent identity and privilege abuse
Looks for shared keys, over-broad tokens and agents acting with a user's identity.
Agent-to-agent channels
Checks whether agent coordination channels answer without authentication.
Cascading failures and runaway agents
Design review: loop limits, budget caps, kill switch and human override.
Broken access control and IDOR
Unprotected admin paths, parameter-based access, observable object-id access, method and referer bypasses.
Authentication weaknesses
User enumeration, missing rate limiting, weak reset flows, weak persistent-login cookies.
Information disclosure
Verbose errors, debug pages, backup files, exposed version-control folders, source maps.
Security headers and clickjacking
Missing frame protection, CSP, HSTS, Referrer-Policy, cookie flags, HTTP downgrade paths and e-mail spoofing protection (SPF/DMARC).
CORS misconfiguration
Origin reflection, accepted null origin, credentials over insecure origins.
CSRF protection
Missing tokens, missing SameSite, method-dependent validation.
Host header handling
Host switching, reset-link poisoning, header-based authentication bypass.
OAuth and redirect handling
Loose redirect_uri, implicit flow accepted, open redirects usable for token theft.
JWT and token validation
Unverified signatures, alg none, kid path traversal, injectable key references, weak keys.
GraphQL exposure
Open introspection, default hidden endpoints, private fields listed to anyone.
API authorization matrix
Every documented route is called once without credentials; answers with data are proven.
SSRF and internal routing
Read-only checks for fetchers that follow attacker-supplied destinations, including routing-based variants.
Business logic and client trust
Prices, roles and quantities accepted from the browser; skipped steps in a flow.
Web cache deception
Observable unkeyed inputs only; the real user cache is never polluted.
Race conditions
Single-attempt checks on public idempotent endpoints; never any over-consumption.
NoSQL operator injection
Operator probes on public authentication endpoints, read-only.
Injection risk in code (SQL, command, template, path)
Dangerous code patterns are reported with the code as evidence; never exploited on the target.
Deserialization, XXE and upload handling
Vulnerable parsers, unsafe deserialization and upload validation read from code and dependencies.
Request smuggling and protocol edge cases
Not tested: proving it requires desynchronising a live server, which belongs in a scoped penetration test.
Active exploitation (RCE, brute force, data extraction)
Never performed. Hackaiz proves risk without damaging, extracting or overloading anything.
Database row-level security and role grants
Table-by-table read attempt as an anonymous client, plus grants and privileged functions when connected.
Storage and bucket exposure
Public buckets, listable objects and unintended public files.
Exposed secrets and keys
Keys shipped to the browser or committed, with the privilege of each key verified before alerting.
Tenant and account isolation
Checks that policies scope rows to their owner rather than to any signed-in user.
Data services exposed on the internet
One harmless read request to Elasticsearch, ClickHouse, Meilisearch, CouchDB and MongoDB default interfaces; reported only when the reply proves no credential was needed.
Database backups and recovery
With a read-only key: presence and age of the last restore point and whether point-in-time recovery is on.
Project sign-in settings
With a read-only key: open signup, email confirmation, password strength, token lifetime and rotation, and edge functions callable without a token.
Vulnerable dependencies
Known advisories matched against the project's real lockfile versions.
CI/CD and build pipeline
Workflow permissions, untrusted triggers, secret exposure in build logs, container base images.
GDPR and consent
Trackers before consent, missing legal pages, personal data sent to third parties.
Data subject rights readiness
Whether export, deletion and retention paths exist in the product at all.