Documentation

How Pentest Agent works

Pentest Agent is autonomous security intelligence for the open web. It runs public, non-invasive checks on a website, explains what it finds, and publishes an evidence-backed report anyone can verify.

Getting started

There are four ways to request a check:

  • • The website: paste a domain or a token contract address.
  • • X: mention @PentestAgent with a website, a CA, or under a project's post.
  • • The API: free endpoint or pay-per-scan with x402.
  • • The Pons monitor: launches on Pons (Robinhood Chain) are checked automatically.

Every check produces a public report page with the grade, every finding and its evidence, a fix plan, a live agent trace, a screenshot and a claims check.

Scans & grading

A scan covers transport security (HTTPS, redirects, HSTS, legacy TLS), HTTP security headers, cookies, DNS and email posture, frontend signals, and Web3 frontend checks: wallet libraries, known-malicious package releases, CDN-loaded wallet code, public source maps and contract verification.

Findings come from deterministic checks, never from a model's opinion. Each site starts at 100 and loses points per issue by severity; heuristic (“potential”) findings count half, and checks that couldn't run are excluded. The full rubric is on the methodology page.

Token (CA) checks

Paste a contract address instead of a domain. For Pons tokens on Robinhood Chain we read the token's on-chain metadata; for other chains we use DexScreener. We then check the website the token lists.

Because anyone can copy a project's website and X handle into a new token, the token is only shown as confirmed when the project's website lists the contract address, or the project's own X account has posted it. Otherwise the report says it's not confirmed.

Website & claims

Every scan visits the site in a real browser, captures a screenshot, reads the domain's registration date, and lists the social, docs and audit links it shows.

The AI extracts factual claims as verbatim quotes (supply, tax, liquidity locks, renounced ownership, audits, partners, team, age). Each claim is checked against data and labelled matches, doesn't match or can't verify. “Can't verify” never means false.

X agent

@PentestAgent check example.com
@PentestAgent check 0x…            (token contract address)
@PentestAgent check this           (as a reply under a project's post)

Free for followers of @PentestAgent, up to 2 checks per account per day (resets 00:00 UTC). Replies include the grade, finding counts, token confirmation and a link to the report, with a report card image attached.

API & x402

Base URL https://api.pentestagent.dev. The free endpoint is rate limited; premium ($0.10) and bulk ($0.05/domain) scans are paid per request with x402 in USDC on Base or Solana, or USDG on Robinhood Chain. No signup, no API keys.

POST https://api.pentestagent.dev/v1/scans            { "target": "example.com" }   free
POST https://api.pentestagent.dev/v1/premium/scan     { "target": "0x…" }           x402
GET  https://api.pentestagent.dev/v1/scans/{id}                                       status + JSON report

Full reference and a working TypeScript example: For Agents.

Pons monitor

When a monitoring window is open, the agent watches every new launch on Pons V2. Tokens that list a real website, reach a $10k market cap and are confirmed as the project's own get a premium check. Results appear under Explore → Pons launches, and the most notable one per 15-minute round can be posted from @PentestAgent.

Safety & opt-out

Checks are passive and bounded: a few dozen ordinary requests for pages and scripts any visitor loads, private and internal addresses refused, strict timeouts and rate limits. Paying never expands what we test. Domain owners can opt out or dispute a finding; see opt-out & abuse reporting.

FAQ

Does a good grade mean a project is safe?
No. It means the externally visible controls we check are in place. It says nothing about the token contract, team or liquidity.
Do you hack or attack websites?
No. We only request what any browser would load, with strict limits, and never attempt exploitation or credential testing.
Why isn't my token confirmed?
List your contract address on your website, or post it from your project's X account, and run the check again.
Can I hide my report?
Domain owners can opt out of public assessments; reports for opted-out domains are hidden.
Does Pentest Agent have a token?
See the $PENTEST page: the official contract address is only ever published there and by @PentestAgent.