Static
SKILL.md & manifest — injection, secret exfil, agent authorization.
PI-*SEC-*AGT-*MAN-*
PRE-INSTALL SECURITY · ROBINHOOD
Audit agent skills
before they touch your wallet.
Security layer for AI agent skills & x402 APIs. Scans SKILL.md, contracts, and payment endpoints — deterministic verdict with evidence.
phylax audit
$phylax --skill ./SKILL.md
▸Loading rules from /rules/*.yaml
▸Running static scan...
▸Running onchain scan (Robinhood Chain)...
▸Running endpoint scan...
ALLOWscore100· findings0
Read-only
Never modifies your environment
Stateless
No data stored between scans
Deterministic
Same input → same output
Evidence-first
Proof on every finding
Three layers. One verdict.
Independent scanners merge findings into a single deterministic score.
Onchain
Bytecode on Robinhood Chain — approvals, proxies, honeypot simulation.
CON-*LIQ-*
Endpoint
x402 payment URLs — HTTPS, schema, redirects, price sanity.
X402-*
30+ open-source YAML rules
Every finding cites a rule ID. Full definitions in the repo.
PI-001criticalFund transfer instruction
PI-002criticalOverride safety instructions
SEC-001criticalPrivate key / seed request
AGT-001criticalUnlimited spend approval
AGT-002highAuto-execute without confirmation
CON-020criticalHoneypot detection
CON-010highUnlimited approval
CON-011highProxy + single owner
X402-001highInvalid 402 schema
X402-041highHTTP endpoint
Deterministic ALLOW · WARN · DENY
Score starts at 100. Rule hits subtract severity weights. Every finding ships with evidence — no black boxes.
score=100−Σ(weight × hits)
critical −40high −20medium −10low −3
ALLOW
score ≥ 80
Safe to install with standard caution.
WARN
score 50–79
Review findings before proceeding.
DENY
score < 50
Do not install this skill.
audit-output.json
1{
2 "verdict": "DENY",
3 "score": 0,
4 "findings": [
5 { "id": "PI-001", "severity": "critical",
6 "evidence": "transfer all USDC..." },
7 { "id": "SEC-001", "severity": "critical",
8 "evidence": "provide your private key..." }
9 ]
10}
Machine-readable JSON with rule IDs, severity, and file references.
Audit a skill right now
Paste a SKILL.md or an owner/repo ref. Phylax runs the free static engine and returns a deterministic verdict with cited evidence — same output as the CLI and API.
skill-input
POST https://usephylax.com/api/audit · mode=fast · free
audit-output.json
Verdict, score, and findings appear here.
Try a sample, or paste your own SKILL.md.
Try a sample, or paste your own SKILL.md.
Read-only · deterministic · free fast engine. Deep mode (honeypot sim) is $0.05 USDC via x402.
npm, API, embed badge
Fast mode is free (CLI + API). Deep mode ($0.05 USDC) on x402 Cloud — honeypot simulation on Robinhood Chain.
audit.ts
import { audit } from "phylax-skill-audit"
const result = await audit({
skill_source: "owner/repo/SKILL.md",
chain_id: 4663,
mode: "fast",
});
console.log(result.verdict); // ALLOW | WARN | DENY
npm v0.2.4Robinhood 4663SSRF-hardened
Full documentationLive badge preview
Badge
[](https://usephylax.com)
API
curl -X POST https://usephylax.com/api/audit \
-d '{"skill_source":"owner/repo","mode":"fast"}'Security layer for skills & x402 APIs
Security layer for skills & x402 endpoints. Phylax audits what you install and what you pay for — it does not compete with x402 Cloud hosting.
Fast auditFree
Static scan, bytecode heuristics, x402 HEAD probes. CI & pre-install gates.
usephylax.com/api/audit
Deep audit$0.05 USDC
Honeypot simulation on Robinhood Chain + full onchain checks. Paid per request.
x402 Cloud
x402 · audit-deep
# Deploy once: x402 deploy → $0.05 USDC/request
POST https://usephylax.com/api/audit?mode=deep
{
"skill_source": "owner/repo/SKILL.md",
"chain_id": 4663
}$PHYLAX is live on Robinhood Chain
Only the contract below is official $PHYLAX. Verify the address before you trade — anything else is not us.
$PHYLAX · Robinhood ChainLIVE
Contract address · via Virtuals
0xff36bcb3cb9e3e2dac718ccaec3bc5ed489d9f0dDon't install blind.
Run Phylax before every skill install. Takes ~200ms.
npm install phylax-skill-audit