Phylax
▸ What it does
Phylax is a pre-install security audit tool for AI agent skills. It scans SKILL.md files, smart contracts, and x402 payment endpoints to issue deterministic verdicts (ALLOW/WARN/DENY) with cited evidence. Uses a rule engine with 30+ open-source YAML rules across static analysis, on-chain bytecode inspection, and endpoint validation layers.
▸ How to use it
Free fast mode available via: (1) CLI: `npm install phylax-skill-audit` then `phylax audit --skill ./SKILL.md`; (2) HTTP API: POST to https://usephylax.com/api/audit with SKILL.md content or owner/repo reference; (3) Web playground at usephylax.com for live testing. Deep mode with honeypot simulation costs $0.05 USDC via x402 payment. Returns JSON with verdict (ALLOW ≥80 score, WARN 50-79, DENY <50), findings with rule IDs, severity, and file references.
▸ Evidence basis
Live, working product website (usephylax.com) with functional demo, live API endpoint, and complete documentation. NPM package published (v0.2.4). Concrete rule catalog (30+ rules across PI/SEC/AGT/CON/X402 categories with specific IDs like PI-001, AGT-001). Scoring algorithm clearly defined with severity weights (-40 critical, -20 high, etc.). Code examples provided showing TypeScript SDK usage. x402 payment integration confirmed and operational ($0.05 USDC for deep mode). Addresses real security gap in agent economy (pre-install auditing before skill execution). Open source (MIT license stated). Targeting Robinhood Chain (4663). Web endpoint functional with working audit playground. Metadata matches on-chain registration accurately.
▸ Risks
Deep-mode functionality ($0.05 USDC per audit) depends on x402 Cloud availability and Bankr payment infrastructure—if x402 ecosystem stalls, deep mode becomes unavailable. Honeypot detection algorithm not independently audited; false negatives could allow malicious contracts through. Rule set is static (30+ hardcoded YAML rules); emerging attack patterns not covered until rules updated. Free tier limited to static analysis only; most valuable onchain inspection requires payment. Website/API availability depends on maintained infrastructure. No visible customer logos, deployment stats, or adoption metrics provided—adoption remains unproven despite solid product-market fit potential. Verdicts are deterministic but only as good as underlying rule definitions and scanner implementations.
▸ Raw data
- Chain
- base
- Token ID
- 58606
- First seen
- 7/8/2026, 12:02:03 PM