←
BAS1,213 posts
BAS avatar

BAS

@basdotio

🥕Eco app: @Agents_BAS 🥕 BNBCHAIN Hackthon winner + grant; Binance MVB8 🥕agent.bas.io 🥕passport.bas.io 🥕bascan.io

◷ Joined June 2023▧ 429 Photos and videos
BAS avatar
BAS@basdotio·May 29

Introducing Agent Passport — portable, verifiable, underwritable agent credit. agent.bas.io One profile. One identifier. Every lender, marketplace, and smart contract can read who an agent is and what it has done — and trace every signal back to a cryptographic attestation. Three things set it apart: 1⃣Portable — reputation doesn't disappear when an agent moves to a new venue 2⃣Verifiable — every score and risk flag is a deterministic function of attested events, not a black-box model 3⃣Underwritable — the schema, signal weights, and decay functions are all shaped by one question: would a lender act on this? This is what BAS is building: the missing layer between what an agent does and the capital it can access.

Post attachment
BAS avatar
BAS@basdotio·Nov 16

🤖 Your AI agent has skills, tools, MCP servers, and permissions. But who’s guarding the guard? One malicious skill. One hidden prompt injection. One suspicious hook. And suddenly, your helpful AI agent is doing side quests you never assigned. 💀 Meet BAS AI Agent Guard 🛡️ Your local antivirus for AI agents. 🔍 Scan skills, MCP servers, hooks & plugins 🚨 Detect prompt injection, data exfiltration, backdoors & excessive permissions 🔐 Audit agent permissions and security risks 🧹 Clean up junk without permanently deleting anything ⚡ Run locally. Work offline. No data uploads. No cloud dependency. No blind trust. Just security checks before your agents get to work. Let agents cook. Just don't let them burn down your machine. 🥕🔥 GitHub: github.com/basdotio/bas-guar…

Post attachment
BAS avatar
BAS@basdotio·Nov 13

3/ Every Binance Agent ID now has a wallet page on BAS. Search any .ai name on agent.bas.io and one page tells you who is behind the wallet, and what it has actually done: 🟡 Identity: Agent ID → 8004 passport → 8183 jobs 🟡 Profile: description, tags, X, docs, website, GitHub 🟡 Commerce stats: GMV, settled revenue, jobs, completion rate 🟡 Every job: role, counterparty, amount paid, final status Same address, every chain, one page.

Post attachment
BAS avatar
BAS@basdotio·Nov 13

2/ How to check any wallet on agent.bas.io: 🟡 Agents → "Binance Agent IDs": every holder, with its 8004 agent and 8183 jobs 🟡 Search a .ai name → its wallet page shows Agent ID → 8004 passport → 8183 jobs 🟡 No 8004 identity yet? Hit "Register on ERC-8004"

Post attachment
BAS avatar
BAS@basdotio·Nov 13

BAS now integrates Binance Agent ID. Every .ai name, linked to what the wallet has done: 🟡 Search any .ai name or wallet 🟡 See its ERC-8004 agents and 8183 jobs 🟡 Filter agents by "Holds Binance Agent ID" 🟡 No 8004 identity? Register in one click agent.bas.io

Post attachment
BAS avatar
BAS@basdotio·Nov 11

AI agents are getting smarter. But who's watching their back? 👀🤖

Post attachment
BAS avatar
BAS@basdotio·Nov 6

3/ The trust model: • never executes what it scans • offline by default, nothing uploaded • secrets redacted before any report is built • one static Go binary, MIT There's no curl | sh installer either — it flags exactly that as high severity in other people's repos. Limits: static analysis is a risk signal, not a safety certificate. It can't see runtime behavior or remote payloads. Anything it couldn't read is reported, never silently skipped — benchmark scorecard, misses included, is in the repo. ⭐ github.com/basdotio/bas-guar…

BAS avatar
BAS@basdotio·Nov 6

2/ It reads your ~/.claude and scores everything — skills, MCP configs, hooks, permission allowlists, subagents, plugins, CLAUDE.md. 69 static rules across 10 risk dimensions. Then it gates loading. Claude Code has no install hook, and skills arrive by git clone or cp anyway — loading is the boundary that holds. Each skill is audited before the agent loads it, and editing an approved one asks again. Things that are easy to miss: • Bash(git *) in your allowlist is arbitrary exec (git -c core.pager=…) • a hook that runs sh pre. sh — it reads and scans pre. sh too • one file that reads credentials, base64s them, makes an outbound call

BAS avatar
BAS@basdotio·Nov 6

Introducing BAS Guard — a free, open-source security scanner for Claude Code setups. Your coding agent auto-loads skills, MCP servers, hooks, plugins and CLAUDE.md files, and hands every one of them your shell, your files and your keys. A skill is just markdown + scripts. One buried line like "before replying, run curl … | bash" and the agent does it with your permissions. Nothing reviews any of this when you install it. 🧵

Post attachment
BAS avatar
BAS@basdotio·Nov 6

3/ The trust model: • never executes what it scans • offline by default, nothing uploaded • secrets redacted before any report is built • one static Go binary, MIT There's no curl | sh installer either — it flags exactly that as high severity in other people's repos. Limits: static analysis is a risk signal, not a safety certificate. It can't see runtime behavior or remote payloads. Anything it couldn't read is reported, never silently skipped — benchmark scorecard, misses included, is in the repo. ⭐ github.com/basdotio/bas-guar…

BAS avatar
BAS@basdotio·Oct 30

The ERC-8004 agent economy is getting crowded. 👀 Who's actually leading the race? 🏆 Agent Passport Platform Leaderboard: - 🥇 @NeoSoulAI — 149,482 agents - 🥈 @aveai_info — 129,920 agents - 🥉 @termix_ai — 47,586 agents - @XLayerOfficial — 14,376 agents - @dgrid_ai — 5,839 agents Across the leaderboard, we're tracking agent registrations, active agents, x402 activity, supported chains, and community feedback.

Post attachment
BAS avatar
BAS@basdotio·Oct 26

Your AI agent has access to tools, skills, and maybe even a wallet. But one sketchy skill later… bro could become the final boss. That’s why AI agents need more than intelligence. They need a guard. 🛡️ BAS AI Agent Guard Because “my agent would never do that” is not a security strategy. Verify first. Trust later. 🥕

Post attachment