Your AI agent trusts every skill it loads. Most people never read them.
agentsec 0.4.0 now flags hidden prompt-injection & exfiltration instructions inside a skill's own text (OWASP AST05) — plus over-broad allowed-tools grants.
npx agentsec 🛡️
2
54
Introducing AgentSec — security auditing for AI agent skills with built-in web3 support.
OWASP AST10 covers generic skill risks. Our AST-10 Web3 Annex extends it with 12 onchain-specific rules: signing authority, Permit2 capture, blind signing, RPC pinning, kill-switch, MCP drift, oracle/slippage, key material.
🧵
3
1
13
91,777
We ran the full annex against 11 production DEX router agent skills:
🥇 @odosprotocol 88 B
🥈 SushiSwap 71 C
🥉 CowSwap 70 C
Uniswap 49 D
PancakeSwap 49 D
KyberSwap 49 D
Across 49 D
deBridge 49 D
0x 48 D
LI.FI 45 D
1inch 26 F
2
2
6
7,175
Audit your own skill in 5 seconds:
npx agentsec
Full per-rule matrix, JSON + HTML + SARIF outputs, OWASP-aligned scoring.
github.com/semiotic-ai/agent…
Building an agent skill or auditing one? agentsec.sh
DMs open for outreach + securing agent skills.
2
229