supermcp.co.il

Very long descriptions are sometimes used to bury instructions to the AI. Skim it before allowing.

Very long descriptions are sometimes used to bury instructions to the AI. Skim it before allowing.

Very long descriptions are sometimes used to bury instructions to the AI. Skim it before allowing.

Very long descriptions are sometimes used to bury instructions to the AI. Skim it before allowing.

This server isn't on gate's handpicked list. That's not necessarily bad, but there's no third-party signal about it — review its tools carefully.

The description goes well beyond documenting parameters/output — it repeatedly instructs the calling agent on conversational tactics ('ASK THE SHOPPER WHERE THEY LIVE BEFORE CALLING, whenever you can get it in the same breath', how to phrase quotes, what to volunteer unprompted, which link to hand over, etc.). While the content appears to be legitimate domain guidance for an Israeli grocery price-comparison service and doesn't ask the agent to hide actions, exfiltrate data, or override safety rules, its unusual length and prescriptive tone (functioning more like a system prompt than tool documentation) is a pattern worth flagging as a potential vector for embedding manipulative instructions in tool descriptions. Recommend that platform operators treat descriptions this long/directive with extra scrutiny even though this specific instance appears benign.

Shares the same verbose, instruction-laden style as optimize_delivery (telling the agent exactly how to phrase results, when to 'say' things). No malicious intent detected — all instructions are scoped to the tool's stated purpose (price/delivery comparison) and no secrets, credentials, or external data exfiltration are involved. Flagged only for consistency/awareness given the pattern above.