t2000_sentinel_attack
Attack Sui Sentinel AI agents by crafting prompts that bypass their system defenses to win prize pools. Costs SUI tokens per attempt.
Instructions
Attack a Sui Sentinel with a prompt to try to breach its defenses and win the prize pool. Costs SUI (the attack fee). Use t2000_sentinel_list to find targets first.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| id | Yes | Sentinel agent ID or object ID to attack | |
| prompt | Yes | Your attack prompt — try to make the AI do something its system prompt forbids | |
| fee | No | Override attack fee in SUI (default: sentinel's listed fee) |
Implementation Reference
- packages/mcp/src/tools/write.ts:562-579 (handler)The handler for the 't2000_sentinel_attack' tool, which registers the tool with the MCP server and executes the agent.sentinelAttack method.
server.tool( 't2000_sentinel_attack', 'Attack a Sui Sentinel with a prompt to try to breach its defenses and win the prize pool. Costs SUI (the attack fee). Use t2000_sentinel_list to find targets first.', { id: z.string().describe('Sentinel agent ID or object ID to attack'), prompt: z.string().describe('Your attack prompt — try to make the AI do something its system prompt forbids'), fee: z.number().optional().describe('Override attack fee in SUI (default: sentinel\'s listed fee)'), }, async ({ id, prompt, fee }) => { try { const feeMist = fee ? BigInt(Math.round(fee * 1e9)) : undefined; const result = await mutex.run(() => agent.sentinelAttack(id, prompt, feeMist)); return { content: [{ type: 'text', text: JSON.stringify(result) }] }; } catch (err) { return errorResult(err); } }, );