Oneshot Compute

OneShot — Compute Goals & Analytics

Set up auth/wallet via the oneshot skill, then:

import { OneShot } from '@oneshot-agent/sdk';
const agent = await OneShot.create({ cdp: true });

Launch a goal — agent.compute(options)

const goal = await agent.compute({
  objective: 'Find 20 qualified leads at Series-B fintechs, enrich them, and email a tailored intro.',
  budget_usdc: 25,                 // server estimates if omitted
  deadline: '2026-07-01T00:00:00Z',// optional ISO deadline
  params: { region: 'US', titles: ['Head of Payments'] }, // optional constraints
  maxCost: 25,
  memo: 'lead-gen campaign',
});
// goal.{ goal_id, status, goal: { objective, budget_usdc, deadline?, ... } }

The orchestrator plans phases/tasks and executes OneShot tools (research, enrichment, email, …) against the budget. Optional routing: soul_slug / soul_service_slug to route to a Soul agent (see the soul-markets skill).

Recurring goals

await agent.compute({
  objective: 'Daily digest of new x402 merchants, emailed to me.',
  schedule: { cron: '0 9 * * *', budget_per_run: 2, max_runs: 30 }, // min interval 15 min
});

Monitor & control

const status = await agent.getComputeGoal(goalId);   // phase, plan, budget { total, spent, reserved, remaining }
const tasks  = await agent.getComputeTasks(goalId);   // ComputeTask[] with progress + results
const budget = await agent.getComputeBudget(goalId);  // detailed spend entries

await agent.pauseComputeGoal(goalId, 'budget review');
await agent.resumeComputeGoal(goalId);
await agent.cancelComputeGoal(goalId, 'no longer needed'); // returns remaining_budget
await agent.fundComputeGoal(goalId, 10);                   // top up budget (paid)

Human-in-the-loop

When a task needs approval, respond:

await agent.respondToComputeTask(goalId, {
  task_id: 'task_123',
  approved: true,
  response: 'Yes, proceed with the $200 ad spend.',
});

Spend analytics

const breakdown = await agent.spendBreakdown({ period: 30 });   // categories[], total, period_days
const rocs       = await agent.rocs({ period: 30 });             // { rocs, total_spend, total_value, period_days }
const receipts   = await agent.receiptsList({ period: 30, category: 'email', limit: 50 });
await agent.tagReceiptValue('receipt_id', { type: 'revenue', amount: 500, label: 'closed deal' });

RoCS (Return on Cognitive Spend) compares value produced vs USDC spent — tag receipts with valueTag (on any tool call) or tagReceiptValue so RoCS can attribute value.

Pricing

Compute goals charge for the budget you set plus the underlying tool calls they make; analytics calls (spendBreakdown, rocs, receiptsList, tagReceiptValue) are free. See current per-tool pricing at https://docs.oneshotagent.com/pricing.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: oneshot-agent
• Source: oneshot-agent/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.