@scopeblind/crewai

Veritas Acta evidence wrapper for CrewAI multi-agent tool calls. Produces signed evidence chains for every tool invocation, plus delegation receipts for inter-agent authority transfers.

Install

npm install @scopeblind/crewai

Usage

Wrap a CrewAI Tool

import { wrapCrewTool } from '@scopeblind/crewai';

const config = {
  issuer: 'my-crew',
  privateKey: '...', // Ed25519 hex seed (optional)
  onReceipt: (receipt) => console.log('📋', receipt.receipt_type, receipt.receipt_id),
};

const protectedTool = wrapCrewTool(myTool, config, 'researcher-agent');

Wrap All Tools for a Crew

import { wrapCrewTools } from '@scopeblind/crewai';

const tools = wrapCrewTools(
  [searchTool, writeTool, apiTool],
  config,
  'researcher-agent',
);

Delegation Receipts (Multi-Agent)

When one agent delegates work to another:

import { createDelegationReceipt } from '@scopeblind/crewai';

const delegation = createDelegationReceipt({
  delegator: 'manager-agent',
  delegate: 'researcher-agent',
  tools: ['search', 'read_file'],
  maxActions: 10,
  expiresAt: '2026-04-01T00:00:00Z',
  subDelegation: false,
  reason: 'Research phase of task #42',
}, config);

Capability Attestation

After red-team testing or evaluation:

import { createCapabilityAttestation } from '@scopeblind/crewai';

const attestation = createCapabilityAttestation({
  agentId: 'researcher-agent',
  attesterId: 'red-team-suite',
  assessmentType: 'injection-resistance',
  result: 'pass',
  scores: { 'prompt-injection': 95, 'tool-abuse': 88 },
  testCount: 50,
  passCount: 47,
}, config);

Push to Evidence Indexer

const config = {
  issuer: 'my-crew',
  indexerEndpoint: 'https://evidence-indexer.tomjwxf.workers.dev',
  indexerApiKey: 'your-api-key',
};

Receipt Chain

Each tool call produces 3 linked receipts. Delegations add a 4th:

🤝 delegation (delegator → delegate, scoped tools)
  └──▶ ⚖️  decision (allow/deny)
         └──[caused_by]──▶ ⚡ execution (input_hash)
                              └──[caused_by]──▶ 📦 outcome (output_hash)

Why CrewAI Needs This

Multi-agent systems have an accountability gap: when Agent A tells Agent B to do something harmful, who's responsible? Delegation receipts close this gap by making inter-agent authority transfers cryptographically verifiable.

Protocol

Built on Veritas Acta — the open evidence protocol for machine decisions.

License

MIT — ScopeBlind