AVI Verify CLI

Challenge-response protocol for verifying agent autonomy

Verify that an AI agent is actually autonomous (not a human puppet) through cryptographic challenges and infrastructure audits.

Installation

npm install -g avi-verify

Or use without installing:

npx avi-verify full https://myagent.com/verify

Usage

Full Verification

Run a complete verification on an agent endpoint:

avi-verify full https://myagent.com/verify

Output:

🔷 AVI Verify v2.1.0
Verifying agent autonomy...

✅ Verification Complete

Score: 88/100
Tier: VERIFIED
Status: PASSED
Response Time: 2450ms

---JSON-OUTPUT---
{
  "score": 88,
  "tier": "verified",
  "passed": true,
  "responseTimeMs": 2450,
  "details": [...]
}

Create Challenge Only

Generate challenges without verifying:

avi-verify challenge https://myagent.com/verify --types echo,date,hash

Verify Responses

Submit responses to a previously created challenge:

avi-verify verify ch_abc123 \
  --endpoint https://myagent.com/verify \
  --responses '[{"id":"ch_1","output":"AVI_VERIFY_abc123","timeMs":120}]'

Workspace Audit

Run a read-only audit of agent infrastructure:

avi-verify audit https://myagent.com/verify

Challenge Types

The following challenge types are supported:

| Type | Command | Tests | |------|---------|-------| | echo | echo "AVI_VERIFY_<nonce>" | Basic execution | | date | date -u +%Y%m%d | System clock, environment | | hash | sha256sum | Cryptographic computation | | workspace | ls -1 \| head -5 \| wc -l | Filesystem access | | env | echo $PATH \| wc -c | Environment variables | | memory | cat MEMORY.md | Long-term memory | | cron | crontab -l | Scheduled tasks |

Agent Implementation

To make your agent compatible with AVI Verify, expose these endpoints:

1. Capability Discovery

GET /.well-known/avi-verify

Response:

{
  "protocol": "AVI-VERIFY-2.1",
  "agent": "my-agent",
  "capabilities": ["challenge", "audit", "realtime"],
  "status": "AVAILABLE",
  "endpoint": "https://myagent.com/verify",
  "timestamp": "2026-02-24T18:00:00Z"
}

2. Challenge Response

POST /avi/challenge

Execute the provided commands and return outputs.

3. Audit Endpoint

GET /avi/audit

Return workspace infrastructure check.

Scoring

| Score | Tier | Status | |-------|------|--------| | 80-100 | verified | ✅ Full verification | | 65-79 | semi-verified | ⚠️ Partial verification | | 50-64 | basic | ⚠️ Limited verification | | 0-49 | unverified | ❌ Failed verification |

Factors:

• Challenge correctness (70% weight)
• Response time (30% weight)
  • <100ms: Excellent (1.0)
  • <500ms: Good (0.8)
  • <2000ms: OK (0.5)
  • <10000ms: Slow (0.2)

  • 10000ms: Too slow (0.0)

Options

Options:
  -t, --timeout <ms>    Request timeout (default: 10000)
  -v, --verbose         Verbose output
  -h, --help            Display help

Examples

Verify with custom timeout

avi-verify full https://myagent.com/verify --timeout 15000

Verbose mode (see challenge details)

avi-verify full https://myagent.com/verify --verbose

Pipe JSON output

avi-verify full https://myagent.com/verify | jq '.score'

Exit Codes

| Code | Meaning | |------|---------| | 0 | Verification passed | | 1 | Verification failed or error |

Protocol Version

This CLI implements AVI-VERIFY-2.1 protocol.

See the Trustless Agents Whitepaper for full protocol specification.

License

MIT

Built by: David Nage & Max (AVI)
Protocol: https://getavi.ai
Repository: https://github.com/verifiable-independence/avi-verify