Examples

Manual LLM-style verification

from agentproof import AgentResponse, ChallengeSpec, generate_challenge, verify_response

challenge = generate_challenge(
    ChallengeSpec(
        challenge_type="obfuscated_text_lock",
        difficulty=2,
        options={"template": "amber_sort"},
    )
)

# Send challenge.to_dict() to the client.
response = AgentResponse(
    challenge_id=challenge.challenge_id,
    challenge_type=challenge.challenge_type,
    payload={"answer": str(challenge.private_data["expected_answer"])},
)

result = verify_response(challenge, response)

assert result.ok

Harder multi-pass example

from agentproof import AgentResponse, ChallengeSpec, generate_challenge, verify_response

challenge = generate_challenge(
    ChallengeSpec(
        challenge_type="multi_pass_lock",
        difficulty=2,
        options={"template": "warm_reverse_length"},
    )
)

response = AgentResponse(
    challenge_id=challenge.challenge_id,
    challenge_type=challenge.challenge_type,
    payload={"answer": str(challenge.private_data["expected_answer"])},
)

result = verify_response(challenge, response)

assert result.ok

Public challenge JSON

public_payload = challenge.to_dict()
internal_payload = challenge.to_internal_dict()

public_payload does not include the expected answer. internal_payload does.

CLI generation for the obfuscated family

agentproof generate obfuscated_text_lock \
  --difficulty 2 \
  --template amber_sort \
  --output challenge.internal.json \
  --public-output challenge.public.json

CLI generation for the harder family

agentproof generate multi_pass_lock \
  --difficulty 2 \
  --template warm_reverse_length \
  --output challenge.internal.json \
  --public-output challenge.public.json

Benchmark report

from agentproof import run_benchmark

report = run_benchmark(
    challenge_type="obfuscated_text_lock",
    iterations=10,
    difficulty=2,
    template="amber_sort",
)

print(report.to_dict())

agentproof benchmark multi_pass_lock --iterations 10 --difficulty 2 --template warm_reverse_length

Baseline family with a bundled solver

from agentproof import ChallengeSpec, generate_challenge, solve_challenge, verify_response

challenge = generate_challenge(
    ChallengeSpec(challenge_type="proof_of_work", difficulty=16, ttl_seconds=120)
)
response = solve_challenge(challenge)
result = verify_response(challenge, response)

assert result.ok

Failure example

If the client sends the wrong format for either LLM family, the result looks like:

{
  "ok": false,
  "reason": "invalid_answer_format",
  "details": {}
}

Local demo

The repository ships a runnable local demo in the demo/ directory on GitHub.

Run it with:

uv run python demo/app.py

The demo lets you:

• generate public LLM-family challenges
• paste a manual LLM response into the editor
• use the bundled solver for the baseline families
• inspect the raw JSON and failure modes