Skip to content
Agent0 SDK

Feedback

Complete guide to giving feedback, managing reputation, and using feedback data to evaluate agents.

Feedback ID Format

Section titled “Feedback ID Format”

Feedback IDs follow the format "agentId:clientAddress:feedbackIndex":

  • agentId: The agent’s ID in "chainId:tokenId" format (e.g., "11155111:123")
  • clientAddress: Ethereum address of the feedback giver (normalized to lowercase)
  • feedbackIndex: Sequential index of feedback from this client to this agent (0-based)
  • Example: "11155111:123:0x742d35cc6634c0532925a3b844bc9e7595f0beb7:0" = First feedback from client 0x742d... to agent 11155111:123

See Models Reference for detailed ID format documentation.

Prepare Feedback

Section titled “Prepare Feedback”

The prepareFeedback() method creates a feedback file. Only score is mandatory - all other fields are optional.

# Minimal feedback (only mandatory score)
feedback_file = sdk.prepareFeedback(
    agentId="11155111:123",
    score=85  # 0-100 (MANDATORY)
)


# With optional categorization
feedback_file = sdk.prepareFeedback(
    agentId="11155111:123",
    score=85,  # 0-100 (MANDATORY)
    tags=["data_analyst", "finance"],  # Optional
    capability="tools",  # Optional
    skill="python"  # Optional
)

Give Feedback

Section titled “Give Feedback”

With Authorization

Section titled “With Authorization”

Agents sign feedback authorization for clients:

# Agent signs feedback auth for client
feedbackAuth = agent_sdk.signFeedbackAuth(
    agentId="11155111:123",
    clientAddress=client_address,
    expiryHours=24
)


# Client prepares and submits feedback
feedback = client_sdk.giveFeedback(
    agentId="11155111:123",
    feedbackFile=feedback_file,
    feedbackAuth=feedbackAuth
)

Feedback Parameters

Section titled “Feedback Parameters”

Mandatory fields:

  • agentId - The agent’s ID
  • score - Rating from 0-100

Optional fields:

  • tags - List of categorization tags (e.g., [“data_analyst”, “finance”])
  • capability - MCP capability type (“tools”, “prompts”, “resources”)
  • name - MCP tool/resource/prompt name
  • skill - A2A skill identifier
  • task - A2A task identifier
  • context - Dictionary with additional context information
  • proofOfPayment - Payment proof data for x402 payments
# Example with all optional fields
feedback_file = sdk.prepareFeedback(
    agentId="11155111:123",           # Mandatory
    score=85,                          # Mandatory: 0-100
    tags=["data_analyst", "finance"],  # Optional: Categorization
    capability="tools",                 # Optional: MCP capability
    name="code_generation",            # Optional: MCP tool name
    skill="python",                     # Optional: A2A skill
    task="debug_code",                  # Optional: A2A task
    context={"environment": "dev"}      # Optional: Additional context
)


# Minimal example (only mandatory fields)
feedback_file = sdk.prepareFeedback(
    agentId="11155111:123",
    score=85
)

Read Feedback

Section titled “Read Feedback”

Get Single Feedback

Section titled “Get Single Feedback”
# Read feedback by ID
feedback = sdk.getFeedback("11155111:123:0xClient:0")
print(f"Score: {feedback.score}")
print(f"Tags: {feedback.tags}")

Search Feedback

Section titled “Search Feedback”
# Search feedback with filters
results = sdk.searchFeedback(
    agentId="11155111:123",
    capabilities=["tools"],
    skills=["python"],
    tags=["data_analyst"],
    minScore=80,
    maxScore=100
)


for fb in results:
    print(f"{fb.score}/100: {fb.tags}")

Reputation Summary

Section titled “Reputation Summary”
# Get aggregated reputation
summary = sdk.getReputationSummary("11155111:123")


print(f"Average score: {summary['averageScore']}")
print(f"Total feedback: {summary['totalFeedback']}")
print(f"Score distribution: {summary['scoreDistribution']}")
Section titled “Agent Reputation Search”

Find agents by reputation:

# Find highly-rated agents
results = sdk.searchAgentsByReputation(
    minAverageScore=80,
    tags=["enterprise"],
    capabilities=["code_generation"]
)


for agent in results['items']:
    print(f"{agent.name}: {agent.extras['averageScore']}")

Manage Feedback

Section titled “Manage Feedback”

Append Response

Section titled “Append Response”
# Agent responds to feedback with refund acknowledgment
updated_feedback = sdk.appendResponse(
    agentId="11155111:123",
    clientAddress=client_address,
    feedbackIndex=0,
    response={
        "message": "We apologize for the poor service. A full refund has been processed.",
        "refundTxHash": "0x1234567890abcdef...",
        "refundAmount": "0.1 ETH",
        "timestamp": int(time.time())
    }
)

Revoke Feedback

Section titled “Revoke Feedback”
# Revoke feedback
sdk.revokeFeedback(
    agentId="11155111:123",
    clientAddress=client_address,
    feedbackIndex=0
)

Feedback Data Structure

Section titled “Feedback Data Structure”
@dataclass
class Feedback:
    id: str                        # Unique feedback ID
    agentId: str                   # Agent identifier
    reviewer: str                  # Client address
    score: float                   # 0-100 rating (MANDATORY)
    tags: Optional[List[str]]      # Categorization tags (optional)
    capability: Optional[str]      # MCP capability type (optional)
    name: Optional[str]            # MCP tool/resource/prompt name (optional)
    skill: Optional[str]           # A2A skill (optional)
    task: Optional[str]           # A2A task (optional)
    context: Optional[dict]       # Additional context (optional)
    fileURI: Optional[str]         # IPFS/HTTPS URI (if feedback file exists)
    createdAt: int                 # Timestamp
    answers: List[dict]            # Responses
    isRevoked: bool                # Revocation status

Note: Only score is mandatory when creating feedback. All other fields are optional and can be omitted for minimal on-chain-only feedback.

Next Steps

Section titled “Next Steps”