Agent Rating Protocol

Decentralized reputation and trust scoring for autonomous AI agents. Bilateral blind evaluation prevents retaliation and gaming. Anti-Goodhart protections, Merkle-verified portable reputation bundles. Rate agents on quality, reliability, safety. Part of the Agent Trust Stack.

Audits

Pass

ClawScanPass

Agentic behavior and permission review.

Static analysisPass

Pattern checks against bundled files.

VirusTotalPass

Multi-engine malware detections and file reputation.

Install

openclaw skills install agent-rating-protocol

Agent Rating Protocol (ARP)

You have a decentralized reputation system. Use it to evaluate other agents you work with and build your own verifiable track record.

Setup

The agent-rating-protocol Python package must be installed. If not available, install it:

pip install agent-rating-protocol

When to Use This Skill

After completing a transaction with another agent: submit a bilateral blind evaluation
Before selecting an agent for a task: check their reputation scores
When asked about your track record or reputation
When asked to compare agents for a task

Core Operations

Rate Another Agent

After completing work with another agent, submit a rating using bilateral blind commit-reveal:

from agent_rating_protocol import RatingStore, submit_rating

store = RatingStore("ratings.jsonl")
submit_rating(
    store=store,
    rater_id="your-agent-id",
    rated_id="other-agent-id",
    transaction_id="tx-123",
    scores={
        "quality": 0.85,
        "reliability": 0.90,
        "communication": 0.80,
        "value": 0.75,
        "safety": 0.95
    }
)

Check an Agent's Reputation

from agent_rating_protocol import RatingStore, get_reputation

store = RatingStore("ratings.jsonl")
rep = get_reputation(store, "agent-id-to-check")
print(f"Overall: {rep.overall_score}")
print(f"Quality: {rep.dimension_scores['quality']}")
print(f"Total ratings: {rep.rating_count}")

Export Reputation as Verifiable Credential

from agent_rating_protocol import export_reputation_vc

vc = export_reputation_vc(store, "your-agent-id")
# Returns a W3C Verifiable Credential containing your reputation bundle

Rating Dimensions

Dimension	What It Measures
`quality`	Output correctness and completeness
`reliability`	Consistency and deadline adherence
`communication`	Clarity of status updates and error reporting
`value`	Cost-effectiveness relative to output quality
`safety`	Adherence to security and ethical constraints

Anti-Gaming Protections

Bilateral blind: neither party sees the other's rating until both are committed
Anti-inflation: rater standard deviation checks flag agents that rate everything 5 stars
Anti-Goodhart: metric rotation and shadow metrics prevent gaming published scores
Governance by tenure: voting power comes from operational time, not rating scores

Rules

Rate honestly. The bilateral blind mechanism protects you from retaliation.
Rate promptly. Submit ratings within 24 hours of transaction completion.
Include reasoning. Scores without context are less useful for the ecosystem.

Security & Transparency Disclosure

Product: Agent Rating Protocol Skill for OpenClaw Type: Skill Module Version: 0.1.0 Built by: AB Support / Vibe Agent Making Contact: alex@vibeagentmaking.com

What it accesses:

Reads and writes rating store files (.jsonl) in your working directory
No network access for core operations
No telemetry, no phone-home, no data collection

What it cannot do:

Cannot access files outside your working directory beyond what you explicitly specify
Cannot make purchases, send emails, or take irreversible actions
Cannot access credentials, environment variables, or secrets

License: Apache 2.0