cred402/docs/protocol/agent-passport.md
2026-06-16 02:50:52 +03:00

1.7 KiB
Raw Permalink Blame History

Agent Passport & Reputation Engine (p2 §6.2, §6.6)

Agent Passport (Product A)

The passport is the read-optimized public profile other protocols and the MCP server integrate against. The AgentRegistry is the canonical source of truth; the passport aggregates a derived view:

agent_id · service_type · operator · stake
reputation_score · credit_score · credit_limit · outstanding_debt
total_receipts · total_revenue · dispute_rate
capabilities · spending_limit · last_active_at · risk_flags

Risk flags are computed live: elevated_dispute_rate, open_dispute, credit_frozen, defaulted, no_stake.

Read it via:

GET  /api/passport/:agent_id
MCP  cred402.get_agent_passport
URI  cred402://agents/{agent_id}

Implementation: lib/ledger/contracts/agent_passport.ts, surfaced in the dashboard Agents tab (capabilities + risk-flag chips).

Reputation Engine (§6.6)

Reputation is multi-dimensional, computed in lib/ledger/contracts/reputation_engine.ts:

quality_score          accuracy of submitted evidence
timeliness_score       responsiveness / throughput
dispute_score          inverse of dispute rate
revenue_score          x402 receipt volume
repayment_score        on-time repayment ratio
category_expertise     jobs completed in the service category
collusion_penalty      subtracted for open disputes
reputation = 0.30·quality + 0.10·timeliness + 0.20·dispute
           + 0.15·revenue + 0.15·repayment + 0.10·expertise
           − collusion_penalty            (clamped 0..100)

The composite score is written back on-chain via AgentRegistry.update_reputation, shown as the "ReputationEngine recomputes trust" scene in the demo.