[Security] agent_registry register_transaction and complete_transaction are unauthenticated, letting reputation be fabricated

[david87131]

Why this matters

In agent.rs, register_transaction never calls require_auth on anyone, so any caller can create AgentTransaction records with attacker-chosen agent and parties. complete_transaction only compares the supplied agent to stored state but never calls agent.require_auth, so anyone can complete on the agent's behalf and bump completed_agreements. Combined with rate_agent (no rater != agent guard), one actor can manufacture transactions, complete them, and rate any verified agent at will.

Acceptance criteria

• register_transaction authenticates a legitimate party and validates parties are distinct/real
• complete_transaction calls require_auth on the agent or authorized completing party
• rate_agent rejects self-rating and only allows a real party of a completed transaction to rate, once
• Tests assert a non-party cannot register, complete, or rate, and an agent cannot rate itself

Files to touch

• contracts/agent_registry/src/agent.rs

Out of scope

• Weighting reputation by transaction value
• Cross-contract verification against the rental contract

[david87131]

Superseded by the consolidated contract audit issues (#67-#104) which cover the same findings more comprehensively. Closing to avoid duplicate coverage.

[Vyacheslav-Tomashevskiy]

Opened #211 which addresses all four acceptance criteria: register_transaction now authenticates the caller (must be the agent or a listed party) and validates parties (non-empty, distinct, agent excluded); complete_transaction calls agent.require_auth() and rejects double-completion; rate_agent rejects self-rating. 8 new adversarial tests, cargo test 31 passed, fmt+clippy clean.

RTC payout wallet: RTCd1554f0f35576faf01d386a6be1c947f560dd0b7