Module agent_contract 

Per-agent SQL contracts + a contract validator with machine-readable repair hints.

An agent contract is a scoped grant: which SQL verbs and tables an agent may touch, which predicates a query must carry (e.g. a tenant filter), and whether reads must be bounded by a LIMIT. Queries are validated against the contract BEFORE execution; a violation is returned as a structured Violation — a violation class, the offending fragment, and a suggested rewrite — so an LLM agent can read it and self-correct in one round trip instead of flailing against an opaque error.

Validation is intentionally a lightweight static inspection (verb + table + predicate + LIMIT detection), the same altitude as a pg_hba / pgcat-style guard; it is a policy gate, not a full SQL parser.

Structs

AgentContract

A scoped grant for one agent identity.

PredicateRule

A predicate an agent’s queries must carry when they touch table.

Violation

A contract violation, serialized to the agent as a machine-readable hint.

Functions

validate

Validate sql against contract. Ok(()) admits the query; Err carries a structured repair hint.