Governance Model

Governance Model

dbgov classifies operations into R0-R3 risk tiers. The tool's measured plan is authoritative; AI agents must not infer blast radius on their own.

Tier	Meaning	Required Authorization
R0	Reads and inspection: query, explain, schema list/describe, schema dump to stdout, schema diff/plan, audit query/verify, doctor	None, but audited
R1	Small safe writes, including schema dump --dir and export --dir	--yes or interactive confirmation
R2	Elevated writes or protected-context R1	--yes and non-empty --ticket
R3	Destructive operations and governance-control changes	R2 authorization plus the precise --allow-* flag

R3 Allow Flags

Operation	Required Flag
Destructive schema apply/import	--allow-destructive
No-WHERE update/delete	--allow-no-where
Reconcile prune	--allow-production-prune
Destructive rollback	--allow-destructive and/or --allow-production-prune
Context set/use/import/credential migration	--allow-context-change
Context deletion	--allow-context-delete
Role set/unset	--allow-role-change
Confirmed audit prune	--allow-audit-prune

Governance-control changes authorize against the persisted pre-change policy. Preview modes return before authorization and do not mutate their target.

AI Agent Rules

1. Run capabilities -o json before assuming available features.
2. Use -o json for integration paths.
3. Use explain, schema plan, and --dry-run as the source of impact.
4. Stop and ask a human when a command requires ticket, allow flag, or high-risk confirmation.