vaultkit 1.1.6 → 1.2.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 42e3300980d3a6c3041304eea365fad63d8bfb2a5b56a49c32ccb89d21366ec8
-  data.tar.gz: f353b00284834a0c3eed06f4af09da181d4bb3f622f61bbcfcc949d4014bf109
+  metadata.gz: 8490da7d9c28171310faf52f3a8b754dd00798d75837ec76eee4b3f3b5ce6aab
+  data.tar.gz: 53ce9b61c202b464b66d9fe3f4d195597a6992c20d5f707247a33da93f196e89
 SHA512:
-  metadata.gz: fcf5ba06e5f6256f705de1d228b70c52a94fe4ce26c11d82d3e2dbab30397119ea33d511acdac0d1e383cb3e501c1286e303490f3035a2b9991cee8b04863d87
-  data.tar.gz: 537cf82edb8000c9a0c5d273c14e1958198022a16a6918b5f5f2ac7f673b1f9df7b1b26c5daf95acbb6bc0a6f3395ff430e73cc09ee6f8934f2179336b09789a
+  metadata.gz: e3f239d9f906313cac41a736d2e352b8be1b429e04fe0b08e8bb24190ee4e186510dd6141306dc1834da7750a2d710e39e246fcfe52b76c1915b2c5196731c1e
+  data.tar.gz: 822c8f6472e7bd4372b439bde33964329cbf93be371407873bbf2725750122633ba145d3d1635a84aebe0454a97d389f13a2014e1999cc2da96dfe60dc3b0a68

data/README.md

@@ -701,6 +701,25 @@ vkit scan production_pg
 # Summary: 2 new datasets, 3 new fields, 1 modified field
 ```
 
+**Classifier Suggestions:**
+
+A rescan keeps the labels people reviewed (from a deployed bundle or a manual edit). When the classifier disagrees with one of them, the scan lists it as a suggestion under its dataset, marked `?`. Suggestions are not changes: they don't count toward the changed total, and applying the scan leaves them alone.
+
+```bash
+vkit scan production_pg
+
+# Output:
+#   settings                 ✓
+#     ? key_value  low/internal → high/credential (suggested)
+#
+#   No changes detected
+#
+#   ? = classifier suggestion, not applied. To accept one, edit
+#     datasets/registry.yaml and redeploy the bundle.
+```
+
+To accept a suggestion, change the field's sensitivity and tags in `datasets/registry.yaml`, then build and deploy the bundle. Control planes that predate suggestions don't send them, and the output is unchanged.
+
 **Apply Changes:**
 ```bash
 vkit scan production_pg --mode apply

data/lib/vkit/cli/base_cli.rb

@@ -331,11 +331,17 @@ module Vkit
           option :name, required: true, desc: "Human-readable name (e.g. billing-bot)"
           option :expires_in, type: :string, desc: "Token lifetime (e.g. 1h, 24h, 30d)"
           option :role, type: :string, default: "agent", desc: "Role assigned to this token"
+          option :tenant, type: :string, desc: "Tenant this token acts for on tenant-scoped datasets"
+          option :clearance, type: :string, desc: "Clearance policies see: low, medium or high"
+          option :region, type: :string, desc: "Region policies see: EU, US, CA or APAC"
           def create
             Commands::AgentTokensCreateCommand.new.call(
               name: options[:name],
               expires_in: options[:expires_in],
-              role: options[:role]
+              role: options[:role],
+              tenant: options[:tenant],
+              clearance: options[:clearance],
+              region: options[:region]
             )
           end
 

data/lib/vkit/cli/commands/agent_tokens_create_command.rb

@@ -26,6 +26,9 @@ module Vkit
 
             puts "🧠 Name:       #{token["name"]}"
             puts "🎭 Role:       #{token["role"]}"
+            puts "🏢 Tenant:     #{token["tenant"]}" if token["tenant"]
+            puts "🛡  Clearance:  #{token["clearance"]}" if token["clearance"]
+            puts "🌍 Region:     #{token["region"]}" if token["region"]
             puts "🆔 Token ID:   #{token["id"]}"
             puts "🏷  Prefix:     #{token["prefix"]}"
             puts "⏳ Expires At: #{token["expires_at"] || "never"}"
@@ -47,7 +50,10 @@ module Vkit
           body = {
             name: options.fetch(:name),
             role: options[:role],
-          }
+            tenant: options[:tenant],
+            clearance: options[:clearance],
+            region: options[:region]
+          }.compact
 
           if options[:expires_in]
             body[:expires_at] = parse_expires_in(options[:expires_in]).iso8601

data/lib/vkit/cli/commands/scan_command.rb

@@ -19,51 +19,88 @@ module Vkit
               body: { datasource: datasource_name, mode: mode }
             )
 
-            diff     = response["diff"] || {}
-            datasets = diff["datasets"] || []
-            changed  = datasets.select { |d| d["changes"].values.any?(&:any?) }
-            clean    = datasets.reject { |d| d["changes"].values.any?(&:any?) }
+            print_diff(response)
+          end
+        end
 
-            if changed.empty?
-              datasets.each { |d| puts "  #{d["name"].ljust(24)} ✓" }
-              puts
-              puts "  No changes detected"
-            else
-              changed.each do |dataset|
-                added    = dataset.dig("changes", "added_fields")   || []
-                removed  = dataset.dig("changes", "removed_fields") || []
-                modified = dataset.dig("changes", "changed_fields") || []
-
-                puts "  #{dataset["name"]}"
-
-                added.each do |f|
-                  name = f.is_a?(Hash) ? f["name"] : f
-                  type = f.is_a?(Hash) && f["type"] ? " (#{f["type"]})" : ""
-                  puts "    + #{name}#{type}  ⚠️  unclassified"
-                end
-
-                removed.each do |f|
-                  name = f.is_a?(Hash) ? f["name"] : f
-                  puts "    - #{name}"
-                end
-
-                modified.each do |f|
-                  name = f.is_a?(Hash) ? f["name"] : f
-                  puts "    ~ #{name}"
-                end
-
-                puts
+        private
+
+        def print_diff(response)
+          diff     = response["diff"] || {}
+          datasets = diff["datasets"] || []
+          changed  = datasets.select { |d| d["changes"].values.any?(&:any?) }
+          clean    = datasets.reject { |d| d["changes"].values.any?(&:any?) }
+
+          if changed.empty?
+            datasets.each { |d| print_clean(d) }
+            puts
+            puts "  No changes detected"
+          else
+            changed.each do |dataset|
+              added    = dataset.dig("changes", "added_fields")   || []
+              removed  = dataset.dig("changes", "removed_fields") || []
+              modified = dataset.dig("changes", "changed_fields") || []
+
+              puts "  #{dataset["name"]}"
+
+              added.each do |f|
+                name = f.is_a?(Hash) ? f["name"] : f
+                type = f.is_a?(Hash) && f["type"] ? " (#{f["type"]})" : ""
+                puts "    + #{name}#{type}  ⚠️  unclassified"
+              end
+
+              removed.each do |f|
+                name = f.is_a?(Hash) ? f["name"] : f
+                puts "    - #{name}"
               end
 
-              clean.each { |d| puts "  #{d["name"].ljust(24)} ✓" }
+              modified.each do |f|
+                name = f.is_a?(Hash) ? f["name"] : f
+                puts "    ~ #{name}"
+              end
+
+              print_suggestions(dataset)
 
               puts
-              puts "  #{changed.size} changed · run with --apply to update baseline"
             end
 
+            clean.each { |d| print_clean(d) }
+
             puts
-            puts "  ✅ Applied" if response["applied"]
+            puts "  #{changed.size} changed · run with --apply to update baseline"
           end
+
+          # Suggestions are reviewed labels the classifier now disagrees with.
+          # Older control planes don't send them; applying a scan never changes them.
+          if datasets.any? { |d| suggestions_for(d).any? }
+            puts
+            puts "  ? = classifier suggestion, not applied. To accept one, edit"
+            puts "    datasets/registry.yaml and redeploy the bundle."
+          end
+
+          puts
+          puts "  ✅ Applied" if response["applied"]
+        end
+
+        def print_clean(dataset)
+          puts "  #{dataset["name"].ljust(24)} ✓"
+          print_suggestions(dataset)
+        end
+
+        def print_suggestions(dataset)
+          suggestions_for(dataset).each do |s|
+            puts "    ? #{s["name"]}  #{label(s["current"])} → #{label(s["suggested"])} (suggested)"
+          end
+        end
+
+        def suggestions_for(dataset)
+          dataset["suggestions"] || []
+        end
+
+        def label(classification)
+          classification ||= {}
+          tags = Array(classification["tags"])
+          [classification["sensitivity"], *(tags.join(",") unless tags.empty?)].compact.join("/")
         end
       end
     end

data/lib/vkit/policy/bundle_compiler.rb

@@ -3,15 +3,21 @@ require "yaml"
 require "digest"
 require "time"
 require_relative "policy_validator"
+require_relative "match_check"
+require_relative "tenancy_check"
+require_relative "validation_error"
 
 module Vkit
   module Policy
     class BundleCompiler
       FORMAT_VERSION = "v1"
 
+      # Raises if a policy can never match (see MatchCheck).
       def self.compile!(org_slug:, bundle_version:, policies_dir:, registry_dir:, source: {})
         policies = load_policies(policies_dir)
-        registry = load_registry(registry_dir)
+        raw_registry = load_raw_registry(registry_dir)
+        registry = normalize_registry(raw_registry)
+        check_matches!(policies, registry, raw_registry)
 
         bundle = {
           "bundle" => {
@@ -49,12 +55,25 @@ module Vkit
         end
       end
 
-      def self.load_registry(dir)
+      def self.check_matches!(policies, registry, raw_registry)
+        errors = MatchCheck.call(policies, registry).errors + TenancyCheck.call(raw_registry)
+        return if errors.empty?
+
+        raise ValidationError, <<~MSG
+          These policies or datasets would never protect anything as written:
+
+          #{errors.map { |e| "  - #{e}" }.join("\n")}
+
+          Fix or remove them. If they came from a policy pack, `vkit policy pack upgrade` installs
+          versions that can match.
+        MSG
+      end
+
+      def self.load_raw_registry(dir)
         path = File.join(dir, "registry.yaml")
         raise "Missing datasets/registry.yaml" unless File.exist?(path)
 
-        raw = YAML.load_file(path)
-        normalize_registry(raw)
+        YAML.load_file(path) || {}
       end
 
       # Normalization
@@ -138,8 +157,11 @@ module Vkit
           {
             "name" => name.to_s,
             "datasource" => data["datasource"].to_s,
+            # A field name confines every query to the requester's tenant;
+            # false (a deliberately shared table) only matters to TenancyCheck.
+            "tenant_column" => (data["tenant_column"] if data["tenant_column"].is_a?(String)),
             "fields" => normalize_fields(data["fields"] || {})
-          }
+          }.compact
         end
       
         datasources =

data/lib/vkit/policy/match_check.rb

@@ -0,0 +1,120 @@
+# frozen_string_literal: true
+
+require "set"
+
+module Vkit
+  module Policy
+    # Finds policies that can never match, by checking them against the
+    # registry being bundled and against what the control plane evaluates:
+    # a key the engine ignores, a tag or sensitivity no field can have, a
+    # dataset that isn't in the registry, or a context value VaultKit never
+    # sets. Requester clearance and region are set by admins on users and
+    # agent tokens (vaultkit-inc/vaultkit#76), so they're checked like any
+    # other value.
+    class MatchCheck
+      # Labels the scanner's classifier assigns. Always valid, so a pack can
+      # cover secrets before a database has any.
+      # unknown marks a column nobody has reviewed yet.
+      CLASSIFIER_TAGS = %w[pii financial credential internal unknown].freeze
+      SENSITIVITIES = %w[high medium low unknown].freeze
+
+      MATCH_KEYS = %w[dataset fields context].freeze
+      FIELD_KEYS = %w[any all contains sensitivity].freeze
+      CONTEXT_KEYS = %w[environment dataset_region requester_role requester_region requester_clearance time].freeze
+      ENVIRONMENTS = %w[production staging development sandbox].freeze
+      REGIONS = %w[EU US CA APAC].freeze
+      CLEARANCES = %w[low medium high].freeze
+
+      Result = Struct.new(:errors, keyword_init: true)
+
+      def self.call(policies, registry)
+        new(registry).call(policies)
+      end
+
+      def initialize(registry)
+        fields = Array(registry["datasets"]).flat_map { |ds| Array(ds["fields"]) }
+        @datasets = Array(registry["datasets"]).map { |ds| ds["name"].to_s }.to_set
+        @tags = (CLASSIFIER_TAGS + fields.flat_map { |f| Array(f["tags"]).map(&:to_s) }).to_set
+        @sensitivities = (SENSITIVITIES + fields.map { |f| f["sensitivity"].to_s }).reject(&:empty?).to_set
+      end
+
+      def call(policies)
+        errors = []
+        policies.each do |policy|
+          where = "#{policy['__file'] || policy['id']} (#{policy['id']})"
+          problems(policy).each { |p| errors << "#{where}: #{p}" }
+        end
+        Result.new(errors: errors)
+      end
+
+      private
+
+      def problems(policy)
+        match = policy["match"] || {}
+        return ["match must be a mapping"] unless match.is_a?(Hash)
+
+        unknown_keys(match.keys, MATCH_KEYS, "match") +
+          dataset_problems(match["dataset"]) +
+          field_problems(match["fields"]) +
+          context_problems(policy["context"], "context") +
+          context_problems(match["context"], "match.context") +
+          mask_problems(policy, match)
+      end
+
+      def unknown_keys(keys, allowed, path)
+        (keys.map(&:to_s) - allowed).map do |key|
+          next "#{path}.category isn't supported: the engine ignores it. Use `contains: [<category>]` instead." if key == "category"
+
+          "#{path}.#{key} isn't supported, so the engine ignores it. Allowed: #{allowed.join(', ')}"
+        end
+      end
+
+      def dataset_problems(dataset)
+        return [] if dataset.nil? || @datasets.include?(dataset.to_s)
+
+        ["dataset #{dataset} isn't in datasets/registry.yaml. Scan the datasource, or fix the name."]
+      end
+
+      def field_problems(fields)
+        return [] if fields.nil?
+        return ["match.fields must be a mapping"] unless fields.is_a?(Hash)
+
+        problems = unknown_keys(fields.keys, FIELD_KEYS, "match.fields")
+        tags = Array(fields["contains"]).map(&:to_s)
+        unknown = tags.reject { |t| @tags.include?(t) }
+        problems << "no field has the tag #{unknown.join(', ')}. Known tags: #{@tags.to_a.sort.join(', ')}" if unknown.any?
+        if fields.key?("sensitivity") && !@sensitivities.include?(fields["sensitivity"].to_s)
+          problems << "sensitivity #{fields['sensitivity']} never occurs. Use one of: #{@sensitivities.to_a.join(', ')}"
+        end
+        problems
+      end
+
+      def context_problems(context, path)
+        return [] if context.nil?
+        return ["#{path} must be a mapping"] unless context.is_a?(Hash)
+
+        unknown_keys(context.keys, CONTEXT_KEYS, path) + context.filter_map do |key, value|
+          case key.to_s
+          when "requester_clearance"
+            "#{path}.requester_clearance #{value} isn't one of: #{CLEARANCES.join(', ')}" unless CLEARANCES.include?(value.to_s)
+          when "requester_region"
+            "#{path}.requester_region #{value} isn't one of: #{REGIONS.join(', ')}" unless REGIONS.include?(value.to_s)
+          when "time"
+            "#{path}.time only works under match.context." if path == "context"
+          when "environment"
+            "#{path}.environment #{value} isn't one of: #{ENVIRONMENTS.join(', ')}" unless ENVIRONMENTS.include?(value.to_s)
+          when "dataset_region"
+            "#{path}.dataset_region #{value} isn't one of: #{REGIONS.join(', ')}" unless REGIONS.include?(value.to_s)
+          end
+        end
+      end
+
+      def mask_problems(policy, match)
+        return [] unless policy.dig("action", "mask") == true
+        return [] if match["fields"].is_a?(Hash) && (match["fields"].keys.map(&:to_s) & %w[any contains sensitivity]).any?
+
+        ["a mask policy needs match.fields (any, contains or sensitivity) to say which fields to mask"]
+      end
+    end
+  end
+end

data/lib/vkit/policy/packs/ai_safety/metadata.yaml

@@ -1,8 +1,8 @@
 __pack_meta:
   name: ai_safety
-  version: "1.1.0"
+  version: "2.0.0"
   layer: 30
-  description: "Secure defaults for AI/agent access: least-privilege, approvals for sensitive, environment controls."
+  description: "Agents need approval for personal, financial and unreviewed data in production, and for high-sensitivity data anywhere."
   dependencies:
     - starter
   priority_band:

data/lib/vkit/policy/packs/ai_safety/policies/01_require_approval_agents_sensitive_prod.yaml

@@ -0,0 +1,17 @@
+id: ai_require_approval_agents_sensitive_prod
+description: Agents need security approval for personal or financial data in production.
+priority: 1200
+
+match:
+  fields:
+    contains: [pii, financial]
+
+context:
+  environment: production
+  requester_role: agent
+
+action:
+  require_approval: true
+  approver_role: security
+  reason: Agents reading personal or financial data in production need security review.
+  ttl: 1h

data/lib/vkit/policy/packs/ai_safety/policies/02_require_approval_agents_high_sensitivity.yaml

@@ -0,0 +1,16 @@
+id: ai_require_approval_agents_high_sensitivity
+description: Agents need security approval for high-sensitivity data in every environment.
+priority: 1300
+
+match:
+  fields:
+    sensitivity: high
+
+context:
+  requester_role: agent
+
+action:
+  require_approval: true
+  approver_role: security
+  reason: Agents reading high-sensitivity data need security review, in any environment.
+  ttl: 1h

data/lib/vkit/policy/packs/ai_safety/policies/03_require_approval_agents_unknown_prod.yaml

@@ -0,0 +1,17 @@
+id: ai_require_approval_agents_unknown_prod
+description: Agents need security approval in production for columns nobody has reviewed yet.
+priority: 1400
+
+match:
+  fields:
+    sensitivity: unknown
+
+context:
+  environment: production
+  requester_role: agent
+
+action:
+  require_approval: true
+  approver_role: security
+  reason: This column's sensitivity is unknown, so an agent needs security review to read it in production.
+  ttl: 1h

data/lib/vkit/policy/packs/financial_compliance/metadata.yaml

@@ -1,8 +1,8 @@
 __pack_meta:
   name: financial_compliance
-  version: "1.0.0"
+  version: "2.0.0"
   layer: 40
-  description: "Financial controls: approvals, masking, audit-friendly TTLs for financial + payment data."
+  description: "Finance approval for financial data in production, and short-lived grants for financial fields."
   dependencies:
     - starter
   priority_band:

data/lib/vkit/policy/packs/financial_compliance/policies/01_require_approval_financial_prod.yaml

@@ -0,0 +1,16 @@
+id: fin_require_approval_financial_prod
+description: A finance manager approves requests for financial data in production.
+priority: 2100
+
+match:
+  fields:
+    contains: [financial]
+
+context:
+  environment: production
+
+action:
+  require_approval: true
+  approver_role: finance_manager
+  reason: Financial data in production needs finance approval.
+  ttl: 30m

data/lib/vkit/policy/packs/financial_compliance/policies/02_short_ttl_for_financial_grants.yaml

@@ -0,0 +1,12 @@
+id: fin_short_ttl_financial
+description: Grants that include financial fields expire after 30 minutes.
+priority: 2400
+
+match:
+  fields:
+    contains: [financial]
+
+action:
+  allow: true
+  reason: Financial grants expire quickly.
+  ttl: 30m

data/lib/vkit/policy/packs/starter/metadata.yaml

@@ -1,8 +1,8 @@
 __pack_meta:
   name: starter
-  version: "1.0.0"
+  version: "2.0.0"
   layer: 10
-  description: "Secure-by-default baseline policies for VaultKit."
+  description: "Secure defaults: secrets denied, PII and high-sensitivity fields masked, approval for high-sensitivity data in production, unreviewed columns flagged."
   dependencies: []
   priority_band:
     min: 0

data/lib/vkit/policy/packs/starter/policies/01_deny_credentials.yaml

@@ -0,0 +1,11 @@
+id: deny_credentials
+description: Never return passwords, tokens, API keys or other secrets.
+priority: 95
+
+match:
+  fields:
+    contains: [credential]
+
+action:
+  deny: true
+  reason: Credentials are never returned. Use the service that owns them instead.

data/lib/vkit/policy/packs/starter/policies/02_mask_pii_by_default.yaml

@@ -1,16 +1,15 @@
 id: mask_pii_by_default
-description: "Mask PII fields by default unless explicitly allowed."
-
+description: Mask personal data for every requester, including after approval.
 priority: 80
 
 match:
   fields:
-    contains: ["pii"]
+    contains: [pii]
 
 action:
   mask: true
-  reason: "PII is masked by default for safety."
+  reason: Personal data is masked by default.
   ttl: 1h
 
 masking:
-  default_method: redact
+  default_method: full

data/lib/vkit/policy/packs/starter/policies/03_mask_high_sensitivity.yaml

@@ -0,0 +1,15 @@
+id: mask_high_sensitivity
+description: Mask high-sensitivity fields, such as bank accounts, even when they aren't personal data.
+priority: 82
+
+match:
+  fields:
+    sensitivity: high
+
+action:
+  mask: true
+  reason: High-sensitivity data is masked by default.
+  ttl: 1h
+
+masking:
+  default_method: full

data/lib/vkit/policy/packs/starter/policies/{03_require_approval_high_sensitivity.yaml → 04_require_approval_high_sensitivity.yaml}

@@ -1,5 +1,6 @@
 id: require_approval_high_sensitivity
-description: "Require approval before accessing high sensitivity data."
+description: A data owner approves requests for high-sensitivity data in production.
+priority: 85
 
 match:
   fields:
@@ -8,10 +9,8 @@ match:
 context:
   environment: production
 
-priority: 85
-
 action:
   require_approval: true
   approver_role: data_owner
-  reason: "High sensitivity data requires manual review in production."
-  ttl: "1h"
+  reason: High-sensitivity data in production needs a data owner's approval.
+  ttl: 1h

data/lib/vkit/policy/packs/starter/policies/{06_default_ttl.yaml → 05_default_ttl.yaml}

@@ -1,10 +1,9 @@
 id: default_ttl_for_allows
-description: "Apply a default TTL to allow decisions."
+description: Grants for allowed requests last four hours.
+priority: 10
 
 match: {}
 
-priority: 10
-
 action:
   allow: true
-  ttl: "4h"
+  ttl: 4h

data/lib/vkit/policy/packs/starter/policies/06_flag_unknown.yaml

@@ -0,0 +1,12 @@
+id: flag_unknown_sensitivity
+description: Allow columns nobody has reviewed yet, and record why, so they show up in the audit log.
+priority: 20
+
+match:
+  fields:
+    sensitivity: unknown
+
+action:
+  allow: true
+  reason: This column's sensitivity is unknown. Review its label in datasets/registry.yaml.
+  ttl: 1h

data/lib/vkit/policy/schema/policy_bundle.schema.json

@@ -102,6 +102,11 @@
             "properties": {
               "name": { "type": "string", "minLength": 1 },
               "datasource": { "type": "string", "minLength": 1 },
+              "tenant_column": {
+                "type": "string",
+                "minLength": 1,
+                "description": "Column holding the tenant id. Every query on this dataset is confined to the requester's tenant."
+              },
               "fields": {
                 "type": "array",
                 "minItems": 1,

data/lib/vkit/policy/tenancy_check.rb

@@ -0,0 +1,54 @@
+# frozen_string_literal: true
+
+module Vkit
+  module Policy
+    # Checks the tenant columns in datasets/registry.yaml. A dataset with
+    # `tenant_column: <field>` is confined to the requester's tenant on every
+    # query. The risk is a tenant table nobody marked, so a dataset that has
+    # another dataset's tenant column must either be marked too, or say
+    # `tenant_column: false` to declare every tenant may read all of it.
+    class TenancyCheck
+      def self.call(raw_registry)
+        new(raw_registry || {}).call
+      end
+
+      def initialize(raw_registry)
+        @datasets = raw_registry.to_h { |name, data| [name.to_s, data || {}] }
+      end
+
+      def call
+        @datasets.flat_map { |name, data| problems(name, data) }
+      end
+
+      private
+
+      def problems(name, data)
+        column = data["tenant_column"]
+        fields = (data["fields"] || {}).keys.map(&:to_s)
+
+        case column
+        when false then []
+        when nil then unmarked(name, fields)
+        when String
+          return [] if fields.include?(column)
+
+          ["#{name}: tenant_column #{column} isn't one of its fields, so no query on it could be confined to a tenant."]
+        else
+          ["#{name}: tenant_column must be a field name, or false for a table every tenant may read."]
+        end
+      end
+
+      def unmarked(name, fields)
+        (fields & tenant_columns).map do |column|
+          owners = @datasets.select { |_, d| d["tenant_column"] == column }.keys.join(", ")
+          "#{name} has #{column}, the tenant column of #{owners}, but no tenant_column. Add " \
+            "`tenant_column: #{column}` to confine it to one tenant, or `tenant_column: false` if every tenant may read all of it."
+        end
+      end
+
+      def tenant_columns
+        @tenant_columns ||= @datasets.values.map { |d| d["tenant_column"] }.grep(String).uniq
+      end
+    end
+  end
+end

data/lib/vkit/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Vkit
-  VERSION = "1.1.6"
+  VERSION = "1.2.0"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: vaultkit
 version: !ruby/object:Gem::Version
-  version: 1.1.6
+  version: 1.2.0
 platform: ruby
 authors:
 - Nnamdi Ogundu
@@ -95,24 +95,24 @@ files:
 - lib/vkit/core/registry_diff_printer.rb
 - lib/vkit/core/table_formatter.rb
 - lib/vkit/policy/bundle_compiler.rb
+- lib/vkit/policy/match_check.rb
 - lib/vkit/policy/packs/ai_safety/metadata.yaml
-- lib/vkit/policy/packs/ai_safety/policies/01_deny_agent_prod_without_clearance.yaml
-- lib/vkit/policy/packs/ai_safety/policies/02_require_approval_for_sensitive_in_prod.yaml
-- lib/vkit/policy/packs/ai_safety/policies/03_mask_sensitive_by_default_for_agents.yaml
+- lib/vkit/policy/packs/ai_safety/policies/01_require_approval_agents_sensitive_prod.yaml
+- lib/vkit/policy/packs/ai_safety/policies/02_require_approval_agents_high_sensitivity.yaml
+- lib/vkit/policy/packs/ai_safety/policies/03_require_approval_agents_unknown_prod.yaml
 - lib/vkit/policy/packs/financial_compliance/metadata.yaml
-- lib/vkit/policy/packs/financial_compliance/policies/01_require_approval_for_financial_prod.yaml
-- lib/vkit/policy/packs/financial_compliance/policies/02_mask_payment_tokens.yaml
-- lib/vkit/policy/packs/financial_compliance/policies/03_deny_non_admin_access_pci_in_prod.yaml
-- lib/vkit/policy/packs/financial_compliance/policies/04_short_ttl_for_financial_grants.yaml
+- lib/vkit/policy/packs/financial_compliance/policies/01_require_approval_financial_prod.yaml
+- lib/vkit/policy/packs/financial_compliance/policies/02_short_ttl_for_financial_grants.yaml
 - lib/vkit/policy/packs/starter/metadata.yaml
-- lib/vkit/policy/packs/starter/policies/01_deny_sensitive_without_clearance.yaml
+- lib/vkit/policy/packs/starter/policies/01_deny_credentials.yaml
 - lib/vkit/policy/packs/starter/policies/02_mask_pii_by_default.yaml
-- lib/vkit/policy/packs/starter/policies/03_require_approval_high_sensitivity.yaml
-- lib/vkit/policy/packs/starter/policies/04_block_cross_region.yaml
-- lib/vkit/policy/packs/starter/policies/05_limit_production_access.yaml
-- lib/vkit/policy/packs/starter/policies/06_default_ttl.yaml
+- lib/vkit/policy/packs/starter/policies/03_mask_high_sensitivity.yaml
+- lib/vkit/policy/packs/starter/policies/04_require_approval_high_sensitivity.yaml
+- lib/vkit/policy/packs/starter/policies/05_default_ttl.yaml
+- lib/vkit/policy/packs/starter/policies/06_flag_unknown.yaml
 - lib/vkit/policy/policy_validator.rb
 - lib/vkit/policy/schema/policy_bundle.schema.json
+- lib/vkit/policy/tenancy_check.rb
 - lib/vkit/policy/validate_bundle.rb
 - lib/vkit/policy/validation_error.rb
 - lib/vkit/utils/banner.rb

data/lib/vkit/policy/packs/ai_safety/policies/01_deny_agent_prod_without_clearance.yaml

@@ -1,15 +0,0 @@
-id: ai_deny_agent_prod_without_clearance
-description: "Deny agent access in production unless requester_clearance is high/admin."
-priority: 1100
-
-match: {}
-
-context:
-  environment: production
-  requester_role: agent
-  requester_clearance: low
-
-action:
-  deny: true
-  reason: "Automation/agents in production require elevated clearance."
-  ttl: "1h"

data/lib/vkit/policy/packs/ai_safety/policies/02_require_approval_for_sensitive_in_prod.yaml

@@ -1,16 +0,0 @@
-id: ai_require_approval_sensitive_prod
-description: "Require approval when accessing sensitive fields in production."
-priority: 1200
-
-match:
-  fields:
-    contains: ["pii", "financial", "phi", "secrets"]
-
-context:
-  environment: production
-
-action:
-  require_approval: true
-  approver_role: security
-  reason: "Sensitive data access in production requires security review."
-  ttl: "1h"

data/lib/vkit/policy/packs/ai_safety/policies/03_mask_sensitive_by_default_for_agents.yaml

@@ -1,18 +0,0 @@
-id: ai_mask_sensitive_for_agents
-description: "Mask sensitive fields by default for agents."
-priority: 1300
-
-match:
-  fields:
-    contains: ["pii", "financial", "phi", "secrets"]
-
-context:
-  requester_role: agent
-
-action:
-  mask: true
-  reason: "Agents should not receive raw sensitive fields by default."
-  ttl: "4h"
-
-masking:
-  default_method: redact

data/lib/vkit/policy/packs/financial_compliance/policies/01_require_approval_for_financial_prod.yaml

@@ -1,16 +0,0 @@
-id: fin_require_approval_financial_prod
-description: "Require approval for financial-category fields in production."
-priority: 2100
-
-match:
-  fields:
-    category: financial
-
-context:
-  environment: production
-
-action:
-  require_approval: true
-  approver_role: finance_manager
-  reason: "Financial data access in production requires finance approval."
-  ttl: "1h"

data/lib/vkit/policy/packs/financial_compliance/policies/02_mask_payment_tokens.yaml

@@ -1,17 +0,0 @@
-id: fin_mask_payment_tokens
-description: "Mask payment tokens / card-related fields everywhere."
-priority: 2200
-
-match:
-  fields:
-    contains: ["pci", "card", "payment_token"]
-
-context: {}
-
-action:
-  mask: true
-  reason: "Payment tokens/card-related data must never be exposed in raw form."
-  ttl: 24h
-
-masking:
-  default_method: redact

data/lib/vkit/policy/packs/financial_compliance/policies/03_deny_non_admin_access_pci_in_prod.yaml

@@ -1,15 +0,0 @@
-id: fin_deny_non_admin_pci_prod
-description: "Deny PCI-tagged data access in production unless admin."
-priority: 2300
-
-match:
-  fields:
-    contains: ["pci"]
-
-context:
-  environment: production
-
-action:
-  deny: true
-  reason: "PCI-tagged data in production is restricted to admins."
-  ttl: "1h"

data/lib/vkit/policy/packs/financial_compliance/policies/04_short_ttl_for_financial_grants.yaml

@@ -1,14 +0,0 @@
-id: fin_short_ttl_financial
-description: "Short TTL for financial access grants to reduce risk window."
-priority: 2400
-
-match:
-  fields:
-    category: financial
-
-context: {}
-
-action:
-  allow: true
-  reason: "Financial grants must expire quickly."
-  ttl: "30m"

data/lib/vkit/policy/packs/starter/policies/01_deny_sensitive_without_clearance.yaml

@@ -1,15 +0,0 @@
-id: deny_sensitive_without_clearance
-description: "Deny access to highly sensitive data without sufficient clearance."
-
-match:
-  fields:
-    sensitivity: high
-
-context:
-  requester_clearance: low
-
-priority: 90
-
-action:
-  deny: true
-  reason: "High sensitivity data requires elevated clearance."

data/lib/vkit/policy/packs/starter/policies/04_block_cross_region.yaml

@@ -1,14 +0,0 @@
-id: block_cross_region
-description: "Block cross-region access to prevent data residency violations."
-
-match: {}
-
-context:
-  requester_region: US
-  dataset_region: EU
-
-priority: 95
-
-action:
-  deny: true
-  reason: "Cross-region data access is not permitted."

data/lib/vkit/policy/packs/starter/policies/05_limit_production_access.yaml

@@ -1,16 +0,0 @@
-id: limit_production_access
-description: "Restrict production data access to approved roles."
-
-match: {}
-
-context:
-  environment: production
-  requester_role: analyst
-
-priority: 70
-
-action:
-  require_approval: true
-  approver_role: admin
-  reason: "Production access for analysts requires approval."
-  ttl: "2h"